Optimizing radiation safety in Chronic Total Occlusion (CTO) procedures: Dosimetric assessment with and without the implementation of Egg Nest radioprotective shielding
Τ. Μ. Axakali*, M. Α. Kouri**,***, Ε. Ioannidis****, Κ. Manousopoulos****, I. Papadopoulos****, P. Varelas****, Κ. Fillipou****, Ρ. Manwlakou****, I. Karalis****, I. Tsiafoutis****, E. Kounadi***, N. Kalyvas* and G. Fountos*
*Department of Biomedical Engineering, Radiation Physics, Materials Technology and Biomedical Imaging Laboratory, AKTYBA, University of West Attica
**2nd Department of Radiology, Medical Physics Unit, Medical School, National and Kapodistrian University of Athens
***Medical Physics, GHA Korgialeneio Mpenakeio-Hellenic Red Cross
****Hemodynamic Laboratory, GHA Korgialeneio Mpenakeio-Hellenic Red Cross, Athens, Greece
Background: Cardiovascular diseases remain the leading cause of mortality, with a higher incidence in males. As a result, Interventional Cardiology has become increasingly important, particularly in the treatment of Chronic Total Occlusion (CTO)—a complete coronary artery blockage causing significant ischemia and angina. CTO is managed using a catheter-based approach via the femoral or radial artery,1,2 followed by guidewire navigation and standard Percutaneous Coronary Intervention (PCI), including balloon angioplasty and stent placement.3 CTO-PCI has proven effective in relieving symptoms and reducing antianginal medication in patients with ischemia and myocardial hibernation.4 However, the procedure involves extended radiation exposure, as indicated by elevated Air-Kerma values.5,6
Materials and method: Radiation doses were measured using real-time, active personal dosimeters worn by interventional cardiologists and nursing staff, recording exposure every minute. This approach enabled accurate assessment of radiation based on role, exposure duration, and proximity to the source. Measurements were taken before and after introducing the Egg Nest shielding system under consistent conditions. The Egg Nest mitigates scattered and leakage radiation, reducing exposure for all personnel. Equivalent Dose and Effective Dose (mSv) were calculated using dosimeter data, adjusted by radiation and tissue weighting factors. Estimated Cancer Risk (ECR) was derived from Effective Dose using established risk coefficients. Statistical analysis compared pre- and post-shielding doses to determine effectiveness.
Results: The analysis confirmed a significant reduction in radiation exposure across all personnel following Egg Nest implementation, evidenced by decreased equivalent and effective doses. This reduction translated into a lower ECR, emphasizing the shielding’s efficacy in enhancing occupational safety. The real-time dosimetry not only validated these improvements but also revealed procedural inefficiencies and sources of avoidable exposure. These findings highlight the critical role of dynamic radiation monitoring and targeted protective measures in optimizing long-term safety in interventional cardiology.
Discussion: The results emphasize the need for continuous radiation monitoring and improved protection strategies in interventional cardiology. Real-time dose tracking revealed exposure inefficiencies, leading to enhanced safety protocols. The Egg Nest shielding significantly reduced radiation, underscoring the value of ongoing optimization to minimize long-term exposure risk.
Conclusion: Minimizing radiation exposure is vital to reduce long-term risks in interventional cardiology. Ongoing optimization of protection measures and innovative shielding is key to safeguarding healthcare staff, with future research needed to assess sustained effectiveness.
Volumetric Modulated Arc Therapy (VMAT) versus optimized dynamic conformal arc (OptDCA) techniques for Linac-based stereotactic radiotherapy of single brain metastases
M. Giannopoulou*, T. Stroubinis**, D. Stasinou**, M. Psarras**, A. Zygogianni*,***, M. Protopapa*, V. Kouloulias**** and K. Platoni*,**
*Department of Applied Medical Physics, School of Medicine, National and Kapodistrian University of Athens, Attikon University Hospital, Athens, Greece
**Department of Radiation Oncology and Stereotactic Radiosurgery, Mediterraneo Hospital, Athens, Greece
***1st Department of Radiology, School of Medicine, National and Kapodistrian University of Athens, Athens, Greece
****Department of Radiation Oncology, School of Medicine, National and Kapodistrian University of Athens, Attikon University Hospital, Athens, Greece
Introduction: Stereotactic radiotherapy (SRT) and stereotactic radiosurgery (SRS) are highly precise radiation treatment techniques used for the management of brain metastases offering excellent local tumor control with minimal impact on surrounding healthy tissues.1 Two Linac-based techniques used for these treatments are Volumetric Modulated Arc Therapy (VMAT) and Dynamic Conformal Arc (DCA). VMAT enables highly conformal dose distribution through continuous modulation of the multi-leaf collimator (MLC), gantry speed, and dose rate, potentially enhancing dose conformity and sparing organs at risk (OARs). In DCA, both the gantry speed and dose rate remain constant, while the MLCs adjust slightly to match the shape of the target for each gantry angle. The dose distribution of the DCA plan is usually not as conformal as that of the VMAT plan. Therefore, performing optimization of the DCA plan (optDCA) can potentially enhance dose conformity.2 This study aims to evaluate and compare the plan quality metrics, dosimetric parameters and the delivery efficiency for single brain metastases SRS/SRT plans for VMAT and optDCA techniques.
Material and Methods: Ten patients with a single brain metastasis who were previously treated with VMAT techniques using an Edge Linac were retrospectively randomly selected. New optDCA plans were generated using the same geometry and energy (6 flattening filter free (FFF) with dose rate 1400 MU/min) such as the original VMAT using the treatment planning system (TPS) Eclipse (version 15.6.06). The optDCA plans were optimized utilizing parameters that limit the modulation of the plan, such as aperture shape controller and MU objective. Paddick conformity index (PCI), gradient index (GI), number of Monitor Units (MU) and brain V12/20Gywere recorded for each plan. Gamma Passing Rates (GPR) were calculated for the criteria of 3%/1 mm and 5%/1 mm dose difference and distance-to-agreement, respectively. These indices were statistically assessed to evaluate differences between VMAT and OptDCA. Statistical analysis with non-parametric Wilcoxon signed rank test was performed.
Results: No statistically significant difference between the two techniques was found for GI (p = 0.726), GPR [3%/1 mm (p = 0.507), 5%/1 mm (p = 0.917)], V12 (p = 0.646), and V20 (p = 0.678). However, a statistically significant difference was observed for PCI (p = 0.005) and MU (p = 0.028), suggesting that VMAT demonstrated superior conformity however requiring a higher number of MU compared to OptDCA.
Discussion: In this study, it was observed that the VMAT technique offers better PCI compared to OptDCA, which is expected due to the higher modulation of the MLCs. However, OptDCA requires fewer MU due to the smaller modulation of the beam and lower complexity. OptDCA seems to be a promising technique for single brain metastasis SRS/SRT. Future steps will involve the evaluation for its further implementation in multiple metastases cases.
Conclusion: Both techniques generate clinically acceptable plans with comparable dosimetric parameters. The optDCA technique provides similar plans to the VMAT technique with less complexity and shorter treatment delivery time.
References
1. C. Velten, et al. “Single isocenter treatment planning techniques for stereotactic radiosurgery of multiple cranial metastases,” Phys Imaging Radiat Oncol, vol. 17, pp. 47–52, Jan. 2021.
2. D. Pokhrel, et al. “Dynamic conformal arcs-based single-isocenter VMAT planning technique for radiosurgery of multiple brain metastases,” Medical Dosimetry, vol. 46, no. 2, pp. 195–200, Jun. 2021.
Keywords: SRS, Radiosurgery, Single brain metastases, Single isocenter, VMAT, DCA, dosimetric parameters
Nanoparticles in medical imaging: Advantages and challenges
A. Tsantiri*, S. Zisiadi*, S. Vorrias*, Z. Tsouris*, A. Plousi** and E. P. Efstathopoulos**
*MSc Program in Nanomedicine, National Kapodistrian University of Athens
**Medical Physicist, Department of Applied Medical Physics, National and Kapodistrian University of Athens
Introduction: Nanotechnology has significantly advanced in recent years, enabling important applications in clinical diagnostics, particularly in medical imaging. The use of nanoparticles (NPs) as contrast agents has emerged as a promising alternative to traditional agents, aiming to improve both sensitivity and specificity in imaging techniques.
Materials and methods: This literature review focuses on studies published in the past decade, highlighting key applications of nanoparticles in various imaging modalities such as Magnetic Resonance Imaging (MRI), Computed Tomography (CT), Positron Emission Tomography (PET), Single Photon Emission Computed Tomography (SPECT), ultrasound, and optical imaging. Particular emphasis was given to each nanoparticle category’s types, mechanisms of action, and physicochemical properties.
Results: In MRI, iron oxide-based nanoparticles have been widely used as T2 agents, demonstrating notable biocompatibility and magnetic properties. Gadolinium-, manganese-, and gold-based nanoparticles have also been explored, showing improved image enhancement and reduced toxicity. In CT, gold, bismuth, and superparamagnetic iron oxide nanoparticles (SPIONs) serve as efficient contrast agents due to their high atomic number and X-ray attenuation. Furthermore, nanoparticles are increasingly utilized in PET, SPECT ultrasound imaging, and novel modalities such as optical imaging.
Discussion: The multifunctional nature of nanoparticles allows for multimodal imaging and potential integration of therapeutic agents (theranostics). While their advantages are considerable, including increased specificity, prolonged circulation time, and lower toxicity, limitations such as biocompatibility, clearance, and regulatory challenges persist.
Conclusions: Nanoparticles represent a transformative advancement in medical imaging. Their role as contrast agents is expanding, with promising prospects in personalized diagnostics and therapy. Further research is essential to overcome current limitations and fully realize their clinical potential.
References
1. J. Wahsner, E. M. Gale, A. Rodriguez-Rodriguez and P. Caravan (2019) Chemistry of MRI Contrast Agents: Current Challenges and New Frontiers, Chemical Reviews, 119:957–1057.
2. S. M. Dadfar, K. Roemhild, N. Drude, S. von Stillfried, R. Knüchel-Clarke and T. Lammers (2019) Iron oxide nanoparticles: Diagnostic, therapeutic and theranostic applications, Advanced Drug Delivery Reviews, 138:302–325.
3. T. Bauerle, M. Saake and M. Uder (2021) Gadolinium-based contrast agents: What we learned from acute adverse events, nephrogenic systemic fibrosis and brain retention, Rofo, 193:1010–1018.
4. C. Henoumont, M. Devreux and S. Laurent (2023) Mn-Based MRI Contrast Agents: An Overview, Molecules, 28:21
5. J. A. Park, H. S. Choi, Y. J. Lee, H. C. Kim and J. S. Kim (2013) Gadolinium Complex of (125)I/(127)I-RGD-DOTA Conjugate as a Tumor-Targeting SPECT/MR Bimodal Imaging Probe, ACS Med Chem Lett, 4:216–219.
6. D. H. Hu, J. Y. Zhang, F. Gao, J. Liang, Y. Wang and X. W. Li (2013) Hybrid gold-gadolinium nanoclusters for tumor-targeted NIRF/CT/MRI triple-modal imaging in vivo, Nanoscale, 5:1624–1628.
7. X. Luo, L. Wang, J. Duan, Q. Li, Y. Ma and Z. Xu (2017) Dual-functional lipid-like nanoparticles for delivery of mRNA and MRI contrast agents, Nanoscale, 9:1575–1579.
Keywords: Nanoparticles, Contrast Agents, Medical Imaging, MRI, CT, Theranostics
Dosimetry in non-standard fields of 1.5T MR-Linacs: Correction factors for two commercially available OSL dosimeters
V. Margaroni*, P. Karaiskos* and E. P. Pappas*
*Medical Physics Laboratory, Medical School, National and Kapodistrian University of Athens, Athens, Greece
Introduction: In small and non-standard field dosimetry in MR-Linac systems, the TRS-483 code-of-practice needs to be extended to account for potential changes in the detector’s response associated with the presence of the strong magnetic field.1 Optically Stimulated Luminescence (OSL) dosimeters are excellent candidates for Quality Assurance procedures and remote audit tests in MR-Linacs.2 The aim of this study is to implement a Monte Carlo based framework for the determination of the relevant correction factors for two commercially available Optically Stimulated Luminescence (OSL) dosimeters under non-standard irradiation fields.
Materials and methods: The EGSnrc V2019 Monte Carlo software package was employed throughout this study. Phase space files for two clinical fields (i.e., 3 x 3 and 2 x 2 cm2), fclin, and the machine-specific reference 10 x 10 cm2 field, fmsr, of the Unity 1.5T/7MV MR-Linac (Elekta, UK) were provided by the vendor and used as the source models. The nanoDotsTM (Landauer Inc., USA) and the myOSLchipTM (RadPro International GmbH, Germany) commercially available dosimeters were considered in this study. They comprise active volumes made of Al2O3 (disk of 5 mm in diameter and 0.2 mm thick, density ρ = 1.41 g/cm3) and BeO (square disk of 4.65 x 4.65 mm2 and 0.5 mm thick, density ρ = 2.85 g/cm3), respectively.
Both detectors were modelled according to blueprints provided by the corresponding manufacturers (Figure 1). OSL dosimeters were simulated at a depth of 5 cm inside an RW3 (PTW, Germany) solid phantom. The magnetic field was always perpendicular to the irradiation beam and parallel to the treatment couch. Gantry angle was fixed at 0°. Two detector orientations were considered: (i) coronal and (ii) axial (Figure 1). The correction factors were determined as the field output factor multiplied by the ratio of detector readings under the fmsr and fclin field in the presence of the static magnetic field B, i.e.,3,4:
The two commercially available OSL dosimeters modelled in EGSnrc. The two detector orientations with respect to the magnetic field are defined.
Results: The calculated magnetic field correction factors are presented in Table 1.
Magnetic field correction factors for the two OSL detectors and both orientations. Overall combined uncertainties at 68% confidence level are given in the parentheses.
OSL dosimeter type
OSL dosimeter orientation
Nominal field size (cm2)
3 x 3
2 x 2
nanoDotsTM
Coronal
1.017 (6)
1.016 (6)
Axial
1.012 (6)
0.995 (6)
myOSLchipTM
Coronal
1.008 (6)
1.018 (6)
Axial
1.002 (6)
0.994 (6)
Conclusion: A set of correction factors for two commercially available OSL detectors were determined, enabling more options for experimental dosimetry, Qaulity Assurance procedures and remote audit tests in non-standard MR-Linac irradiation fields.
References
1. de Pooter et al, Phys Med Biol 66 (2021) 05TR02
2. Episkopakis et al, Phys Med Biol 68 (2023) 225002
3. Blum et al, Phys Med Biol 66 (2021) 155003
4. Margaroni et al, Phys Med Biol 70 (2025) 025011
Keywords: OSLD, small field, dosimetry, correction factors, MR-Linac, Unity
Acknowledgements
Elekta Ltd is acknowledged for providing the phase space files. Landauer Inc. and RadPro International GmbH are acknowledged for providing detailed schematics of the corresponding OSL dosimeters.
This work was supported by computational time granted from the Greek Research and Technology Network (GRNET) in the National HPC facility – ARIS – under project ID pr017028.
Evaluation of a deep learning image reconstruction algorithm on radiation dose and image quality in CT examinations
M. Patsioti*, A. Ploussi*, G. Christopoulos** and E. P. Efstathopoulos*
*Department of Applied Medical Physics, Medical School, National and Kapodistrian University of Athens, Athens, Greece
**2nd Department of Radiology, Medical School, National and Kapodistrian University of Athens, Athens, Greece
Abstract: Deep learning image reconstruction algorithms in medical imaging reduce radiation dose and improve image quality, by employing artificial intelligence. The aim of this study was to investigate the effect of the deep learning image reconstruction algorithm, Advanced intelligent Clear-IQ Engine (AiCE), integrated into Aquilion ONE Prism (Canon Medical Systems) CT scanner on radiation dose and image quality in CT examinations.
Materials and methods: A cylindrical TOS-SS phantom was used for data acquisition. The phantom was scanned using the Aquilion ONE Prism CT scanner at 6 different dose levels by modifying the tube current (100/200/300/400/500/600 mA). Images were reconstructed using AiCE (Standard level), FBP, and AIDR-3D algorithms. CTDIvol and DLP were recorded from dose report and effective dose (ED) was calculated. Parameters such as noise (SD), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were calculated for image quality evaluation.
Results: Effective dose ranges from 1.25 to 7.80 mSv. In comparison with the FBP algorithm, the noise reduction in AiCE images ranged from 34.72% to 39.72%, depending on the x-ray tube current, while in the case of the iterative reconstruction algorithm AIDR-3D, the reduction varied from 32.50% to 36.63% (Table 1). The SNR and CNR values were comparable in both AiCE and AIDR-3D images, while both parameters were higher compared to the FBP images (Figure 1).
Noise reduction (%) using AiCE and AIDR-3D algorithms compared to the FBP, for six different x-ray tube current values.
Algorithm
Noise reduction (%)
100 mA
200 mA
300 mA
400 mA
500 mA
600 mA
AiCE
39.72
36.70
34.73
35.50
34.72
35.57
AIDR-3D
36.63
34.41
32.54
32.87
33.14
32.50
Reconstructed images acquired at 120 kVp and 300 mA using the AiCE, FBP, and AIDR-3D algorithms, respectively, along with the corresponding SNR and CNR values for the acrylic region of interest.
Conclusion: The use of deep learning image reconstruction algorithms can reduce image noise, especially at lower doses, while improving image quality by increasing the SNR and CNR parameters. Additional measurements at lower exposures (<100 mA) are needed to explore the behaviour of AiCE algorithmon on radiation dose and image quality.
References
1. J. Greffier, Effect of a new deep learning image reconstruction algorithm for abdominal computed tomography imaging on image quality and dose reduction compared with two iterative reconstruction algorithms: a phantom study, Quant Imaging Med Surg 2022;12(1):229-243.
2. Higaki et al., Deep Learning Reconstruction at CT Phantom Study of the Image Characteristics, Academic Radiology, Vol 27, No 1, January 2020.
Introduction: Aggregation-induced emission (AIE) molecules are at the forefront of research as they present unique optical properties. These molecules in their aggregate state present enhanced emission and can be utilized in bioimaging applications.1 Curcumin and quercetin are two natural substances that have been actively studied as anticancer agents. It has been proven that these molecules, upon encapsulation in polymeric nanoparticles, can be stabilized and enhance their emission. In this study, the synthesis and self-assembly study of an amphiphilic copolymer was carried out. Encapsulation of curcumin and quercetin was performed. Stability and physicochemical studies, and preliminary biocompatibility tests proved the successful formulation of nanocarriers.2
Materials: Linear and statistical amphiphilic copolymers of poly(oligoethylene glycol methylether methacrylate-co-methyl methacrylate), P(OEGMA-co-MMA), were synthesized in varying comonomer ratios by controlled polymerization. Self-assembly studies were carried out in water. Curcumin and quercetin were loaded in the polymeric nanocarriers while preliminary biocompatibility was tested via the addition of fetal bovine serum in all copolymer-drug solutions.
Methods: Copolymer synthesis was carried out by reversible addition fragmentation chain transfer (RAFT) polymerization and confirmed with the Size Exclusion Chromatography (SEC), Proton Nuclear Magnetic Resonance Spectroscopy (1H-NMR), and Attenuated Total Reflectance (ATR)-Fourier Transform Infrared (FTIR) Spectroscopy. Nanoformulations were prepared with the co-solvent protocol and studied with Dynamic Light Scattering (DLS), Electrophoretic Light Scattering (ELS), Fluorescence (FS), and UV–Vis Absorption Spectroscopy (UV–Vis).
Results: Copolymer characterization proved the successful synthesis of the materials with low polydispersity indexes. Self-assembly studies revealed the formation of nanoaggregates for all copolymers in water. The maximum encapsulation efficiency that was achieved for curcumin was 54%, and 49% for quercetin. The nanosystems proved to be stable for 20 days. Photophysical characterization confirmed the AIE phenomenon. FBS assays also showed no interaction between the nanocarriers and the proteins.
Discussion: MMA and OEGMA are two well-known and biocompatible monomers. The hydrophilic properties of OEGMA comonomer and the hydrophobic ones of MMA resulted in the self-assembly of the synthesized copolymers in water and the creation of nanoaggregates. The hydrophobic drugs were encapsulated in the hydrophobic domains of these nanostructures. Due to the entrapment of the drugs, the aromatic rings of these molecules were stabilized and aggregated, therefore the AIE phenomenon was observed.
Conclusions: Herein, we focused on the synthesis and self-assembly properties of these novel copolymers as well as on encapsulation studies for the drug molecules, which also show photophysical properties.
References
1. Pantelaiou, M.A., Vagenas, D., Karvelis, E.S., Rotas, G., Pispas, S. Co-Assembled Nanosystems Exhibiting Intrinsic Fluorescence by Complexation of Amino Terpolymer and Its Quaternized Analog with Aggregation-Induced Emission (AIE) Dye. Nanomaterials 2024, 14, 1631, DOI:10.3390/nano14201631.
2. Pantelaiou, M.A., Vagenas, D., Pispas, S. Poly(Oligoethylene Glycol Methylether Methacrylate-Co-Methyl Methacrylate) Aggregates as Nanocarriers for Curcumin and Quercetin. Polymers 2025, 17, DOI:10.3390/polym17050635.
Keywords: Amphiphilic random copolymers, Nanocarriers, Nanomedicine, Curcumin, Quercetin
Assessing the dosimetric effect of golden beam data and machine equivalence in VMAT and IMRT treatment plans
E. Anousis*,**, G. Patatoukas*, M. Chalkia*, N. Kollaros*, K. Zourari*, N. Trogkanis*, D. Michaletou* and K. Platoni*
*Department of Applied Medical Physics, Attikon University Hospital, Medical School, National and Kapodistrian University of Athens, Athens, Greece
**Department of Medicine, MSc Program in Medical Physics – Radiation Physics, University of Patras, Patra, Greece, Medical Physics Unit
Introduction: In radiotherapy Golden Beam Data (GBD) consist of measurements like Percentage Depth Dose (PDD), beam profiles, and output factors for various field sizes and energies, obtained under standardized conditions representing a “reference” linear accelerator (LINAC). The aim of this work is to assess the impact that the differences between measured data and GBD may have on the dosimetry of radiotherapy treatment plans.
Materials and methods: The clinical aspect of the use of GBD was examined via the comparison of Dose Volume Histograms (DVH) for Planning Target Volumes (PTVs) and Organs At Risk (OARs) for 69 patients of different treatment sites (24 lung, 15 breast, 15 prostate and 15 head and neck cancer patients) based on Normal Tissue Constraint Guidelines.
Originally delivered treatment plans of Attikon’s two LINACs (VB1, VB2) were recalculated using the Anisotropic Analytical Algorithm (AAA) for the simulated LINAC (VB3) and dose related parameters were analyzed and compared.
Results: The differences between VB1 and VB2 were insignificant (Figure 1), a fact that confirms the equivalence of the LINACS. Furthermore, comparison with the VB3 showed differences in volumetric DVH parameters (Figure 2).
Average dose differences between VB1-VB2 and initial machine-VB3 for treatment sites of lung and breast. The largest differences are highlighted.
DVH curve comparison of a prostate plan calculated in VB1 (■ points) and VB3 (▲ points).
Conclusion: While GBD provides a standardized approach to beam modeling, measured data could ensure increased accuracy, reducing possible dosimetric uncertainties. Careful validation of beam data selection is recommended to optimize clinical treatment quality.
References
1. Beam modeling and beam model commissioning for Monte Carlo dose calculation-based radiation therapy treatment planning: Report of AAPM Task Group 157
2. Y. A. M. Yousif et al, “Golden beam data provided by linear accelerator manufacturers should be used in the commissioning of treatment planning systems”
Keywords: Radiation Therapy, Golden Beam Data, Dosimetry
Software solutions for shielding calculations in radiotherapy rooms
C. Ganou*, A. Ganos**, S. Papatheodorou***, V. Metaxas*, C. Dimitroukas*,*** and G. Panayiotakis*,***
*Department of Medical Physics, School of Medicine, University of Patras, Greece
**Department of Civil Engineering, University of Patras, Greece
***Department of Medical Physics, University Hospital of Patras, Greece
Introduction: Effective shielding in radiation therapy is crucial to protect staff and the public from harmful radiation. Linear accelerators produce high-energy beams that can penetrate walls, making proper shielding design essential. Shielding calculations determine barrier thickness based on radiation type, energy, and exposure conditions. Designing these shields must balance safety, cost, and space. Existing methods often prioritize dose estimation, but they tend to be complex and expensive, making them less accessible for smaller departments. This study aims to develop a user-friendly software tool for simplifying shielding calculations for radiotherapy rooms with linear accelerators, offering an accessible solution for radiation protection planning.
Materials and methods: The “Radiotherapy Infrastructure Shielding Calculations” (RISC) software, developed in MATLAB, allows users to calculate shielding thicknesses without needing MATLAB installed. It features a user-friendly GUI with separate interfaces for primary and secondary barrier calculations. The primary barrier interface includes tabs for primary radiation, patient scattered and leakage radiation, IMRT techniques, and shielding cost. The secondary barrier interface offers tabs for scattered and leakage radiation, IMRT techniques, and shielding cost. RISC uses NCRP 1511 methods to compute barrier thickness with concrete (density: 2.35 g/cm3) for photon energies ranging from 4 to 30 MV. The software also estimates the cost of radiotherapy rooms with linear accelerators, supporting both conventional and IMRT techniques, and includes instantaneous dose rate (IDR) calculations. It comes with two supporting files: a “Terminology” document that defines the parameters and a “User’s Manual” for detailed guidance.
Results: The RISC has been validated against all comparative examples in NCRP 151, as well as calculations from shielding reports for the Varian IX linear accelerator at Methodist Hospital of Willowbrook2 and the Elekta Infinity at the University Hospital of Patras.3
Discussion: The RISC software simplifies shielding calculations for primary and secondary barriers in radiotherapy rooms with linear accelerators, providing detailed and efficient results. Future updates will expand its capabilities to include various radiation treatments (IMRT, IGRT, VMAT, TBI), radiotherapy machines (gamma knife, cyber knife, Co-60), radiation types (photons, protons, electrons), and shielding materials (heavy concrete, steel, lead, earth, wood, BPE). Additional features will include calculations for maze door, skyshine, and groundshine radiation, along with graphical representations of barrier thicknesses. RISC is also a valuable educational tool for graduate students and medical physicists learning shielding calculations.
Conclusions: The RISC software is user-friendly, providing fast and accurate shielding thickness and cost calculations for primary and secondary barriers in radiotherapy rooms with linear accelerators using conventional or IMRT techniques. Validated against NCRP 151 examples and previous shielding reports, it is freely available online as a standalone application. With an intuitive GUI, RISC simplifies input parameter selection, making it useful for graduate education and assisting medical physicists in planning new radiotherapy facilities. As an open project, it will be periodically updated with new features, including additional treatments, machines, shielding materials, and specialized calculations for skyshine, groundshine, and maze door shielding.
References
1. NCRP Report No 151 (Bethesda, MD: National Council on Radiation Protection and Measurements) (2005)
2. Richardson, S., Shielding Calculation Report of Varian iX. Methodist Hospital-Willowbrook (2007)
3. RISC software. Available from: RISC Software - Google Drive
Introduction: Radiomics analysis of prostate cancer (PCa) relies on MRI-derived quantitative features. Variations in magnetic field strength introduce scanner-related variability, potentially affecting feature stability. This study evaluates the robustness of radiomic features and the effectiveness of ComBat harmonization in mitigating field strength effects.
Material and methods: We analyzed 207 PCa lesions (152 clinically significant, 55 insignificant) from local and PROSTATEx21 datasets. Features were extracted from Original, Gradient, LoG, and Wavelet images. ComBat harmonization was used to address multicenter variability.2 Stability across 1.5T and 3T MRI was assessed using p-values and Cohen’s d. Normality was tested via Shapiro-Wilk; comparisons used t-tests or Mann-Whitney U accordingly. Effect sizes were interpreted as large (d > 0.8), small (d < 0.2), with significance at p < 0.05.
Results: Before ComBat harmonization, Wavelet features had the highest number of statistically significant differences (599), followed by LoG (209), Original (88), and Gradient (73), indicating substantial variability due to field strength effects. LoG-derived features exhibited the highest stability, whereas Gradient and Wavelet-derived features were the most unstable, demonstrating significant shifts. Applying Combat normalization dramatically improved stability by reducing the number of statistically significant features to 152 (Wavelet), 11 (LoG), 5 (Original), and 3 (Gradient).
Discussion: These findings emphasize the need for careful feature selection in radiomics studies.3 The results suggest that Combat normalization is a good approach for harmonizing MRI-derived features across different field strengths. The high instability observed in Wavelet and Gradient features highlights the importance of considering scanner effects when selecting radiomic biomarkers.
Conclusion: ComBat harmonization enhances feature stability across MRI field strengths, supporting its application in multi-center radiomics studies. Future research should explore additional harmonization techniques to further improve reproducibility.
References
1. Litjens, G., Debats, O., Barentsz, J., Karssemeijer, N., & Huisman, H. (2017). SPIE-AAPM PROSTATEx Challenge Data (Version 2) [dataset]. The Cancer Imaging Archive.
2. Orlhac F, Eertink JJ, Cottereau AS, Zijlstra JM, Thieblemont C, Meignan M, Boellaard R, Buvat I. (2022) A Guide to ComBat Harmonization of Imaging Biomarkers in Multicenter Studies, J Nucl Med., Feb;63(2):172-179.
3. Peerlings J., Woodruff H.C., Winfield J.M., Ibrahim A., Van Beers B.E., Heerschap A., Jackson A., Wildberger J.E., Mottaghy F.M., DeSouza N.M., Lambin P. (2019) Stability of radiomics features in apparent diffusion coefficient maps from a multi-centre test-retest trial, Sci Rep 9, 4800
This work has been partially supported by project MIS 5154714 of the National Recovery and Resilience Plan Greece 2.0 funded by the European Union under the NextGenerationEU Program.
Low gamma passing rate analysis for VMAT and IMRT treatment plans
K. Outsikas*,**, G. Patatoukas*, M. Chalkia*, K. Zourari*, N. Kollaros*, E. Kypraiou*** and K. Platoni*
*Department of Applied Medical Physics, ‘Attikon’ University Hospital, Medical School National Kapodistrian University, Athens, Greece
**Department of Medicine, Department of Physics, MSc Medical Physics and Radiation Physics, University of Patras, Patras, Greece
***Department of Clinical Radiation Oncology, ‘Attikon’ University Hospital, Medical School National Kapodistrian University, Athens, Greece
Introduction: Patient Specific Quality Assurance (PSQA) is a standard procedure used to compare the calculated distribution with the delivered one. For this purpose, Gamma Passing Rate (GPR), Dose Difference (DD) and Distance to Agreement (DTA) metrics are commonly used. This study aimed to identify parameters systematically associated with low GPR values in VMAT and IMRT treatment plans for Head & Neck, Prostate, Lung and Breast cancer patients.
Materials and methods: Radiotherapy treatment plans were categorized as “Low” or “High” GPR using institution-specific Tolerance and Action Limits. Geometry and dose related metrics such as Plan Averaged Beam Area (PA), Edge Metric (EM) and Dose Rate (DR) among others were calculated and assessed for various treatment sites and irradiation phases. Correlation and statistical analyses were performed against GPR values, using the 3%/2 mm and 2%/2 mm DD/DTA criteria.
Results: Although no strong correlations were found, site- and irradiation phase-specific trends were observed. Statistical comparisons between “Low” and “High” GPR groups showed statistically significant differences in certain metrics values. ROC analysis and scatter plots suggested potential threshold values for metrics that could help flag plans more likely to fail (Figure 1).
Scatter plot of EM which reflects the aperture complexity. Head & Neck low risk irradiation phase.
Conclusion: Correlation analysis revealed the direction of impact for several parameters. Further subcategorization of treatment plans, not only by treatment site but also by irradiation phase, revealed stronger correlations and clearer trends. It was observed that some parameters-although negatively affecting plan’s deliverability (PSQA GPR value) might be compensated for by others with a positive impact, thereby weakening the observed correlations. Thus, further research and potentially further sub-categorization based on metrics’ values may provide insights towards better understanding of deliverability’s fluctuations (Table 1).
ROC & Spearman’s analysis for EM as a predictive metric for identifying low GPR plans.
Statistic
Value
Cutoff Point
0.119 mm−1
Low GPR plans above cutoff
11/11
Sensitivity
0.440
Specificity
1
Positive predictive value
1
Area under curve (AUC)
0.702
rs against GPR (2%/2 mm)/p-value
−0.402/0.004
References
1. Miften, M. et al. Tolerance limits and methodologies for IMRT measurement-based verification QA: Recommendations of AAPM Task Group No.218 Med Phys 45, e53–e83 (2018).
2. Younge, K.C. et al. Penalization of aperture complexity in inversely planned volumetric modulated arc therapy. Med Phys 39, 7160–7170 (2012).
I would like to thank Dr. Platoni, Dr. Patatoukas, M. Chalkia, K. Zourari and N. Kollaros for their guidance and support during this research.
The theranostics era: From treating what you see to seeing what you treat
A. Stratis*
GE Healthcare*
Andreas.Stratis@gehealthcare.com
Introduction: Recent breakthroughs in medtech and radiopharma engineering have ushered in a new era of Nuclear Medicine-based personalized diagnosis and treatment. Theranostics, combining targeted radionuclide therapy with precise diagnostic processes, epitomizes precision care, promising long-term survival and cure within a multidisciplinary framework. While not a new concept, new tracers and advanced imaging systems have unlocked novel clinical strategies, offering advanced tools for targeted therapy and monitoring. In theranostics, patients are not imaged solely to detect metastatic disease but also to determine whether cancer cells express a specific therapeutic target, aiming a high therapeutic index with efficacy markedly surpassing toxicity. This work aims to present the theranostic pathway’s conceptual framework and provide clinical evidence for its efficacy in neuroendocrine tumors (NETs) and prostate cancer.
Building a personalized treatment strategy: The fundamental principle of theranostics involves creating pharmaceutical targeting ligands that bind specifically to overexpressed biomarkers unique to each type of cancer (Figure 1). Neuroendocrine tumors (NETs) overexpress somatostatin receptors (SSTRs), targeted by SSTR-ligands which are labeled with 68Ga for PET/CT imaging and 177Lu for therapy. Similarly, prostate tumors, overexpressing prostate-specific membrane antigen (PSMA), are imaged using 18F- or 68Ga-PSMA, and treated with 177Lu-PSMA.1,2
The development process of a theranostic tracer.
Advanced radiopharmaceutical engineering is required to produce a ligand that, on one end, perfectly matches the overexpressed biomarker and, on the other end, can be bound to different radionuclides through a linker chemical compound. Recent advances in cyclotron technology have given access to different PET/CT isotopes and have expanded the range of theranostic applications. Once the hot-radioactive isotope has been produced, sophisticated synthesis systems are employed to chemically bind the isotope (hot part) with the linker-ligand pair (cold part) and formulate the end theranostic product for imaging purposes.
Modern PET/CT scanners, with advanced reconstruction algorithms, AI technologies, and ultra-sensitive detectors, leverage theranostic tracers in clinical practice. These systems enhance diagnostic confidence with sub-2 mm spatial resolution and exceptional small lesion detectability, operating at low scan times and radiation doses, empowering physicians to ‘treat what they see.’ Post-therapy, patients are monitored using state-of-the-art SPECT/CT scanners, enabling doctors to ‘see what they treat.’ Advanced AI-powered dosimetry tools evaluate the radiotoxicity in sensitive organs to determine whether to continue, modify, or discontinue treatment.
The prostate Ca paradigm:Figure 2 shows the theranostic pathway of a 78-year-old prostate cancer patient with biochemical failure. The patient was imaged with 8 mCi of 18F-PSMA using a GE HealthCare OMNI Legend 32 PET/CT scanner (total time, 8 mins) to determine eligibility for 177Lu-PSMA therapy. Multiple metastatic lesions were detected, and 220 mCi of 177Lu-PSMA were administered in the first cycle. At 24 hours post-treatment, imaging with a CZT-based system (Discovery 870 CZT, GE HealthCare) was performed to monitor treatment and conduct dosimetry, assessing doses to critical organs and comparing against toxicity levels. All pathological lesions shown on PET/CT absorbed the Lu-PSMA, indicating a highly personalized treatment strategy.
An example of advanced theranostics.
Discussion: In the era of precision care, NM-theranostics require multidisciplinary advanced technology synergies to deliver great therapy outcomes.
References
1. Strosberg J et al. Phase 3 trial of (177)Lu-dotatate for midgut neuroendocrine tumors. N Engl J Med. 2017;376:125–35.
2. Emmett L et al. Lutetium177 PSMA radionuclide therapy for men with prostate cancer: a review of the current literature and discussion of practical aspects of therapy. J Med Rad Sci. 2017;64:52–60
Keywords: Theranostics, Precision Care, Personalized medicine
Is AI ending the compromise between dose and image quality? The lung CT paradigm
Introduction: While X-ray (XR) remains the frontline modality for assessing chest pathology, published evidence indicates that the 3D nature of CT imaging provides enhanced diagnostic confidence and increased clinical value. Taekker et al. reported that Ultra-low-Dose CT (ULDCT) is more sensitive than XR for detecting pneumonia and pneumothorax, and equally accurate for diagnosing cardiogenic pulmonary edema and pleural effusion.1 Kroft et al. highlighted ULDCT’s value, showing a 20% reduction in false-positive and false-negative XR results.2 Gheysens et al. (2022) demonstrated ULDCT’s efficiency in detecting solid lung nodules (> 50 mm3) in lung cancer screening, regardless of patient size.3 Despite the significant advantages of CT imaging over XR, the radiation dose burden remains a matter of concern. A novel deep learning-based image reconstruction (DLIR) algorithm, Truefidelity™ (TF), has been recently introduced in GE HealthCare CT systems, featuring a DNN trained with high quality FBP datasets to learn how to differentiate noise from signals, and to intelligently suppress the noise without impacting anatomical and pathological structures. The objective of this study was to investigate whether the implementation of TF on a GE HealthCare Revolution APEX CT scanner can reduce thorax CT radiation doses to XR levels, while preserving the superior diagnostic quality inherent to 3D CT imaging over planar imaging.
Methodology: A series of scans with varying exposure parameters (kV, pitch), DLIR strengths (low, medium, high), and edge enhancement filters (E1-E3) were conducted on a 39.7 cm chest region of a CT whole body phantom (PBU-60, Kyoto Kagaku). All scans used a fixed current of 10 mA, keeping CTDIvol below 0.3 mGy. Image quality was subjectively evaluated based on the “European Guidelines on Quality Criteria for Computed Tomography”, and the top-ones qualified for objective IQ evaluation based on Contrast-to-Noise (CNR) metrics and dose-independent figures of merit (FOMs). To this end, regions of interest (ROIs) of 0.5 cm2 were drawn bilaterally in the anterior, middle, and posterior lung parenchyma of the Kyoto phantom, to measure CTlung,mean (HU) and SDlung. Similar ROIs were placed in the right inferior pulmonary vein and left ventricle to assess CTpv and CTheart, respectively. The parameters yielding the best combination of subjective and objective image quality (IQ) were assessed, and then applied in clinical practice.
Results: For lung parenchyma imaging, helical acquisitions obtained at 140 kV, 10 mA, with a pitch of 1.375 and reconstructed at 0.625 mm slices with TF at high strength provided the highest CNR and recorded a top-5 performance on the dose-independent FOM. These scan parameters resulted in a striking 0.2 mGy CTDIvol, reducing the radiation risk to an effective dose of 0.117 mSv. For mediastinum imaging, axial scans at 120 kV, 10 mA reconstructed at 0.625 mm slices with TF at high strength demonstrated the top performance at 0.21 mGy CTDIvol, and an effective dose of 0.12 mSv.
Discussion: DLIR reconstructed images provide enhanced 3D diagnostic insights, reaching the global mean effective dose of a complete chest X-ray (UNSCEAR) and reducing lung CT radiation risk below the mean effective dose of a planar chest X-ray in Greece (0.13 mSv).
Conclusions: Deep learning has enhanced diagnostic confidence by providing much richer information to doctors reviewing chest images with the 3D data of CT at dose levels comparable to 2D imaging. While ending the trade-off between dose and image quality is challenging, ongoing research promises a bright future in CT imaging.
References
1. Tækker M et al. Diagnostic accuracy of ultra-low-dose chest computed tomography in an emergency department. Acta Radiol. 2022;63(3):336-344.
2. Kroft LJM et al. Added Value of Ultra-low-dose Computed Tomography, Dose Equivalent to Chest X-Ray Radiography, for Diagnosing Chest Pathology. J Thorac Imaging. 2019; 34(3):179-186.
3. Gheysens G et al. Detection of pulmonary nodules with scoutless fixed-dose ultra-low-dose CT: a prospective study. Eur Radiol. 2022; 32(7):4437-4445.
Keywords: TrueFidelity, Deep Learning, lung/thorax CT screening, ULD
Hemodynamic and anatomic effects of the anterior cerebral artery in aneurysm formation
N. Tzaneti*, D. Zantzas*, E. Sidiropoulos**, K. Zisis*, K. Papadopoulou* and A. Kalfas*
*Aristotle University of Thessaloniki, Department of Mechanical Engineering, Thessaloniki, Greece
Introduction: Cerebral aneurysms affect 0.5–6% of the population, with 31–35% located in the anterior communicating artery (ACoA), exhibiting a higher rupture risk. Rupture can lead to severe complications and even death. Hemodynamic factors, especially wall shear stress (WSS), are strongly linked to aneurysm formation, progression, and rupture.1,2 Clinical challenges in prognosis, treatment, and rupture risk assessment serve as the main motivation for this study, which aims to identify hemodynamic and geometric parameters related to the formation of ACoA aneurysms through computational fluid dynamics (CFD).
Methods: Three-dimensional models of the ACoA region were reconstructed from time-of-flight magnetic resonance angiography images of seven healthy individuals and seven patients with ACoA aneurysms.3 Blood flow was modeled as laminar and Newtonian with a density of 1060 kg/m3, and viscosity of 0.0035 Pa·s. Boundary conditions included rigid and non-slip walls, zero pressure conditions at the outlets and both steady and transient velocity at the inlets.2 The primary hemodynamic parameters analyzed included wall shear stress (WSS) and static pressure. Statistical analysis of the results was also performed. The workflow of the study is shown in Figure 1.
Workflow of the study.
Results: The results demonstrated statistically significant lower WSS values in the aneurysm area compared to the corresponding area in healthy models, as is shown in Figure 2a where the time-averaged WSS (TAWSS) in the case of transient flow is presented. Among the geometric parameters, only the angle between the segments Α1 and A2 of the anterior cerebral artery (ACA) showed statistical significance, with a smaller angle correlating with aneurysm presence. Statistical correlation analysis of hemodynamic and geometric parameters of the aneurysms showed that the left angle between A1 and A2 is positively related to WSS (Figure 2b), indicating the effect of parent artery anatomy on the hemodynamics of the aneurysms.
(a) TAWSS contour plots (b) correlation of left angle ACA-A1/A2 with WSS in steady-state flow.
Conclusions: A statistically significant difference was found between the two groups in WSS and the angle between the A1 and A2 segments, highlighting key factors in aneurysm formation. These findings may enhance early detection and inform treatment strategies.
References
1. P. Texakalidis, A. Sweid, N. Mouchtouris, et al. (2019) Aneurysm Formation, Growth, and Rupture: The Biology and Physics of Cerebral Aneurysms, World Neurosurg., 130: 277-284.
2. F. Liang, X. Liu, R. Yamaguchi, and H. Liu (2016) Sensitivity of flow patterns in aneurysms on the anterior communicating artery to anatomic variations of the cerebral arterial network, J Biomech., 49(15): 3731-3740.
3. T. Di Noto, G. Marie, S. Tourbier, Y. Alemán-Gómez, O. Esteban, G. Saliou, M. Bach Cuadra, P. Hagmann, and J. Richiardi (2022) Lausanne_TOF-MRA_Aneurysm_Cohort, OpenNeuro. [Dataset]
Introduction: Cardiovascular diseases (CVDs) are the leading cause of mortality globally, primarily driven by atherosclerosis. Drug Eluting Balloons (DEBs) offer an emerging alternative to stents by delivering antiproliferative agents without leaving permanent implants.
Materials and methods: Seven swine coronary arteries were reconstructed using inSilc software from Optical Coherence Tomography (OCT) and angiography data. A detailed mesh was created for both the balloon and arterial wall using ANSYS. The balloon was modeled as a linear elastic polyamide structure (Grilamid L251), while the artery was assigned Mooney-Rivlin hyperelastic properties. Simulations included folding and pleating, followed by balloon deployment using clinically pressure profiles (0.8–1.2 MPa2). Frictional coefficients (μ=0.2) were considered between the balloon and the arterial wall (Figure 1).
The Von Mises stresses distributionand and the max Von Mises stress in all cases.
Results: Simulations showed consistent performance in luminal expansion without excessive stress on the arterial wall. Maximum stress remained below 2.5 MPa. The DEB’s compliance matched the manufacturer’s chart. Stress distribution analysis indicated that stenosed regions experienced the most mechanical remodeling. Strain values remained within physiological tolerance, with no simulated tissue rupture or delamination.
Conclusion: The in silico deployment framework effectively assessed the Everolimus DEB’s performance in realistic coronary artery models. Results highlight its safety and mechanical reliability for Percutaneous Transluminal Coronary Angioplasty (PTCA) (Table 1).
The mechanical properties used.1,3
Geometry
Parameter
Value
Balloon
Young’s Modulus
1100 (MPa)
Poisson ratio
0.4
Density
1010 (kg/m3)
Artery
C10
0.7 (MPa)
C01
4.5 (MPa)
D1
0.0003 (1/MPa)
References
1. M. A. Geith et al. “Experimental and mathematical characterization of coronary polyamide-12 balloon catheter membranes,” PLoS ONE, vol. 15, no. 6, Jun. 2020, doi: 10.1371/journal.pone.0234340.
2. C. S. Katsouras et al. “Safety and Efficacy of an Innovative Everolimus-Coated Balloon in a Swine Coronary Artery Model,” Life, vol. 13, no. 10, Art. no. 10, Oct. 2023, doi: 10.3390/life13102053.
3. C. Noble et al. “Patient specific characterization of artery and plaque material properties in peripheral artery disease,” J. Mech. Behav. Biomed. Mater., vol. 101, p. 103453, Jan. 2020, doi: 10.1016/j.jmbbm.2019.103453.
Keywords: Drug Eluting Balloon, Finite Element Analysis, Coronary Artery Disease
Acknowledgement
This work is supported by the CleverBalloon project, which has received funding from the “Competitiveness, Entrepreneurship and Innovation” (EPAnEK) Operational Program with code Τ2EΔΚ-03677.
Machine learning models for long-term cardiovascular and related events prediction: A multi-study comparative analysis
K. Tsarapatsani*, A. Sakellarios*, V. Tsakanikas*, H. Rudolf**, H. Trampisch***, M. Kleber****, W. März***** and D. Fotiadis*
*Biomedical Research Institute, Foundation for Research and Technology-Hellas, Ioannina, Greece
**Institute for Biostatistics and Informatics in Medicine and Ageing Research, University Medical Center Rostock, Rostock, Germany
***Department of Medical Informatics, Biometry and Epidemiology, Ruhr University Bochum, Bochum, Germany
****Medical Clinic V, Mannheim Medical Faculty, University of Heidelberg, Mannheim, Germany
*****Institute of Public Health, Social and Preventive Medicine, Medical Faculty Mannheim, University of Heidelberg, Mannheim, Germany
Introduction: Cardiovascular disease (CVD) remains the leading cause of death worldwide.1 This work synthesizes findings from nine independent studies employing machine learning (ML) to predict major CVD-related outcomes, including myocardial infarction (MI), stroke, heart failure (HF), atrial fibrillation (AF), and peripheral vascular events, over 5-20 follow-up periods.
Materials and methods: All analyses used clinical and laboratory data from the Ludwigshafen Risk and Cardiovascular Health (LURIC) study and the German Epidemiological Trial on Ankle Brachial Index (getABI). ML models were implemented, including Logistic Regression (LR), Support Vector Machines, Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and Decision Trees. Pre-processing involved imputation, standardization, feature selection, and hyperparameter tuning. Performance metrics included Accuracy, Area Under the Receiver Operating Characteristic Curve (AUC), Sensitivity, Specificity, Precision and F1-score. The best performing model was selected based mainly on the highest accuracy. In cases of comparable accuracy, models with higher AUC and interpretability were prioritized.
Results: Each study identified a best-performing ML model specific to the clinical endpoint under investigation. Table 1 summarizes the highest performing model from each study, with XGBoost and LightGBM consistently demonstrating superior predictive ability. Notably, the highest accuracy (87.56 %) was achieved by LightGBM in the prediction of CVD/cerebrovascular mortality.
Best performing ML models across studies.
Outcome (follow-up)
Best model
Accuracy
All-cause mortality (20 years)
XGBoost
76.00 (%)
MI (10 years)
XGBoost
74.27 (%)
CVD Death (10 years)
LR
72.20 (%)
Stroke (7 years)
LightGBM
68.00 (%)
Amputation/Revascularization (7 years)
RF
73.27 (%)
Fatal MI (10 years)
LightGBM
69.42 (%)
CVD/Cerebrovascular mortality (7 years)
LightGBM
87.56 (%)
AF (7 years)
XGBoost
71.19 (%)
HF (5 years)
LightGBM
68.00 (%)
Conclusions: This work demonstrates the potential of ML-based tools to enhance CVD – risk prediction and enable earlier, more tailored interventions. Performance varied across endpoints, with simpler models occasionally matching the performance of more complex ones, highlighting the importance of outcome-specific model selection. Limitations related to cohort-specific characteristics and potential variability in endpoint definitions-may affect the generalizability of the findings and should be addressed in future research.
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101017424, as part of TIMELY project.
4D flow MRI-enabled patient-specific computational hemodynamics of thoracic aorta: CFD predictions vs. in vivo imaging data
D. Petrolekas*, A. Raptis*, E. Karavasilis**, K. Moulakakis***, I. Kakisis*** and C. Manopoulos*
*Laboratory of Biofluid Mechanics & Biomedical Technology, School of Mechanical Engineering, National Technical University of Athens, Athens, Greece
**Medical Physics Laboratory, School of Medicine, Democritus University of Thrace, Alexandroupolis, Greece
***Department of Vascular Surgery, Attikon University Hospital, National and Kapodistrian University of Athens, Athens, Greece
Introduction: 4D Flow MRI enables in vivo, time-resolved, three-directional velocity measurements but has spatial and temporal resolution limitations. Computational fluid dynamics (CFD) simulations may complement MRI by providing patient-specific hemodynamic predictions.1 This study compares CFD simulations with 4D Flow MRI data to evaluate the accuracy of CFD in replicating in vivo hemodynamics, employing patient-based MRI-derived aortic geometry and boundary conditions.
Methods: Anonymized 4D Flow MRI data, including magnitude (anatomical) and phase (velocity) images across three velocity encoding directions, were acquired from a normal volunteer. The transient velocity field was reconstructed using Ferdian et al.,2 generating 4D velocity arrays (u, v, w) across all cardiac time steps (t). The magnitude images were used to reconstruct the 3D model of the thoracic aorta (TA). Blood flow simulations were conducted in both SimVascular and CRIMSON software using the same TA model, boundary conditions, and solver settings. The blood velocity profile at the inlet surface was reconstructed from the 4D Flow MRI data and was applied as boundary condition. Three-element Windkessel models were attached at the outlets. The resistance and compliance values were iteratively tuned to achieve physiological pressure. Pressure distribution, flow rates, and wall shear stress (WSS) were calculated. CFD data were reconstructed into 4D arrays, downsampled2 to match MRI resolution, and aligned for direct comparison. A descending aorta cross-section was analyzed, computing mean velocity magnitude for comparison.
Results and discussion: Similar flow conditions are predicted by both programs, though CRIMSON exhibits slightly higher velocities and some differences in velocity distribution along the flow domain. MRI comparisons show that while both software tools capture major flow trends, minor discrepancies appear in velocity magnitudes and flow oscillations, particularly in the aortic arch region. Rigid-wall assumption contributes to overestimated systolic velocities and diastolic fluctuations (Figure 1).
Velocity vectors at peak systole by CFD (left). Time-resolved mean velocity magnitude at a descending aorta cross-section: MRI vs SimVascular vs CRIMSON comparison (right).
Conclusions: Patient-specific inlet conditions are essential for replicating physiological flow dynamics, as they more accurately capture realistic flow patterns and WSS variations, than parabolic profiles, which overestimate velocities and introduce extra discrepancies. This approach enables disease progression prediction through WSS analysis and supports virtual testing of interventions. Future work should integrate FSI to improve physiological realism and refine boundary condition calibration for more accurate hemodynamic parameters.
References
1. L. Shahid, et al. (2022) Enhanced 4D Flow MRI-Based CFD with Adaptive Mesh Refinement for Flow Dynamics Assessment in Coarctation of the Aorta. Annals of Biomedical Engineering, 50:1001–16
2. E. Ferdian, A. Suinesiaputra, et al. (2020) 4DFlowNet: Super-Resolution 4D Flow MRI Using Deep Learning and Computational Fluid Dynamics. Frontiers in Physics, 8:138
This research was carried out within the framework of the Action “Flagship actions in interdisciplinary scientific fields with a special focus on the productive fabric”, which is implemented through the National Recovery and Resilience Plan Greece 2.0, funded by the European Union – NextGenerationEU (Project ID: TAEDR-0535983).
Dissection mechanics of tissue components in human aorta
E. Papanikolaou*, D. Sokolis**, D. Iliopoulos*** and C. Manopoulos*
*Laboratory of Biofluid Mechanics and Biomedical Technology, School of Mechanical Engineering, NTUA, Athens, Greece
**Laboratory of Biomechanics, Center of Clinical, Experimental Surgery, and Translational Research, BRFAA, Athens, Greece
***Professor of Cardiac Surgery, National and Kapodistrian University of Athens & 4th Cardiac Surgery Department, Hygeia Hospital, Athens, Greece
Introduction: The study of the mechanical properties of the aneurysmatic ascending aorta is of significant clinical and scientific interest due to its role in systemic circulation and the risk of dissection or rupture. The focus of this research is the examination of the mechanical properties of the aortic wall under radial tensile stress, specifically to explore how the different layers respond to rupture propagation.
Materials and methods: 74 samples were collected from 12 patients (33–76 yr) at Hygeia Hospital. The experimental procedures were conducted at the Center of Clinical, Experimental Surgery & Translational Research (BRFAA). The number of specimens depended on the size of the tissue, with the left lateral quadrant yielding the fewest samples due to its morphology. For the mechanical testing,1 a fully automated tensile testing machine was used, equipped with a force cell with a maximum load of 500 g. The sampling frequency of the machine is set at 10 Hz.
Results: The graph presents the maximum force (Fmax) values in grams for both inner and outer layers across different anatomical regions of the aortic wall: anterior, right lateral (rl), posterior, and left lateral (ll). Fmax represents the maximum force the tissue can withstand before failure. Notable differences are observed between the inner and outer layers across the two areas. The anterior and left lateral regions were grouped together, as were the posterior and right lateral regions, due to their shared biomechanical properties in the aortic wall (Figure 1).
Comparison of Fmax (g) across aortic wall regions and layers.
Discussion: The study found that the intima-media interface exhibits significantly higher mechanical resistance, with greater Fmax and elastic modulus values compared to the media-adventitia interface, indicating that the inner layers of the aorta are stronger and more resistant to rupture. These findings align with previous studies2 that have demonstrated the mechanical vulnerability of the media-adventitia interface. Furthermore, regional differences were observed, with the posterior and right lateral regions exhibiting higher mechanical resistance compared to the anterior and left lateral regions, suggesting that the structural integrity of the aorta varies across different anatomical locations.
Conclusion: The findings highlight the critical role of the intima-media interface in maintaining the mechanical stability of the aortic wall, as well as the vulnerability of the media-adventitia interface to rupture initiation. The regional differences observed in this study also underscore the importance of considering anatomical location when assessing rupture risk. By integrating these findings with existing literature, this study contributes valuable data to the field and provides a foundation for future research aimed at improving the diagnosis and treatment of aortic aneurysms.
References
1. Sommer, G., Gasser, C., Regitnig, P., Auer, M., & Holzapfel, G. (2008). Dissection Properties of the Human Aortic Media: An Experimental Study. Journal of Biomechanical Engineering, 130.
2. Tong, J., Sommer, G., Regitnig, P., & Holzapfel, G. (2011). Dissection Properties and Mechanical Strength of Tissue Components in Human Carotid Bifurcations. Annals of Biomedical Engineering, 39(6), 1703-1719.
This research was implemented in the framework of the Action “Flagship actions in interdisciplinary scientific fields with a special focus on the productive fabric”, which is implemented through the National Recovery and Resilience Fund Greece 2.0 and funded by the European Union— NextGenerationEU (Project ID: TAEDR-0535983)
Structural modelling and simulation of abdominal aortic aneurysms accounting for intraluminal thrombus and calcifications
P. Sarantides*, A. Raptis* and C. Manopoulos*
*Laboratory of Biofluid Mechanics and Biomedical Technology, School of Mechanical Engineering, National Technical University of Athens, 157 72 Zografos, Greece
Introduction: Abdominal aortic aneurysms (AAAs) constitute a pathology of the aorta, which is characterized by a permanent increase in diameter. Although this pathology is well known, the factors influencing its development as well as the likelihood of vessel rupture remain highly unclear. In the present study, the influence of intraluminal thrombus (ILT) and calcifications is examined in the intense response of the arterial wall under normal loading conditions. The purpose of this process is to assess the role of each co-developing vascular pathology in the development of an index for estimating rupture risk.
Materials and methods: Based on the segmentation of computed tomography images, a 3D geometry of the AAA is created. The inability to capture the aneurysmal wall in computed tomography introduces challenges in constructing the overall aneurysm geometry. Therefore a new in-house software is used to integrate the patient-specific 3D geometry of calcifications within the aneurysmal wall, accurately simulating the real case. As a result a finite element analysis-compatible model is created, consisting of spatial meshes of three distinct entities (ILT, calcifications and aortic wall).1 For the aneurysmal wall a transversely isotropic hyperelastic Holzapfel–Gasser–Ogden nonlinear material is applied. The ILT is modeled through an isotropic linear elastic material, and lastly for the calcifications a brittle material model is adopted.2 An intraluminal pressure equal to the mean systolic pressure under normal conditions (80 mmHg) is applied, while the geometry is fixed on its distal and proximal ends.
Results and discussion: The deformation and developing stresses within the wall, thrombus, and calcifications are calculated. The stress concentrations and their maxima across the entire extent of the wall are evaluated. Ιt is evident that the locations of the stress maxima on the ILT and aneurysm wall surface, are not altered by the intraluminal pressure increase. Regarding the ILT, the highest values of stress are observed on the boundaries of the thrombus, where its thickness is minimized. The stress fields along the wall’s surface are modified in the case of integrating or removing the calcifications in the modeling process. There is evidence that the presence of calcifications in the aneurysmal wall can dictate high stress localization.
Conclusions: Following the analysis of the results, it is concluded that calcifications significantly influence local stress concentrations. In conjunction with the thrombus, which acts as a stress-relieving factor for the wall, it becomes evident that the stress distribution is significantly altered by the introduction of these factors (Figure 1).
Meshing of aneurysmal wall (white), intraluminal thrombus (red) and calcification (pink).
References
1. P. Sarantides, A. Raptis, D. Mathioulakis, K. Moulakakis, J. Kakisis and C. Manopoulos (2024) Computational Study of Abdominal Aortic Aneurysm Walls Accounting for Patient-Specific Non-Uniform Intraluminal Thrombus Thickness and Distinct Material Models: A Pre- and Post-Rupture Case. Bioengineering, 11:144
2. C. Luo, X. Yang and J. Li (2022) Mechanical Properties of Single-Crystal Calcite and Their Temperature and Strain-Rate Effects. Materials, 15:4613
This research was carried out within the framework of the Action “Flagship actions in interdisciplinary scientific fields with a special focus on the productive fabric”, which is implemented through the National Recovery and Resilience Plan Greece 2.0, funded by the European Union – NextGenerationEU (Project ID: TAEDR-0535983).
Exploring state-of-the-art deep learning architectures for blood flow prediction in pathological vessels
M. Athanasiou*, A. Raptis* and C. Manopoulos*
*Laboratory of Biofluid Mechanics and Biomedical Technology, School of Mechanical Engineering, National Technical University of Athens, 157 72 Zografos, Greece
Introduction: Accurate hemodynamic assessment is vital for understanding vascular disease progression. While computational simulations provide non-invasive evaluations, they require individual runs for similar anatomies and significant expertise and resources. Accelerating these simulations for real-time analysis could improve clinical decision-making. This study investigates deep learning architectures to predict hemodynamic fields in vascular pathologies.
Materials and methods: A physics-informed neural network (PINN) architecture was employed for hemodynamic predictions in 2D geometries of vessel stenosis. A multi-case PINN strategy was employed, in accordance with.1 The PINN was trained without the use of labeled data, utilizing the parameterized incompressible steady state Continuity and Navier-Stokes equations. The degree of stenosis was set to vary from 20% to 60%, and Reynolds number (Re) from 500 to 2000. To measure the accuracy, computational fluid dynamics (CFD) ground truth data were generated using COMSOL.
Moreover, a graph neural network (GNN) architecture was employed for hemodynamic predictions in 3D abdominal aortic aneurysm (AAA) geometries. A large training set of synthetic hemodynamic data was created, employing 3D patient-based AAA models from the repository of the SAFE-AORTA project and performed 3D transient hemodynamic simulations in SimVascular. To ensure the training set had sufficient variability, we applied varying flow rate inputs and resistance-compliance output parameters. The GNN was trained to predict velocity, pressure and wall shear stress (WSS) fields at peak systole, as well as the time averaged wall shear stress (TAWSS). To overcome the scalability limitation of MeshGraphNets, our model partitions large input graphs into smaller subgraphs to reduce training memory overhead.2
Results and discussion: Overall, the PINNs achieve very low errors but also fail to capture details in the hemodynamic field after the stenosis which may hide within accuracy metrics that are more general by design. Moreover, it was found that the biggest error in velocity prediction occur for higher stenosis degree and lower Re (Figure 1). On the contrary, pressure prediction accuracy drops for bigger Re but increases for bigger degrees of stenosis.
PINN vs CFD for u velocity with stenosis degree = 40% and Re = 750. Normalized root mean squared error of u velocity for various stenosis degrees and Re.
The partitioned GNN architecture seems to maintain the predictive accuracy of full-graph GNNs while substantially enhancing the scalability. Increased accuracy has been achieved for the prediction of hemodynamic quantities of interest on unseen patient geometries at a fraction of the time that a conventional simulation requires
Conclusions: Μulti-case PINNs enable a single training process to handle various geometries and flow conditions, but the accuracy is limited, locally at the stenosis area. The partitioned GNN seems to be a more efficient DL architecture for hemodynamic predictions. A combination of the two architectures will be tested in the future.
References
1. H. S. Wong, W. X. Chan, B. H. Li and C. H. Yap (2024) Strategies for multi-case physics-informed neural networks for tube flows: a study using 2D flow scenarios. Scientific Reports, 14:11577.
2. M.A. Nabian, C. Liu, R. Ranade and S. Choudhry (2024) X-MeshGraphNet: Scalable Multi-Scale Graph Neural Networks for Physics Simulation. arXiv preprint. arXiv:2411.17164
This research was carried out within the framework of the Action “Flagship actions in interdisciplinary scientific fields with a special focus on the productive fabric”, which is implemented through the National Recovery and Resilience Plan Greece 2.0, funded by the European Union – NextGenerationEU (Project ID: TAEDR-0535983).
Modeling of three-dimensional scaffolds for bone regeneration using CFD analysis
O. Ntousi*, P. Siogkas* and D. I. Fotiadis**, Fellow, IEEE
*Unit of Medical Technology and Intelligent Information Systems, Dept. of Materials Science and Engineering, University of Ioannina, Ioannina, Greece
**Biomedical Research Institute, Foundation for Research and Technology Hellas, FORTH-BRI, Ioannina, Greece
fotiadis@uoi.gr, o.ntousi@uoi.gr, psiogkas@uoi.gr
Introduction: Bone tissue regeneration is one of the largest challenges in regenerative medicine, particularly for large defects where natural healing is not successful. Conventional strategies are not effective, and new strategies are required. Tissue engineering (TE) combines biology and engineering to create scaffolds to promote bone healing. In this work, Computational Fluid Dynamics (CFD) is employed to model and optimize 3D scaffolds for bone regeneration. It simulates fluid flow through different scaffold geometries, analyzing the effect of inlet velocities on significant parameters. Two scaffold models, “PCL-50” and “PCL-250,” with different strut architecture and porosity, were compared at inlet velocities 0.05, 0.1, and 0.5 mm/s.1–3
Materials and methods: CFD simulation was performed employing ANSYS 16.2 software. The Navier-Stokes and the continuity equations were employed to simulate pressure, velocity, WSS, and permeability within the scaffold regions. The fluid was defined as cell culture media. A no-penetration boundary was applied to the scaffold surface, and an outer fluid domain minimized boundary effects. Three inlet velocities (0.05, 0.1, and 0.5 mm/s) were tested assuming zero outlet pressure.3
Results: Results showed that scaffold architecture plays an important role compared to inlet velocity in influencing permeability and WSS distribution. Between the two models, the “PCL-50” scaffold had higher permeability and higher WSS values, which are favorable to increased cell proliferation and bone tissue growth. On the other hand, the “PCL-250” scaffold had lower fluid flow within its pores. The results are in agreement with current experimental findings, confirming the effectiveness of the CFD-based method in assessing scaffold performance. In addition, in this work we can emphasize the importance of scaffold geometry design to enhance biophysical indications essential for tissue regeneration. Larger WSS zones, particularly around the unit cell junctions in the “PCL-50” design, reflect regions of enhanced mechanical stimulation, which is important for osteogenic differentiation (Figure 1).
WSS distributions in the “PCL-250” and “PCL-50” scaffolds. The distributions were induced by an inlet velocity 0.5, 0.1 and 0.05 mm/s.
Conclusion: This approach provides significant insight into scaffold geometry and flow condition impacts on biological performance The “PCL-50” scaffold showed the best permeability, similar to human bone, though inlet velocity variations had little impact. Permeability is crucial for osteogenesis and vascularization. WSS was mainly influenced by scaffold architecture, with a linear relationship to fluid flow rate. Higher WSS in the “PCL-50” scaffold supports better cell adhesion, proliferation, and osteogenic differentiation, promoting bone growth.2,3
References
1. S. Zhang, et al., ‘A review on the use of computational methods to characterize, design, and optimize tissue engineering scaffolds, with a potential in 3D printing fabrication’, J Biomed Mater Res, vol. 107, no. 5, pp. 1329–1351, Jul. 2019
2. D. Ali and S. Sen, ‘Permeability and fluid flow-induced wall shear stress of bone tissue scaffolds: Computational fluid dynamic analysis using Newtonian and non-Newtonian blood flow models’, Computers in Biology and Medicine, vol. 99, pp. 201–208, Aug. 2018
3. N. Jusoh et al. ‘CFD Simulation on Permeability of Porous Scaffolds for Human Skeletal System’, HumEnTech, vol. 1, no. 1, pp. 39–47, Feb. 2022
A preliminary comparison of unsupervised and deep learning approaches for automatic abdominal aortic aneurysm segmentation on CT images
D. Arampatzis*, E. Athanasiadis**, E. Kontopodis**, Ι. Theodorakopoulos***, I. Theocharakis**, S. Kostopoulos**, D. Glotsos**, P. Asvestas**, A. Raptis****, Ch. Manopoulos****, K. Moulakakis*****, J. Kakisis***** and I. Kalatzis**
*Department of Statistics and Actuarial – Financial Mathematics, University of the Aegean, Karlovasi, Samos, Greece
**Department of Biomedical Engineering, University of West Attica, Egaleo, Athens, Greece
***Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi, Greece
****School of Mechanical Engineering, National Technical University of Athens, Greece
*****Department of Vascular Surgery, Attikon University Hospital, National and Kapodistrian University of Athens, Athens, Greece
Introduction: Abdominal Aortic Aneurysm (AAA) is a life-threatening condition with a high mortality risk if ruptured, often developing silently.1 This study explores an unsupervised segmentation method based on image intensity patterns, comparing it to deep-learning models like TotalSegmentator2 for improved AAA detection.
Material and methods: The study includes 18 patients with AAA (≥40 mm) undergoing conservative treatment due to aneurysm size or surgical contraindications, regardless of cause, symptoms, or morphology. For AAA segmentation in CT images, the study combines conventional image analysis and deep learning. An in-house algorithm uses intensity-based and morphological operations to detect aorta without large training datasets. Additionally, the nnU-Net framework, TotalSegmentator,2 optimizes segmentation for high accuracy, allowing a comparison of traditional, feature-based, and AI-driven methods.
Results and discussion: The results showed that TotalSegmentator slightly outperformed the proposed method in extracting the 3D aortic volume (Jaccard Index 0.763 ± 0.261, Dice Index 0.828 ± 0.255), while the unsupervised approach demonstrated better slice-wise segmentation accuracy when the aorta was visually distinct (Jaccard Index 0.772 ± 0.183, Dice Index 0.855 ± 0.155).
The discrepancy arises from cases where the aorta blends with neighboring structures; the proposed method struggles in such instances, whereas TotalSegmentator, trained to recognize anatomical structures, can infer plausible aortic shapes even when visibility is poor.
Conclusions: The findings highlight the potential of deep learning models to improve aorta segmentation and indicate that fusing training-free and deep models’ segmentation could lead to a robust system of AAA detection and monitoring, enabling earlier, personalized interventions that reduce rupture risk and improve patient outcomes (Figure 1).
A CT slice depicting the segmented structures’ boundaries in white and the ground truth one in red.
References
1. Bobadilla, J. L. & Kent, K. C., (2012), Screening for abdominal aortic aneurysms. Adv.Surg. 46, 101–109
2. Wasserthal, J. et al. (2023), TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiol. Artif. Intell. 5, e230024
Keywords: X-ray CT, segmentation, aortic aneurysm
Acknowledgement
This research is carried out within the framework of the Action “Flagship actions in interdisciplinary scientific fields with a special focus on the productive fabric”, which is implemented through the National Recovery and Resilience Plan Greece 2.0, funded by the European Union –NextGenerationEU (Project ID: TAEDR-0535983).
Vision transformer for thyroid ultrasound image classification
M. Abadi*, H. Toubakh*, B. Benarabi*, N. Zerhouni**, M. Zervakis*** and M. Antonakakis***
*Department of Electronics and Telecommunications, Faculty of New Technologies of Information and Communication, University of Kasdi Merbah Ouargla, Algeria
**Institut FEMTO-ST, Département AS2M, University of Bourgogne Franche-Comté, Besançon, France
***School of Electrical and Computer Engineering, Technical University of Crete, Akrotiri Campus, GR-73100 Chania, Crete, Greece
Introduction: Thyroid nodule classification remains a significant challenge in medical imaging due to the variability in ultrasound images.1 Vision Transformer (ViT)2 have shown a promising potential for classification in biomedical imaging for diagnostic purposes. In this study, we apply a ViT model on the Algerian Ultrasound Images Thyroid Dataset (AUITD) for thyroid ultrasound classification.
Materials and methods: The AUITD dataset3 includes 1,472 benign, 1,895 malignant, and 171 normal thyroid ultrasound images from hospitals in Setif, Algeria, labeled by experts according to clinical standards. Due to the dataset’s variability and imbalance, we employed a ViT-Tiny model pretrained on ImageNet, chosen for its ability to capture long-range dependencies more effectively than CNNs. Images were resized to 224 × 224, normalized, and augmented. Class imbalance was handled using a Weighted Random Sampler and Focal Loss. Training was conducted over 100 epochs using the Adam optimizer with OneCycleLR.
Results: The proposed model achieved a 98% overall test accuracy, with high precision and recall for benign and malignant cases. However, the normal thyroid class exhibited lower recall (14%), indicating further dataset balancing or augmentation techniques needed. The confusion matrix and classification report reveal the strengths and weaknesses of the approach, highlighting the necessity of additional data for rare cases.
Discussion and conclusion: This study shows that ViT is well-suited for thyroid ultrasound image classification, achieving 98% accuracy. Its global attention mechanism enables better handling of complex patterns compared to CNNs, especially for benign and malignant cases. However, performance remains limited for the underrepresented normal class. Future work will focus on data balancing and self-supervised methods to improve generalization (Table 1).
ViT classification performance.
Precision (%)
Recall (%)
f1-score (%)
Support
Benign
92
100
96
60
Malignant
99
100
99
292
Normal Thyroid
100
14
25
7
Accuracy
98
359
Macro avg
97
71
73
359
Weighted avg
98
98
97
359
References
1. Gao, X., et al. (2022). Deep learning-based thyroid nodule classification in ultrasound images: A review of the methods and models. Frontiers in Oncology, 12, 888005.
2. Chen, J., et al. (2023). Medical image classification with vision transformers: A comprehensive review. Medical Image Analysis, 84, 102707.
Keywords: Vision Transformer, Thyroid Ultrasound, Deep Learning, Medical Image Classification.
Acknowledgement
This study was supported by “the framework of the Action “Flagship actions in interdisciplinary scientific fields with a special focus on the productive fabric”, which is implemented through the National Recovery and Resilience Fund Greece 2.0 and funded by the European Union—NextGenerationEU (Project ID: TAEDR-0535985).
The hematoma fluid and serum proteomes of chronic subdural hematoma patients
G. Mavrovounis*,**, M. Samiotaki***, I. Kakkos****, V. Aidinis***, G. Stranjalis* and T. Kalamatianos*,****
*Neurosurgery, School of Medicine, National and Kapodistrian University of Athens, Athens Greece
**Emergency Medicine, Faculty of Medicine, University of Thessaly, Larissa, Greece
***Biomedical Sciences Research Center ‘Alexander Fleming’, Athens, Greece
****Biomedical Engineering, University of West Attica, Athens, Greece
Introduction: Chronic subdural hematoma (CSDH) is a common neurosurgical condition. Numerous lines of evidence indicate that the subdural fluid characterizing CSDH partly contains blood and its products. Previous research has also identified proteins putatively involved in CSDH pathophysiology.1 Nevertheless, the proteome of CSDH remains poorly understood. Herein, we assessed and compared the hematoma fluid and serum proteomes of CSDH patients.
Materials and methods: Hematoma and serum from 7 patients were analyzed using mass spectrometry-based proteomics. Data processing and t-tests were performed in Perseus, creating volcano plots with FDR set at 0.05 and S0 at 0.1. Gene Ontology (GO) analysis for the hematoma was conducted using Gorilla.2
Results: 789 proteins were identified. 502 passed quality filtering and were included in the final analysis. Of these, 499 were detected in hematoma fluid and 461 in serum, with 458 proteins shared between both sample types. 41 proteins were unique to hematoma, and 3 to serum. The analysis identified 182 proteins, 167 of which had significantly higher levels in hematoma compared to serum and 15 in serum (Figure 1).
Volcano plot showing differential protein expression in hematoma vs. serum.
GO analysis indicated extracellular matrix organization and related terms as the most significantly enriched processes. Inflammatory pathways, including toll-like receptor and pattern recognition receptor signaling were also enriched. Coagulation and fibrinolysis related processes were also significantly represented. Additional highlighted pathways related to wound healing, vascular regulation, apoptosis, hormone/peptide secretion and the control of vascular tone and diameter.
Discussion: Enriched pathways related to matrix remodeling, innate immunity, and vascular regulation may underlie clinical features like membrane formation and postsurgical recurrence. Activation of toll-like receptor and cytokine signaling indicate inflammation and immune responses to blood degradation or injury. Coagulation and fibrinolysis pathways highlight ongoing hemostatic imbalance, while apoptotic and vascular terms reflect cell turnover and vascular instability. These processes are apparent during the postulated evolution of CSDH, from tissue repair and inflammation to bleeding and cycles of coagulation/fibrinolysis.1 Overall, several pathways and related proteins indicated by the present data highlight potential novel targets for biomarker and therapeutic strategy development.
Conclusions: We provide state-of-the-art proteomic data on CSDH. While the serum and hematoma proteomes exhibit considerable overlap, qualitative and quantitative differences are apparent. Our findings can lead to further investigation into biomarkers and therapeutic targets in CSDH.
References
1. Edlmann E, et al. Pathophysiology of chronic subdural haematoma: inflammation, angiogenesis and implications for pharmacotherapy. J Neuroinflammation. 2017;14(1):108.
2. Eden E, et al. GOrilla: a tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC Bioinformatics. 2009;10:48.
Introduction: Type 1 Diabetes Mellitus (T1DM) patients are using artificial pancreas (AP) implementations,1 which aim to replace or support pancreatic functionality by regulating insulin production. A significant amount of research has been conducted to explore glucose control options for commercial applications employing PID-based control formulations1 to optimize glucose production, particularly to insulin dosing calculation. This work presents an AP scheme using a continuous PID controller, which is shown to outperform rival particle swarm optimization (PSO)2 -tuned PID control schemes in a case study also incorporating exogenous glucose disturbance.
Materials and methods: Simulation has been performed in MATLAB/Simulink environment and includes the following parts:
1. Mathematical model for human body: T1DM patient is modeled by Bergmans’ Minimal Model1
2. Continuous glucose monitoring sampled every 1 min outputting a running mean of 5 samples.
3. Insulin pump algorithm:
where is bounded between 0 and .
4. Continuous Parallel PID controller is described by:
5. PSO framework (ITAE and MAPE obj. functions):
Metrics used: Controllers’ performance was evaluated using MSE, which quantifies the mean squared deviation from the setpoint, MAPE, which represents the mean absolute percentage error, and steady-state error (SSE), which corresponds to the error in the last sample for each of the two predefined time periods T0–Tdist and Tdist+1–Tend. MACA measures mean absolute controller actions to determine more conservative or aggressive behavior, while MAPA quantifies pump actions.
Experimental results: The case study models a T1DM patient consuming a 50 g-carbohydrate rich meal at Tdist = 700 min. Results of the simulation (Table 1 and Figure 1) show that the proposed control scheme leverages the controller’s dynamics to a greater extent by performing more aggressive control actions (higher MACA) compared to the other controllers. It also utilizes the pump more efficiently, reflected in higher MAPA. Consequently, the proposed scheme operates more efficiently, achieving better performance in steady-state error, MSE and MAPE. This improvement comes at the cost of a trivial increase in overshoot. Future work will evaluate the results across a wider range of virtual patient scenarios and will also focus on more advanced control techniques, such as Model Predictive Control (MPC) schemes.
Simulation metrics.
PSO-PID
PSO-PID
Proposed PID
Obj.Function (f)
MAPE
ITAE
-
T0 to Tdist
Steady-state error
3.030
3.033
2.071
MSE
11.970
11.981
8.180
MAPE
5.311
5.313
4.155
MACA
0.829
0.828
0.876
MAPA
17.497
17.495
18.313
Tdist+1 to Tend
Steady-state error
0.474
0.476
0.429
MSE
4.063
4.069
1.116
MAPE
5.563
5.569
4.601
MACA
1.379
1.379
1.559
MAPA
24.643
24.643
26.418
Full AP closed loop simulation results.
References
1. S. Echajei, et al. (2025) Implementation of PID control strategies on Bergman model representing Type 1 Diabetes-T1D, Commun. Math. Biol. Neurosci., 2025:1.
2. A. Kapnopoulos and A. Alexandridis (2022) A cooperative particle swarm optimization approach for tuning an MPC-based quadrotor trajectory tracking scheme, Aerospace Science and Technology 127:107725.
Keywords: Artificial pancreas,T1DM,PSO, PID control
Structured data entry workflow for medical equipment inventory management
A. Daskalaki*, S. Zisimopoulos* and A. Dermitzakis*
*Institute of Biomedical Technology (INBIT), Patras, Greece
Introduction: Efficient medical equipment (ME) management is crucial for health-care facilities to ensure optimal performance, safety, and regulatory compliance. An accurate and continuously updated equipment inventory serves as the foundation for effective HTM systems. The proposed inventory management process involves two key stages: (i) data acquisition and (ii) data entry. While data acquisition involves collecting photos of ME’s identification label and relative information, data entry focuses on structuring and integrating extracted information into the inventory system.
This work focuses on the data entry stage, which translates raw image-based data into structured inventory records. Key fields extracted from ME images include Model, Manufacturer, Serial Number, Unique Device Identification (UDI), Year of Manufacturing, and GMDN2/EMDN3 nomenclature codes. Among these, nomenclature assignment remains a complex and time-intensive process due to the variations between classification systems and the lack of standardized mapping.4 This study proposes a structured, hierarchical workflow to streamline the data entry process for large-scale national ME inventories.
Materials and methods: The proposed workflow is implemented by INBIT1 and follows a three-level pyramid structure. Each level varies in workforce size, specialization, and training requirements:
Bottom Level: Handles basic data entry tasks, where equipment images are matched with pre-existing “Triplets”—a combination of Model, Manufacturer, and GMDN/EMDN code—from a predefined database. If a suitable Triplet is found, it is assigned to the registered equipment.
Second Level: Addresses cases where no existing Triplet is available. In this stage, new Triplets are created using the available GMDN codes database. Once verified, these new entries become Local Temporary Triplets, accessible to the bottom-level team for future assignments.
Top Level: Manages complex cases flagged as requiring expert review. This includes handling new or emerging medical technologies requiring new nomenclature codes. Additionally, the top level supervises the entire workflow and validates new Triplets, which are then permanently integrated into the database as Global Validated Data.
INBIT currently utilizes a repository comprising 1,020 GMDN codes, 2,030 manufacturer entries, and 11,300 Triplets. The workflow ensures continuous expansion and refinement of this database, progressively reducing manual effort over time.
Results: The national ME inventory project is expected to take one year and is currently in its second month. At the beginning of data entry, approximately 35% of cases required second- and third-level processing. However, as the database has expanded and more Triplets have been created, this percentage has already started to decrease, currently at around 30%. This early result indicates that the structured approach is progressively minimizing the need for expert intervention, making data entry more efficient and less time-consuming.
Discussion: The structured workflow distributes workload efficiently. Initially, manual intervention is high, but as the database grows, most data is processed at the bottom level, minimizing expert involvement and improving overall efficiency.
Nomenclature assignment is a key challenge due to the lack of direct GMDN-EMDN mapping. The hierarchical approach ensures expert input at higher levels while maintaining rapid lower-level processing. The adaptable workflow allows resource redistribution as the database expands.
Conclusion: A structured, multi-level workflow improves ME inventory management efficiency and accuracy. This scalable approach streamlines large-scale data processing and addresses nomenclature complexities. As the database grows, expert intervention decreases, optimizing resource use and processing time. The proposed framework supports sustainable and efficient HTM systems.
4. A. Daskalaki, M. Marinou, A. Dermitzakis, Medical Devices Nomenclature Systems: Challenges and Considerations in Health-Care Equipment Inventory Management, Proceedings 5th ICEHTMC 2023
Keywords: Medical Equipment Inventory, Health-care Technology Management, Data Entry Workflow, Nomenclature Assignment, GMDN, EMDN
Creating a unified medical equipment inventory in Greece
S. Zisimopoulos*, A. Daskalaki* and A. Dermitzakis*
*Institute of Biomedical Technology (INBIT), Patras, Greece
Introduction: During the last decades, medical technology has become a key factor in all modern healthcare delivery systems. However, WHO 2022 Global atlas of medical devices shows that most countries worldwide do not have any form of national medical equipment (ME) inventory, and when data is available, it mainly refers to high capital value equipment.1 The Institute of Biomedical Technology (INBIT) has undertaken a pivotal role in the Unified Medical Equipment Management System (MEMS) section of a broader National Digital Transformation Project in the Greek Healthcare System. The scope of this project includes all 128 Hospitals of Greece’s public healthcare sector intended to be registered within a 10-month horizon.
Methods: Three Working Groups (WGs) of Biomedical Engineers have been assembled and trained in the inventorying process. These WGs are based in different areas (Athens, Thessaloniki and Northern Greece, Rest of mainland and Islands) for maximum geographic coverage. The process implemented for this project is an adaptation of other regional projects that have been successfully completed in the past,2,3 specialized for a nationwide scale (Figure 1).
ME registration process.
This process includes the creation of an Electronic Record for each individual ME. This is done by the WGs on a room-by-room basis, where each device is labelled with a tag bearing a unique national device ID that will be used throughout its lifecycle. The tag also contains a QR code that can be scanned with a specialized smartphone app to register new ME, view existing device details and report failures. Photographic record of every device is also created, including the unique ID label, manufacturer label, UDI and a full device photo. This will lead to the assignment of model, manufacturer and Global/ European Medical Device Nomenclature (GMDN/EMDN) groups, during the sequential data entry phase of the project, as the data are fed to the web-Praxis MEMS. Finally, data such as acquisition method, operating status, department/area of installation etc. are registered in collaboration with the hospital staff. Due to the challenges of the modern healthcare environment, a robust communication scheme had to be developed. A kick-off meeting with all hospital management units was performed to provide the overarching goals of the project. Each month, a list of hospitals to be registered is created and the Management and Biomedical Eng. Departments are informed in detail by the Social Security e-Governance (IDIKA). Sequentially, the final schedule is made by each WG, and the Hospitals are informed of the exact date and the registration process.
Conclusions: As of the time of writing, 35,084 devices have been registered in 34 Hospitals. These preliminary results indicate steady progress, highlighting the project’s feasibility. The implementation of a Unified MEMS across Greece’s public healthcare sector marks a significant step toward improving national ME practices. Taking advantage of digital transformation, this initiative enhances ME traceability, lifecycle management, and overall healthcare efficiency, through the creation of various national databases. As the project continues, its outcomes will provide valuable insights into future large-scale healthcare improvements that rely on evidence-based decisions such as centralized contracts, homogenized maintenance practices and equipment redistribution.
2. S. Zisimopoulos, A. Dermitzakis, C. Roilos and N. Pallikarakis (2022) Implementation of a Medical Equipment Inventory at a Regional Healthcare System in Greece, IFMBE Proceedings, 87:721-728
3. A. Dermitzakis, S. Zisimopoulos and N.Pallikarakis (2023) Implementation and Streamlining of the Medical Equipment Inventorying Process, J Global Clinical Engineering, Special Issue 5:157
Keywords: Clinical Engineering, Medical Equipment Management, Medical Device Inventory
Advancing universal accessibility in healthcare
G. Papadoulis**, D. Dumi-Sigalas**, C. Sintoris** and A. Dermitzakis*
*Institute of Biomedical Technology, Patras Science Park, Rio, Greece, 26504
**Interactive Technologies Lab, Department of Electrical and Computer Engineering, University of Patras, Campus Rio, 26504, Greece
Introduction: HOSP4ALL is an initiative that aims to enhance healthcare accessibility and inclusivity1 through innovative assistive technologies and user-centered design. Focusing primarily on individuals with visual impairments, it aspires to create hospital environments without barriers, facilitating patient autonomy. A pilot installation was carried out at the University General Hospital of Patras (UGHP), supported by the Panhellenic Association for the Blind (PST) and the Norwegian Association of the Blind and Partially Sighted (NABP), both of which provided highly positive feedback on system effectiveness.
Materials and methods: The core component of HOSP4ALL is a Bluetooth-based indoor positioning system. Bluetooth Low-Energy (BLE) beacons equipped with Angle of Arrival (AoA) technology2 deliver real-time location data to a user’s smartphone, guiding them through audio prompts and haptic feedback. Physical modifications, including tactile paving (haptic tiles) and Braille signage, further support safe, independent navigation. This design was developed collaboratively with user feedback from PST members, ensuring the solutions directly address the needs of blind and low-vision users.
Results: Initial findings suggest significant improvements in accessibility, safety, and user autonomy. Participants reported reliable guidance in navigating hospital corridors and identifying key service points. The PST formally praised the system as a “pioneering step toward equitable healthcare,” recommending expansion to other hospital departments. Additionally, visiting representatives from NABP commended the system’s ease of use and expressed interest in exploring its adoption in Norwegian healthcare facilities.
Discussion: Beyond its primary focus on assisting individuals with visual impairments, HOSP4ALL demonstrates the broader potential of real-time indoor navigation and asset tracking in healthcare. Challenges remain in scaling the system to encompass entire hospital complexes and in integrating it with existing digital services. Nevertheless, the strong user acceptance underscores the feasibility and importance of widespread deployment (Figure 1).
Navigation app environment. Left: the info and destination-selection menu where users choose clinics, ticket desks, entrances, etc. Centre: the live guidance map, providing a visual route for sighted or low-vision users. Right: step-by-step textual instructions that are also delivered through audio and haptic feedback.
Conclusions: HOSP4ALL shows how innovative technology, combined with inclusive design principles,1 can substantially improve the hospital experience for visitors with visual impairments. Positive evaluations from both the PST and NABP validate the pilot’s impact and indicate a clear path for wider adoption. Future work will expand coverage within UGHP, refine the application’s interface, and explore additional features such as SOS alerts, asset tracking, and data analytics for more effective hospital management.
References
1. Center for Inclusive Design and Environmental Access (2020) Principles of All-Inclusive Design in Healthcare Facilities.
The authors gratefully acknowledge the collaboration of the Panhellenic Association for the Blind (PST), and the Norwegian Association of the Blind and Partially Sighted (NABP). Their insights have been integral to the system’s development.
Mapping digital soft skills for implementation of digital scenario based learning educational resources
P. E. Antoniou*, N. Pandria*, A. Rodina-Theocharaki*, S. Konstantinidis* and P. D. Bamidis*
*Medical Physics and Digital Innovation Laboratory, School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki, Thessaloniki, Greece
Introduction: Digital Soft Skills (DSS) such as communication, collaboration, problem solving, and adaptability in digital healthcare context (telemedicine, digitalized health) are in high demand. These skills are essential for the digital workplace. Therefore, the European Higher Education area has continually increased expenditures in the training and promotion of digital soft skills among their student populations.1,2 Such training is especially effective through the use of Digital Scenario Based Learning (D-SBL).
D-SBL uses interactive digital scenarios, in the form of virtual patients, or other serious games, to support active learning strategies such as problem-based learning. Virtual patients have become, for quite a while now, a standard for medical education with digital enablers.3,4 This work presents the conceptualization and mapping of DSS to EU frameworks and healthcare terminologies for use with Digital Scenario Based Learning (D-SBL) in the context of the DISCERN-DSS EU funded project.
Methods: The educational activities that are planned in the DISCERN-DSS aim to implement 15 D-SBL VPs in the equivalent of 2 modules that will be used to educate more than 350 learners on digital soft skills. In that context our team set-out to create a conceptual mind-map for defining a terminology of digital soft skills that can then be used for the identification of DSS learning objectives in Higher Education curricula.
Initial results and conclusion: Taking cues from existing projects,5 the EU DIGICOMP framework6 and previous attempts to standardize the healthcare soft skills landscape, we have created a list of related DSS concepts. After that step we brainstormed, based on past hierarchies, and organized these terms into a conceptual mind-map.7 This work is currently being used to identify learning objectives for D-SBL, and a Delphi study is planned for prioritizing and organizing a concrete list of educational episodes that support DSS teaching.
Through D-SBL-based upskilling of healthcare professionals in DSS we will foster both cross-regional cooperation between EU and SE Asia and upskill the medical professionals of the crucial aptitudes that comprise digital soft skills.
References
1. Szczepanski, M. “Digital Europe programme: Funding digital transformation beyond 2020.” (2019)
3. Antoniou, P. E., Athanasopoulou, C. A., Dafli, E., & Bamidis, P. D. (2014). Exploring design requirements for repurposing dental virtual patients from the web to second life: a focus group study. Journal of medical Internet research, 16(6), e3343.
4. George, Pradeep Paul, Olena Zhabenko, Bhone Myint Kyaw, Panagiotis Antoniou, Pawel Posadzki, Nakul Saxena, Monika Semwal et al. “Online digital education for postregistration training of medical doctors: systematic review by the digital health education collaboration.” Journal of medical Internet research 21, no. 2 (2019): e13269.
6. European Commission: Joint Research Centre, Vuorikari, R., Kluzer, S. and Punie, Y., DigComp 2.2, The Digital Competence framework for citizens – With new examples of knowledge, skills and attitudes, Publications Office of the European Union, 2022, https://data.europa.eu/doi/10.2760/115376
7. Bouwmans M, Lub X, Orlowski M, Nguyen T-V (2024) Developing the digital transformation skills framework: A systematic literature review approach. PLoS ONE 19(7): e0304127. https://doi.org/10.1371/journal. pone.0304127
Introduction: Brain tumors and glioblastoma especially, the most common malignant primary brain tumor, are characterized by poor overall prognosis and rare long-term survival.1 To assist tumor treatment management and surgical planning, technologies simulating brain tumor growth are thus required. In this context, Digital Twins (DTs) represent a promising solution. They are designed to replicate physical systems intending to reproduce and forecast the actions of their real-world counterparts. In light of this potential, this study aims to develop a DT able to predict the progression of the brain tumor using deep learning.
Materials and methods: For this work, the LUMIERE dataset,2 an open-source single center longitudinal glioblastoma MRI dataset with expert RANO evaluation, is used. A subset of the dataset, comprising 39 patients and their MRI scans around 0, 15 and 41 weeks post-operation, is utilized to validate the feasibility of the proposed approach. The MRI scans include segmented tumor masks, which serve as input for the predictive modeling pipeline.
For the prediction of the progression of brain tumors, a Convolutional Long Short-Term Memory (Conv-LSTM) model is deployed. This type of neural network plays a key role in the task of future image frame prediction since it is capable of capturing both spatial and temporal correlations. In line with the aims of this research, Conv-LSTM has been previously employed successfully for the forecasting of pancreatic neuroendocrine tumor growth.3
The model is trained to take as input the tumor segmentation masks from the first two MRI scans of each patient and predict the tumor mask corresponding to the third scan. To prevent overfitting, the dataset subset is augmented by rotating the segmentation masks.
Results: The results represented are preliminary findings intended to demonstrate proof of concept. Using the tumor segmentation masks from two sequential MRI scans, the trained model can effectively generate the segmentation mask for a subsequent scan. The model was evaluated with five-fold cross-validation using the Dice similarity coefficient and Relative Volume Difference (RVD) metrics. The average evaluation metrics remain modest in initial experiments due to dataset imbalance and limited number of training samples. However, it is worth noting that there is a best-case performance in the five-fold cross-validation, achieving 84% Dice similarity coefficient and -5% RVD. As illustrated in Figure 1, the model is able to predict lesion growth patterns with sufficient accuracy producing a visualization of the progressed tumor.
Ground truth (left) and predicted (right) tumor segmentation mask of a patient’s third week scan.
Discussion and future work: This study demonstrates the potential of DTs that utilize deep learning models for predicting brain tumor progression and thus aiding clinical decisions and personalized treatment planning. Future work includes exploring the capabilities of Generative Adversarial Networks (GANs) in utilizing radiomic features from the tumor region, as a more efficient alternative to the model showcased in this research.
References
1. A.C. Tan, et al. (2020) Management of glioblastoma: State of the art and future directions, CA Cancer Journal for Clinicians, 70: 299-312
2. Y. Suter, et al. (2022) The LUMIERE dataset: Longitudinal Glioblastoma MRI with expert RANO evaluation, Scientific Data, 9: 768.
3. L. Zhang, et al. (2020) Spatio-Temporal Convolutional LSTMs for Tumor Growth Prediction by Learning 4D Longitudinal Patient Data, IEEE Transactions on Medical Imaging, 39: 1114-1126.
Keywords: Artificial Intelligence, Deep Learning, Brain Tumor, Digital Twins
Acknowledgment
The computational results were produced using Aristotle University of Thessaloniki High Performance Computing Infrastructure and Resources.
Modeling late-age adverse effects of breast cancer survivors from real-world EHR data
A. Zoiros*, G. Petridis*, B. Merkaj*, A. Billis* and P. D. Bamidis*
Introduction: The emergence of Real-World Data, i.e. Electronic Health Records (EHR), has gained attention as a valuable source for research analysis, however analytical applications remain problematic due to EHR collection focusing on clinical uses.1 Leveraging EHR data in the context of older breast cancer survivors requires addressing additional obstacles, such as the pronounced multimorbidity in the subject population, which results in mixed observations of treatment and adverse effects. Given these challenges, our study focused on predicting the changes in the frequency of adverse effects after hormonotherapy in older breast cancer survivors.
Materials and methods: Our methodology aimed at structuring an integrative framework for drawing explainable insights by pre-processing and analyzing timestamped data from EHR. In summary, the dataset we analyzed included a cohort of older breast cancer patients (>65 years old) treated with hormone therapy, alongside recorded comorbidities and/or symptoms observed before, during, and after treatment. Primary features were extracted from the raw data, corresponding to cohort information and records of 100 adverse effects. The observations involved 500 anonymized patients, aged 65 to 100 years at the time of diagnosis. However, exploring the real-world dataset revealed several quality issues concerning age inconsistencies, low cohort representation and large time gaps between the date of diagnosis and starting date of treatment. Abundance of noisy observations resulted in a significantly reduced final size of 238 subjects. Our target variable was set as the frequency change in subjects with a recorded condition between two periods: prior to and after the treatment’s start date. A custom transformation was also applied to compare identical values originating from different percentage magnitudes, aiming to capture relative growth and maintain control of the original range at the same time. Predictors were limited to the age at diagnosis and the corresponding cohort’s survival duration. The final model was selected by fitting all possible polynomial regression models up to second degree for each condition and evaluating performance metrics. In addition, interpretational simplicity and predictive power of the model were boosted, while bias and artifacts were minimized.
Results: Our analytical framework managed to support the implementation of informative models with desirable performance metrics, given the limitations of the datasets and the trade-offs of boosting interpretability. As an example, the model of anemia scored moderate adjusted R-squared and RMSE values, corresponding to 0.5 and 0.1. Coefficient p-values fell below the threshold of 0.05, confirming statistical significance. Notably, survival duration and age at diagnosis positively correlate with the frequency change in anemic observations (see Figure 1).
Predicted surface for the model of anemia.
Discussion: While the analysis of EHR data can contribute to explaining the increase of recorded condition frequency during breast cancer hormone therapy, mixed signals of comorbidities and side-effects observed in late-age cancer survivors introduce significant biases, which affect statistical inference.
Reference
1. Sauer, C. M., Chen, L. C., Hyland, S. L., Girbes, A., Elbers, P., & Celi, L. A. (2022). Leveraging electronic health records for data science: common pitfalls and how to avoid them. The Lancet Digital Health, 4(12), e893–e898. https://doi.org/10.1016/S2589-7500(22)00154-6
Keywords: EHR, breast cancer, multimorbidity, side effects, real-world data
Acknowledgements
This research was funded by the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 875329.
AI-powered personalised lifestyle coaching for nutrition and physical activity in breast cancer
V. Fiska*, AM. Krooupa*, S. Nikolopoulos* and I. Kompatsiaris*
*Information Technologies Institute (ITI), Centre for Research and Technology Hellas (CERTH), Thessaloniki, Greece
Introduction: Breast cancer (BC) is a condition in which abnormal breast cells develop uncontrollably and produce tumors.1 Approximately half of all BCs arise in women who have no identifiable risk factors other than their gender and age.1 Research shows that regular exercise and a balanced diet can reduce recurrence risk and improve survival in BC.2 Adopting such lifestyle changes can be challenging due to various personal and environmental barriers. The Capability – Opportunity – Motivation – Behavior (COM-B) model3 is a behavior change framework designed to identify key factors influencing behavior: individuals require capability, opportunity, and motivation to adopt and maintain a behavior. By assessing deficits in these areas, interventions can be tailored to address specific barriers to behavior change. This paper presents the Personalized Lifestyle Coaching (PLC) service architecture and concept as part of MELIORA,4 an EU-funded project aiming to empower women at risk of BC, patients, and survivors, to adopt a healthy lifestyle.
Methodology: Artificial intelligence (AI) and digital health technologies will apply COM-B principles to deliver personalized interventions. The proposed PLC Service architecture (Figure 1) features an AI-driven component that integrates user input—from self-reported questionnaires to real-time activity tracking—with output data generated from other AI MELIORA modules to identify individual barriers across the COM-B domains. It will deliver content and recommendations providing through the MELIORA mobile app personalized physical activity and nutrition guidance. For example, if the app detects low activity levels (an Opportunity and/or Capability deficit), it issues customized exercise suggestions, educational resources, and motivational prompts. All data collection processes will adhere to GDPR requirements and receive institutional ethics approvals, with informed consent obtained from participants prior to data collection. Decision-support algorithms—from rule-based systems to machine learning models (including supervised and unsupervised learning, neural networks and real-time adaptive algorithms)—will be evaluated in order to develop a strategy to map the input data to specific behavior-change strategies, with the optimal approach chosen based on pilot testing outcomes.
Discussion and conclusion: The integration of COM-B with AI through the proposed PLC can provide unique benefits compared to conventional behavioral interventions. AI allows for scalable personalization that adapts in real time to a user’s evolving needs. However, challenges remain in ensuring user uptake and adherence. Future research should focus on refining the proposed PLC adaptive algorithms to enhance user engagement while also evaluating long-term outcomes to establish robust, user-centred models for cancer and survivorship care.
2. Wolff, J., et al. App-Based Lifestyle Intervention (PINK! Coach) in Breast Cancer Patients-A Real-World-Data Analysis. Cancers (Basel). 2024 Feb 29;16(5):1020. doi: 10.3390/cancers16051020. PMID: 38473378; PMCID: PMC10930534.
3. Michie, S., van Stralen, M.M. & West, R. The behaviour change wheel: A new method for characterising and designing behaviour change interventions. Implementation Sci 6, 42 (2011). doi: 10.1186/1748-5908-6-42
Keywords: Breast cancer, Artificial Intelligence, Behavior Models, Personalized Interventions
Acknowledgement
The MELIORA project has received funding from the European Union’s Horizon Europe Research and Innovation Programme under Grant Agreement n° 101136791. The content of this article reflects only the authors’ views and the European Community is not liable for any use that may be made of the information contained therein.
Exploring the impact of age and weight on sleep quality: Insights from the URBANOME project
S. N. Ketseridou*, C. Plomariti*, C. Frantzidis**, E. Feleki***, M. Kermenidou****, I. Machairas*, G Kioselaki*, A. Chatzimpaloglou***, S. Karakitsios***, D. Sarigiannis*** and P. D. Bamidis*
*Laboratory of Medical Physics and Digital Innovation, School of Medicine, Aristotle University of Thessaloniki, Thessaloniki, Greece
**School of Engineering and Physical Sciences, University of Lincoln, UK
***Aristotle University of Thessaloniki, Department of Chemical Engineering, Environmental Engineering Laboratory, University Campus, Thessaloniki 54124, Greece
Introduction: Sleep is considered to play an important role in cognitive and physiological functions in humans. Poor sleep quality has been studied as a risk factor for weight gain.1 Furthermore, the quality of sleep has been found to decrease with healthy aging.2 The objective of the present analysis is to evaluate the relationship between sleep quality, age and weight, utilizing data acquired for the interventional study in Thessaloniki as part of the URBANOME Horizon 2020 project.
Materials and methods: A custom risk-scoring methodology identified 121 high-risk volunteers to be included in this analysis based on pneumological exams and a survey regarding Sleep Quality, Quality of Life, Mental Health, Physical Health and Urban Environment, Housing, and Personal Habits. Participants were provided with a Xiaomi Smart Band 7 and a portable sensor to measure particulate matter (PM), humidity and temperature. They were instructed to wear these sensors for 1 week to monitor activity, sleep quality and exposure to environmental factors. The objective was to evaluate the relationship between sleep quality and demographics using two-way ANOVA to analyze weekly durations of light, deep, and REM sleep, as well as wake time.
Results: The tests confirmed that the data followed a normal distribution (Table 1). A statistically significant interaction was observed between sleep stage, weight, and age (p-value = 0.03). As shown in Figure 1, the deep sleep ratio increases with age among participants with lower weight (adjusted p-value = 0.04).
Normality test results.
Data subset
p-value
Older & overweight
0.17
Young & overweight
0.25
Older & normal weight
0.11
Young & normal weight
0.98
Weekly deep sleep ratio by age and weight.
Discussion: The analysis suggests that older adults tend to have a higher deep sleep ratio than younger individuals, while weight affects deep sleep in both age groups, with higher-weighted older adults having a lower deep sleep ratio than lower-weighted, while the opposite effect is noted in younger participants.
Conclusions: The findings of this analysis provide valuable insights into the relationship between sleep quality, age, and weight, highlighting the importance of good physiology to maintain sleep quality across lifespan. Targeted interventions, such as promoting weight management and optimizing sleep environments, could be tailored by age group to enhance sleep quality, offering valuable guidance for urban health policy and personalized public health strategies.
References
1. Patel, S. R., & Hu, F. B. (2008). Short sleep duration and weight gain: a systematic review. Obesity, 16(3), 643-653.
2. Luca, G., Haba Rubio, J., Andries, D., Tobback, N., Vollenweider, P., Waeber, G., ... & Tafti, M. (2015). Age and gender variations of sleep in subjects without sleep disorders. Annals of medicine, 47(6), 482-491.
This work has been funded by the European Union’s (EU) Horizon 2020 research and innovation programme under grant agreement No 945391 URBANOME (https://www.urbanome.eu)
AI-enhanced tele-rehabilitation: Research-oriented modeling for fall risk, treatment effectiveness, and adverse events in balance disorders
E. Karapintzou*, V. Tsakanikas*, B. Nairn**, M. Pavlou***, D.-E. Bamiou**,***, T. Exarchos* and D. I. Fotiadis*, Fellow, IEEE
*Dept. of Materials Science and Engineering, Unit of Medical Technology and Intelligent Information Systems, University of Ioannina, Ioannina, Greece
**Faculty of Brain Sciences, University College London (UCL) Ear Institute, London, UK
***Dept. of Neuro-Otology, University College London Hospitals, London, UK
Introduction: Balance is a critical function for human mobility and independence.1 Balance disorders in older adults increase fall risk and reduce quality of life, with 28-42% of adults over 65 falling annually, emphasizing the need for effective prevention strategies.2 AI advancements are improving balance assessment and fall risk prediction.1 Our study aims to develop and validate AI-driven models for fall risk, treatment effectiveness, and adverse events within a tele-rehabilitation context.
Materials and methods: The study utilizes data from the TeleRehaB project, including retrospective data from the Holobalance project (HOL), the EMBalance project (EMB), and UCL, UK, covering patients aged 16-98 with balance disorders. The dataset comprises clinical, demographic, and symptom-related data. Data preprocessing involves feature encoding, imputation (Simple and Iterative), normalization, and handling of imbalanced classes via SMOTE. Multiple classifiers (XGBoost, KNN, Neural Networks, etc.) were trained using a 10-fold cross-validation strategy. Model interpretability was achieved through SHAP values to explain feature contributions. These models improve clinical decision-making and patient outcomes in rehabilitation settings by addressing key clinical endpoints, including Risk of Fall (RoF), Treatment Effectiveness (TE), and Adverse Events (AE).
Results and discussion: The study addressed three specific clinical use cases: predicting RoF, evaluating TE, and identifying AE. Table 1 details the models, endpoints, and metrics, with some models being particularly relevant to clinical relevance.
Results overview of research-oriented models.
End point
Source
Target
Model
Accuracy
Sensitivity
Specificity
Key aspect
RoF
EMB
FALL
XGB
0.94 ± 0.06
0.93 ± 0.07
0.94 ± 0.02
Tendency to fall
TE
UCL
HADA
XGB
0.94 ± 0.06
0.87 ± 0.10
0.87 ± 0.13
HADD
AE
HOL
SYMPTOMS
RF
0.86 ± 0.14
0.85 ± 0.13
0.80 ± 0.20
WHODAS 7
Results show a high accuracy of 94%, using the EMB dataset, for RoF with the most important feature being the tendency to fall. For TE, UCL data targeting HADA yielded 94% accuracy with the XGBoost classifier and HADD as a key aspect. Lastly, the RF with the HOL dataset and symptoms as the target reached an accuracy of 86% for AE. SHAP values reveal the impact of input variables on research-oriented models’ predictions. For instance, at RoF (Figure 1), Tendency_to_fall and age are dominant drivers, while features like the Dizziness emotional subscore, and the Hospital Anxiety and Depression Scale (HAD_A_D) have moderate influence.
SHAP value for risk of fall.
The study demonstrates that AI models can predict critical endpoints, aid early risk stratification, guide personalized rehabilitation, and improve long-term tele-rehabilitation outcomes, but face limitations like retrospective data and potential bias.
References
1. E. Antoniadou et al., “Reliability and validity of the mCTSIB dynamic platform test to assess balance in a population of older women living in the community,” J Musculoskelet Neuronal Interact, vol. 20, no. 2, pp. 185–193, 2020.
Keywords: AI predictive analytics, Fall Risk, Treatment Effectiveness, Adverse Events, Rehabilitation
Acknowledgement
This work has received funding from the European Union’s Horizon 2021 research and innovation actions under grant agreement No 101057747, as part of the TeleRehaB DSS project.
Swarm Learning with weak supervision for automatic breast cancer detection in MRI
M. Kalogeropoulos*, P. Papachristou* and A. Athanasiou*
*Mitera Hospital, Breast Imaging Department, Athens, Greece
Introduction: Breast cancer screening guidelines increasingly incorporate magnetic resonance imaging (MRI) for early detection,1 leading to a surge in imaging volume. Artificial intelligence (AI)-driven solutions can aid radiologists in interpreting these images. However, developing AI models is often hindered by data privacy constraints and the need for extensive manual annotations. This study investigates the integration of weakly supervised learning and Swarm Learning (SL) to facilitate privacy-preserving breast cancer detection.2
Materials and methods: A combination of weak supervision reducing reliance on detailed manual annotations and SL allowing decentralized model training across institutions was utilized. The Duke dataset was obtained from The Cancer Imaging Archive (TCIA), while the USZ, CAM, UKA, and Mitera Hospital Athens (MHA) datasets were collected by institutional partners within the ODELIA project under approved ethical protocols. The training dataset consisted of 1372 bilateral breast MRI exams from the US, Switzerland, and the UK. External validation was performed on 649 exams from Germany (UKA) and Greece (MHA). We benchmarked several deep learning models including 2D-ResNet50, 3D-ResNet18/50/101, 3D-DenseNet121, and multiple instance learning (MIL) variants (Att-MIL, ViT-MIL, and ViT-LSTM-MIL) (Figure 1).
AUROC comparison of models trained locally, centrally, and via Swarm Learning. 3D-ResNet101 with Swarm Learning achieved the highest performance.
Results: The study demonstrated that 3D-ResNet-101 achieved superior classification accuracy compared to other architectures. Furthermore, models trained using swarm learning (SL) consistently outperformed those trained on individual institutional datasets, highlighting the benefits of decentralized collaborative training. On the UKA dataset, the SL model achieved an AUROC of 0.807, compared to 0.743 (Duke), 0.538 (USZ), and 0.703 (CAM). On the real-world MHA validation cohort, the SL model again outperformed local models, achieving an AUROC of 0.821 versus 0.729 (Duke), 0.520 (USZ), and 0.673 (CAM). Additionally, explainability analyses using GradCAM++ and occlusion sensitivity analysis (OCA) indicated that SL models effectively focused on tumor-relevant regions, reinforcing their potential for improving breast cancer detection in MRI (Figure 2).
Swarm Learning outperforms local models in external validation on the MHA dataset, achieving an AUROC of 0.821.
Discussion: This study confirms that weakly supervised learning can be effectively applied to breast cancer detection in MRI. The integration of SL enables multi-institutional AI training without centralized data sharing, preserving patient privacy and addressing data heterogeneity. These findings support the feasibility of deploying collaborative AI frameworks in real-world clinical environments.
Conclusion: The combination of weak supervision and Swarm Learning enhances breast cancer detection in MRI by enabling decentralized AI model training across multiple institutions. Future work should focus on expanding dataset diversity, optimizing model interpretability, and evaluating the model’s performance in longitudinal settings. Additionally, addressing integration into real-world clinical workflows will be essential for deploying this approach at scale.
References
1. Mann R.M. et al., “Breast Cancer Screening in Women with Extremely Dense Breasts,” Eur Radiol., 2022.
2. Rieke N. et al., “The Future of Digital Health with Federated Learning,” NPJ Digit Med., 2020.
Keywords: Swarm Learning, Weak Supervision, Breast Cancer Detection, Medical AI, MRI
Acknowledgement
This study was supported by the ODELIA consortium.
A functional near infrared spectroscopy acquisition protocol
V. Nikoudi*, E. Kontopodis*, I. Kalatzis*, E. Ventouras* and A. Skouroliakou*
*Department of Biomedical Engineering, University of West Attica, Athens, Greece
Introduction: Functional Near Infrared Spectroscopy (fNIRS) is a novel neurodiagnostic technique that uses infrared radiation and physiologically relies on the hemodynamic response phenomenon.1 This pilot study proposes an fNIRS acquisition protocol for the evaluation of the prefrontal lobe activation during the performance of cognitive tasks.2
Materials and methods: The signals were collected with the 18 optode fNIRS 2000C Imager from BIOPAC with a sample rate of 10Hz. The acquisition protocol (Figure 1) was applied to five healthy adults. It started with a resting state recording that served as a baseline for the calculation of oxygenated hemoglobin (HbO) concentration variations. The cognitive task chosen was the N-back (for N=1 and 2), where the participant is asked for each presented stimulus whether it matches a stimulus N trials before.
N-back task acquisition protocol. RS (resting state), N-back 1 (cognitive task 1-back), N-back 2 (cognitive task 2-back) and Oxygraph: Time variation of HbO (red) and HbR (blue) relative concentrations, calculated after data preprocessing for the 18 optodes.
The raw data were preprocessed by applying a correlation based signal improvement filter, a band pass filter (0.01-0.1 Hz) and a linear detrending filter. The obtained signal was converted to HbO concentration using the modified Beer – Lambert’s law.
Results: The HbO concentration values for each subject were separated in three blocks corresponding to the resting state, the 1-back task period and the 2-back task period. Welch’s ANOVA and Games – Howell post hoc tests were performed on the three blocks. In all cases a statistically significant difference (p < 0.01) is observed between all three blocks. HbO concentration is highest during the 2-back task performance.
Discussion - conclusions: fNIRS is a non-invasive and flexible diagnostic technique that can record activation in the prefrontal cortex. Despite the limited number of participants, the proposed protocol demonstrates reliability, as the small sample rate along with the duration of each task provides a considerable number of measured values per person. Further research should be conducted to investigate the efficacy of the method in normal controls and cognitively impaired subjects, using additional variations of the acquisition as well as the data processing protocol.
References
1. T. Nguyen et al., ‘Investigation of brain functional connectivity in patients with mild cognitive impairment: A functional near-infrared spectroscopy (fNIRS) study’, Journal of Biophotonics, vol. 12, no. 9, p. e201800298, 2019; doi: 10.1002/jbio.201800298.
2. Rahman, M. A., Siddik, A. B., Ghosh, T. K., Khanam, F., & Ahmad, M. (2020). A Narrative review on Clinical applications of FNIRS. Journal of Digital Imaging, 33(5), 1167–1184; doi: 10.1007/s10278-020-00387.
Introduction: The field of wearable haptics has been benefiting from rapid technological development during the current and past decades. Multi-sensor data acquisition and fusion, novel sensors and advances in microcontroller capabilities have been technologically converging, allowing for the development of devices which can aid people to independently navigate through unfamiliar environments during their daily routine.1 Thus, wearable haptic systems could improve mobility and safety for visually impaired people (VIP) beyond the white cane.2 This paper presents a wearable-computing prototype device of our own development, aiming to assist VIP in navigating safely, whilst improving their mobility, confidence and autonomy.
Methods: The system’s efficiency was assessed in three different types of experiments. The first experiment included blindfolded navigation through an obstacle course with randomly placed cardboard boxes comparing the NaviSense prototype to a broomstick. The second experiment evaluated the system’s ultrasonic sensor obstacle detection capability for obstacles of different height. The final experiment tracked user improvement while using the system in which the participant was challenged repeatedly to cross the obstacle course, each time with a different but equivalent layout, to assess performance improvements as they gained experience using our prototype device. The device features a hat equipped with ultrasonic sensors for obstacle detection, vibration motors embedded on a belt to provide haptic feedback for the detected obstacles and the ATmega328P microcontroller to process the sensors’ data. The system’s software is implemented in C and runs on the microcontroller.
Discussion: This study investigated the NaviSense prototype’s potential to enhance the daily mobility and safety of VIP by using its sensing capabilities and vibration feedback. Experimental results, including the performance improvements shown in Figure 1, provided insights into the system’s performance and usability, indicating that user proficiency improves over time. We believe this is due to the relatively intuitive user interface and a reasonable learning curve. Further research is needed to confirm this assessment across multiple users and obstacle course scenarios.
Indicative graph of a participant’s improving performance in a navigational obstacle-course challenge, while using the prototype device for several sessions (subject 11).
Conclusions: Overall, the results support our initial conjecture that a wearable computing device taking advantage of ultrasonic sensing and intuitive tactile feedback could enhance the navigation and obstacle detection capabilities of VIP. However, the study’s limited sample size and short duration, makes it necessary for future work to be lengthier with more participants, to evaluate its long-term impact on independence and confidence.
References
1. F. Barontini, M. G. Catalano, L. Pallottino, B. Leporini, and M. Bianchi, “Integrating Wearable Haptics and Obstacle Avoidance for the Visually Impaired in Indoor Navigation: A User-Centered Approach,” IEEE Trans. Haptics, vol. 14, no. 1, pp. 109–122, Jan. 2021, doi: 10.1109/TOH.2020.2996748.
2. B. Leporini, M. Rosellini, and N. Forgione, “Haptic Wearable System to Assist Visually-Impaired People in Obstacle Detection,” in Proceedings of the 15th International Conference on PErvasive Technologies Related to Assistive Environments, Corfu Greece: ACM, Jun. 2022, pp. 269–272. doi: 10.1145/3529190.3529217
Introduction: There have been several approaches to develop technologically-enhanced gloves aimed at active or active-assisted hand rehabilitation.1 The most common solutions can broadly be classified into two groups: cable, pulley and motor systems implemented as either hard- and soft-robotics solutions, and solely soft-robotics approaches based on actuators made using soft materials.
Combining rehabilitation techniques with serious gaming and virtual reality (VR) has also shown promise, as users have been found to benefit both cognitively and physically from such an approach.2 Some studies have also demonstrated that having a system which combines the simulation of virtual objects with real, physical haptic feedback from a glove is also a promising approach. A pilot study conducted using a VR environment found that participants demonstrated a significant decrease in time required to perform standard functional tasks.3 Integrating these approaches into a single device, we believe it is possible to create a cheap assistive glove which leverages the advantages of each type of system.
Materials and methods: The prototype device comprises two main components. The pneumatic soft-robotics actuator, which provides curling assistance, and a hard-robotics cable-and-pulley system whose purpose is to detect finger position, emulate tactile resistance and generate the feeling of a virtual object being grasped by the patient. We use silicon for the construction of the pneumatic actuators, which are inflated and deflated by a miniature air compressor via electronically-addressable valves. The position of a patient’s finger is determined by fixing a 3D-printed strip of thin plastic at the top of each glove finger. The strip is serrated at one end in a pattern which matches the teeth of a gear, causing it to rotate it as the finger curls and extends (Figure 1).
The red arrow shows the movement of the plastic strip which rotates the gear, sensed by the Hall Effect sensors (blue) and interpreted as finger flexion. The strip’s range (red) is limited by the black stopper component, which is adjusted by a servo motor.
The haptic feedback of the glove for each finger is generated by a servo motor. The hook on the plastic strip prevents the finger from curling past a certain point, as demonstrated in Figure 2. The device is partially assembled and currently undergoing functionality testing. Initial testing includes sensor functionality, curling assistance, haptic feedback, data aggregation and compilation. Our development team includes physical rehabilitation experts, including a neurosurgeon, who will subsequently assess the glove’s utility in a clinical setting. Final pilot testing at the Medical School of the Aristotle University of Thessaloniki will involve actual patients undergoing physical rehabilitation, aiming to validate the glove’s effectiveness in realistic scenarios. During the pilot testing phase, we plan to assess the impact of prototype glove with and without haptic feedback, to determine this particular feature’s efficacy.
Discussion: The microcontroller, translational and client PC software is partially developed, while the prototype hardware is undergoing iterative improvements. Development of real-time data processing and integration of VR is in progress and initial testing.
Conclusion: Our prototype device combines soft pneumatic actuators -for curling assistance, increased safety and comfort- with hard robotic tendons for precise resistive pressure and extension assistance. We believe and aim to prove that it leverages the comparative advantages of each design approach. The device acquires and compiles data into a report suitable to support the attending physician’s decisions. User’s comfort is enabled through use of a breathable glove and cushioning layers between the user’s hand and the hard parts of the glove. The silicon actuators can be swapped out and cast to fit exact finger lengths of the user. The project is currently in the prototyping stage and will be fully assembled for testing shortly. Final pilot testing is planned in collaboration with the Medical School of the Aristotle University. Additionally, the glove will be run through rigorous and durability testing in order to confirm long-term efficacy.
References
1. M. Tiboni and C. Amici, “Soft Gloves: A Review on Recent Developments in Actuation, Sensing, Control and Applications,” Actuators, v. 11, no. 8, Art. no. 8, Aug. 2022
2. K. Mitsopoulos et al., “NeuroSuitUp: System Architecture and Validation of a Motor Rehabilitation Wearable Robotics and Serious Game Platform,” doi: 10.3390/s23063281.
3. H. C. Fischer, et al. “Hand Rehabilitation Following Stroke: A Pilot Study of Assisted Finger Extension Training in a Virtual Environment,” doi:10.1310/tsr1401-1.
Extended abstract: Attenuation or interruption of neural pathways leads to significant motor and sensory impairments. Conditions such as stroke and spinal cord injury disrupt communication between the central nervous system (CNS) and peripheral muscles, inhibiting voluntary muscle activation and sensory feedback. Recovery remains possible through neuroplasticity of the CNS, which enables the reorganization and formation of new neural pathways in response to stimuli.
Electrical Muscle Stimulation (EMS) involves the application of electrical pulse-trains to peripheral nerves or muscle, eliciting contractions even without voluntary control. EMS provides sensory input to the CNS, which is proven to enhance neuroplasticity. Functional Electrical Stimulation (FES) elicits contractions, synchronized with functional activities like walking or grasping. Over time, FES improves performance in Activities of Daily Living (ADLs), leading to independence and better quality of life.
Higher stimulation parameter values (frequency, amplitude, width etc.) typically elicit stronger contractions but increase discomfort, fatigue, and skin irritation. Electrode placement also affects contraction quality as small shifts can significantly alter outcomes.
An adaptive FES system is necessary to optimize electrode placement and stimulation parameters for efficacy and fewer drawbacks. This is achievable through automatic calibration at the start of each therapy session, allowing individualized treatment and home-based rehabilitation. This paper examines optimization methods and in vivo experimental results.
Experimental setup and method: Our prototype employs a microcontroller to drive two waveform generators for electrical stimulation. Muscle response is monitored via accelerometers on the bicep and wrist, measuring localized twitch and flexion-related acceleration, respectively. This setup enables real-time contraction evaluation across 15 electrode configurations and multiple frequency pairs.
Discussion: The graph in Figure 1 is a result of the described process.
Max acceleration by electrode configuration.
Results show significant variability across electrode configurations, highlighting the need for automatic calibration and signal parameter optimization. Stable configurations are identified through median peak acceleration and low variability. To enhance parameter selection, we plan to implement a rule-based AI system and fuse accelerometer data with EMG and flex sensor input. We intend to increase the number of subjects to strengthen our findings and explore real-time parameter updates using machine learning.
Reference
1. Shiyu Luo, Haonan Xu, Yi Zuo, Xiaogang Liu and Angelo H. All (2020) Review of Functional Electrical Stimulation Treatment in Spinal Cord Injury, NeuroMolecular Medicine, 22: 447–463.
Introduction: Cancer patients’ access to healthcare has been limited due to the Covid-19 pandemic,1 making the need for a comprehensive e-health monitoring framework apparent for both patients and health care professionals (HCPs). The eCAN Joint Action2 aimed at employing a patient-centered telehealth ecosystem, consisting of a teleconsultation platform, a patient mobile application for health data collection and a web dashboard addressed to HCPs for the monitoring of the patients. The present work presents the real-world deployment and insights gained from the Greek pilot within the context of the project, emphasizing challenges and lessons learnt.
Materials and methods: During the eCAN Greek pilot study cancer patients following the intervention arm (experiencing care through a telemedicine platform and remote monitoring of symptoms) participated for a period of 8 weeks. Their participation entailed the utilization of a mobile application. The patients were instructed to use the mobile application on a weekly basis to report their level of distress on a scale between 1 and 10, with 1 being no distress. Additionally, the QLQ-C30 questionnaire3 was integrated within the mobile application, to be filled in by patients on a bi-weekly basis. HCPs were instructed to remind the patients to fill out the questionnaires. Patients in the intervention group participated also in weekly psychological teleconsultation sessions through the Edumeet platform.4 After each teleconsultation, the HCP filled in a structured report on the web dashboard, following the SOAP framework. The questionnaire consisted of subjective and objective questions filled by the HCP to evaluate the psychological state of the patient and their experience with telehealth.
The control group followed their regular consultation in-person schedule, as designed by their HCPs. After each consultation, the HCPs filled in the SOAP report, involving the same sections as the one for the intervention group, except for a few questions addressing teleconsultation experiences. The control group reported their distress level to the HCPs, and they filled out the QLQ-C30 during the first and last consultation through the web dashboard, guided by the HCPs.
Results: 18 patients with advanced cancer were invited to participate in the study. Among them, six (6) were immediately excluded, due to non-willingness to participate. One (1) patient dropped out after the initial session, wanting to attend only in-person consultations. This resulted in eleven (11) patients completing the study, six (6) of them belonging to the intervention group, eight (8) females. The mean age of participants was 54,91 years, with a SD = 13,2. Participants of the intervention group missed a mean of 1,6 sessions due to health issues, except for one patient who only attended 5 teleconsultations. Distress data were consistently reported by only one patient, while a mean of 4,6 distress entries with an SD of 2,3 were missing from the intervention group. No patient completed all QLQ-C30 reports, one (1) patient of the intervention group reported less than 0.5 h, 3 reported 0.5-1.5 h, while the other 2 reported 1.5-3 h travel time to the hospital.
Discussion: The missing data observed during the pilot was coherent with the notes taken by the HCPs in the objective sections of SOAP, reporting the difficulty of the patients in the use of the mobile application. Patients’ low digital literacy and lack of access to technology were also reported by the HCPs, during the collection of lessons learnt through a SWOT analysis conducted after the pilot completion. While the far proximity of the patients to specialized cancer care should have acted as a facilitator for the adoption of telehealth in comparison to in-person psychological consultations and despite the satisfaction of the patients reported by the physicians, the adherence to the study protocol was extremely low. More in-depth research, collecting personal views and a higher number of participants is required to extract conclusions towards the application of telehealth services in the Greek reality.
References
1. Filip, ... Savage, W. K. (2022). Global challenges to public health care systems during the COVID-19 pandemic: a review of pandemic measures and problems. Journal of personalized medicine, 12(8), 1295.
3. Aaronson, ... Takeda, F. (1993). The European Organization for Research and Treatment of Cancer QLQ-C30: a quality-of-life instrument for use in international clinical trials in oncology. JNCI: Journal of the National Cancer Institute, 85(5), 365-376.
This work was funded by the eCAN Joint Action GA number 101075326.
Structural and functional optimization of pneumatics-based soft robotic glove using finite and boundary element analysis
A. Panagakis*, V. Fiska*, V. Mantiou*, K. Mitsopoulos*, A. Moraitopoulos*, K. Tagaras*, P. E. Antoniou*, P. D. Bamidis* and A. Athanasiou*
*Lab of Medical Physics and Digital Innovation (iMedPhysLab), School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki (AUTH), Thessaloniki, Greece
Introduction: Soft robotics rapidly evolve focusing on designing robots with flexible materials to enhance safety, adaptability and ease of use during robot-user interaction. Pneumatic soft robotic gloves (pSRGs) have been developed to assist individuals with hand disabilities by providing controlled movements through pump-fed 3D printed actuators. Important aspects of pSRGs include strength, flexibility and user comfort.1 In our design approach, Finite Element Method (FEM) is used to analyze stress distribution, deformation, and material performance, while Boundary Element Method (BEM) is applied to study boundary interactions and reduce computational complexity. This study aims to further develop the pSRG actuators of the NeuroSuitUp/HEROES project.2
Methods: FEM/BEM combined approach will be used to optimize the project’s pSRG development. FEM can discretize the whole model while BEM only the boundaries. In this project we will use the quadratic method, as it provides balance between accuracy and computational cost, in order to determine the values of stress, strain and deformation using FEM and noise reduction of the pump using BEM.3 Simulations in Abaqus utilize Silicone elastic material. We divided the elements using FEM and BEM with quadratic interpolation into 260000 elements while setting the steady point for finger bending and apply 30 kPa of pressure and gravitational pull.
Expected results and discussion: Post-simulation the points of increased load are checked for redesign and reinforcement needs. The results show the maximum bending of the actuator (Figure 1) as well as the places where increased load is applied. Post-model creation, the project will entail Arduino code development and testing values components creation. Finally, it will lead to 3D printing the actuator prototypes, pSRG assembly, testing and validation. The prototype must meet easiness to wear and user comfort needs. The initial testing and safety check will be conducted with the necessary calibrations in order to be deployed for further validation. Final goals include actuator integrity and strength, flexibility and safety as essential parts of the SRGs in order to achieve the project’s purposes of patient rehabilitation.
An individual pneumatic actuator's (“finger”) bending model and Modeling increased load points during bending.
References
1. Fumiya I & Laschi C (2011) Soft Robotics: Challenges and Perspectives. Procedia Computer Science, 7; 99-102
2. Mitsopoulos K, et al. (2023) NeuroSuitUp: System Architecture and Validation of a Motor Rehabilitation Wearable Robotics and Serious Game Platform. Sensors, 23, 3281
3. Hunter P & Pullan A (2001) FEM/BEM NOTES. Auckland, NZ: Department of Engineering Science
Keywords: Soft Robotics, Finite Element Method, Boundary Element Method, Actuator Design, Rehabilitation
Acknowledgement
This research project was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) https://www.elidek.gr under the “2nd Call for H.F.R.I. Research Projects to support Faculty Members & Researchers” (Project Number: 4391).
Monitoring scattered radiation for optimizing staff positioning in fluoroscopically-guided hip interventions
A. Belavgenis*, S. Skiadopoulos*, A. Karahaliou*, C. Dimitroukas*, V. Metaxas*, F. Efthymiou*, P. Megas** and G. Panayiotakis*
*Department of Medical Physics, School of Medicine, University of Patras, Greece
**Orthopaedics Clinics, University Hospital of Patras, Greece
Introduction: Dose monitoring is necessary for optimizing occupational radiation protection in fluoroscopically-guided interventions.1,2 The aim of this study is to monitor scattered radiation in case of a phantom-based study simulating fluoroscopically-guided hip interventions, for optimizing staff positioning.
Materials and methods: The phantom consisted of a titanium total hip replacement implant (femoral stem component with length 19.5 cm and a femoral head), embedded in a plastic water tank (L = 45 cm, W = 36 cm, H = 18 cm). The implant was placed on 6 cm PMMA plates within the water tank. The phantom was imaged with a C-arm Siemens Cios Select Fluoroscopy system, in pulsed fluoroscopy (5pps), employing three fluoroscopy modes [low dose rate (LDR), normal dose rate (NDR) and high dose rate (HDR)], field of view (FOV) = 23 cm, in undercouch geometry. Scattered radiation (dose rate) was measured utilizing an RTI ion chamber survey meter. Measurement points considered various angles around the phantom (0° to 315°; step: 45°) at a distance of 50 cm and height 85 cm from the ground. Dose rate was used for estimating gonad equivalent dose for various staff positions, considering average fluoroscopy time 60 s per intervention, normal fluoroscopy mode, annual workload of 250 interventions and presence of protective apron of 0.5 mm Pb equivalence.
Results and discussion:Figure 1 provides scattered radiation distribution (dose rate) for LDR, NDR and HDR fluoroscopy mode. Table 1 provides scattered radiation (dose rate) and corresponding annual gonad equivalent dose values for optimized staff positions during fluoroscopy-guided hip interventions. Optimized positions for Head surgeon and Assistant surgeon were selected by the lowest dose rate values, corresponding to 450 and 1350, respectively.
Scattered radiation (dose rate) distribution during low dose rate (LDR), normal dose rate (NDR) and high dose rate (HDR) pulsed fluoroscopy (no zoom).
Scattered radiation (dose rate) and corresponding annual gonad equivalent dose for optimized staff positions during hip interventions.
Medical staff
Staff position (distance/angle)
Scattered dose rate (μSv/h)
Gonad equivalent dose (μSv/year)
Head surgeon
50 cm/45°
49.41
10.29
Assistant surgeon
50 cm/135°
42.21
8.79
Nurse
100 cm/315°
18.97
3.95
C-arm operator
200 cm/270°
0.12
0.03
Conclusions: Monitoring spatial distribution of scattered radiation under intervention-specific conditions can be used for reviewing and optimizing staff positioning.
References
1. T. Dorman, B. Drever and S. Plumridge (2023) Radiation dose to staff from medical X-ray scatter in the orthopaedic theatre, Eur J Orthop Surg Traumatol, 33:3059-3065.
2. M. Dadabhoy, P. Waldock, T. Brammar, S. Pryke and R. Coomber (2022) Gonad irradiation from fluoroscopy during upper limb orthopaedic procedures in a UK District General Hospital. Br J Radiol, 95(1133):20211087.
Introduction: Stereotactic radiosurgery (SRS) is a radiotherapy treatment approach, employing high dose gradients and increased dose per fraction. In cranial SRS, stringent dose delivery spatial requirements (<1 mm) may be necessary to ensure treatment efficiency especially if small targets are involved.1 Target localization accuracy often relies on spatial fidelity of Magnetic Resonance (MR) images used in treatment planning due to superior soft-tissue contrast as compared to other imaging methods.
MR-related spatial distortion is distinguished in two types.2 Sequence-independent distortions are related to the nonlinearity of the gradient magnetic fields, while sequence-dependent distortions are related to static magnetic field inhomogeneities, susceptibility-induced spatial offsets and the chemical shift effect.
A proposed method for determining MR-related distortion is by scanning a phantom with high contrast markers (referred to as control points, CPs) and processing the obtained images with a localization algorithm to facilitate distortion detection and evaluation. For this purpose, a custom software was developed, using python programming language.
Materials and methods: A phantom made of acrylic was used for scanning. The phantom consists of 11 parallel acrylic planes which contain 1978 holes in total. A CP is defined as the geometric centroid of each hole.
An algorithm for CP was developed, requiring minimum user input. More specifically, the user enters a reference image (e.g., a geometrically accurate CT scan) and up to two MR scans with opposite read gradient polarity. A threshold level is selected by the user and the rest of the processing is automated. The algorithm applies different shape filters to distinguish between holes and other objects or noise in the created 3D binary image. The geometric centroids of all qualified objects are calculated in the 3D space, serving as CPs. The latter are matched among images after applying a rigid transformation matrix to spatially co-register all images. Finally, the algorithm statistically analyzes the detected distortion of both types and presents relevant distortion maps in a user-friendly graphical interface Table 1.
Statistics of absolute sequence independent distortion.
Results: Threshold sensitivity was estimated to be (0.031 ± 0.001) mm per 10 pixel intensity units. Moreover, the overall uncertainty of the CP localization procedure was estimated to be <0.2 mm.
As a feasibility study, the phantom was MR- and CT-scanned by a Philips Achieva 1.5T MRI scanner and a Siemens Somatom scanner, respectively. The sequence independent distortion was quantified by implementing the developed software and the reversed read gradient polarity method.2 An analysis of the detected distortion magnitude is presented in Table 1. An indicative distortion map, using one of the software’s plotting options is shown in Figure 1.
Quiver plot showing sequence independent distortion vectors on a plane at 61 mm from machine’s isocenter on Y axis.
Conclusion: The developed algorithm demonstrates adequate overall uncertainty of the order of 0.2 mm. Results are user-independent due to the insensitivity of the selected threshold. With the use of this software, the distortion magnitude of MR imaging protocols can be efficiently quantified and visualized for further evaluation in a user-friendly interface.
References
1. Grishchuk et al 2023, Pract Rad Oncol (13): 183-194
Introduction: Innovation in transitional care is essential for improving patient outcomes, reducing hospital re-admissions, and ensuring a seamless healthcare experience. However, its implementation is often hindered by various barriers such as financial constraints1 and communication2 challenges. This study explores the barriers and enablers in Greek healthcare system, incorporating the experiences of healthcare professionals and patients.
Materials and methods: A mixed-methods approach examined perspectives of patients and healthcare professionals in Greece. A structured questionnaire identified challenges and enablers, while semi-structured interviews provided deeper insights. Quantitative data was collected from health care professionals and patients through an online questionnaire that included multiple choice questions for the barriers and enablers of innovation in transitional care. Questionnaire responders were asked if they agree to be contacted for a follow-up interview that validated and expanded on the findings of the quantitative analysis.
Results: Nine (9) patients (8 female, 1 male) and nine (9) healthcare professionals (3 female, 6 male) answered the online questionnaire. Five (5) healthcare professionals and three (3) patients participated in the interviews.
From the patient perspective, the most commonly reported challenges included emotional stress (60%) and financial difficulties (40%). Support services during the transition were scarce, with only 30% receiving a follow-up call or access to a care support group. Satisfaction with the transition process was low, with 44% reporting they were unsatisfied or very unsatisfied. Most participants selected better community resources and financial assistance (70%) and personalized care (60%) as top areas for improvement.
Healthcare professionals identified the creation of clearer reimbursement and funding models to support innovation as the most important need (55%). Technology was seen as a key enabler, enhancing team communication (55%) and supporting remote monitoring (40%). Key areas for innovation included discharge instructions and medication management (66%), data sharing and interoperability (66%), healthcare coordination (55%), and post-discharge home-care services (55%). The main challenges to adopting innovation were staff resistance to change (55%) and difficulty integrating new technologies into existing systems (55%).
All interviews highlighted gaps in post-discharge care, especially the need for better follow-up and digital support. Resistance to new technologies was a major barrier for both patients and professionals, emphasizing the need for improved communication and training to support adoption and enhance care transitions.
Discussion: The study identified key barriers and enablers in care transitions. However, its findings may be influenced by the small sample size and uneven gender representation. Despite these limitations, several notable insights emerged. Financial challenges were a significant concern for both patients and healthcare professionals—patients struggled with affordability, while professionals sought clearer reimbursement and funding models. Additionally, patients lacked follow-up calls and support services, while healthcare professionals recognized technology as a crucial enabler that could address these gaps.
Conclusions: This study reinforces previous research within the context of the Greek healthcare system. Τhe findings highlight the urgent need for innovation in transitional care services, particularly through the integration of advanced technologies.
References
1. Fakha, A., Groenvynck, L., de Boer, B. et al. A myriad of factors influencing the implementation of transitional care innovations: a scoping review. Implementation Sci 16, 21 (2021). https://doi.org/10.1186/s13012-021-01087-2
2. Gass, B., McFall, L., et al. (2023). Perspectives of acute, post-acute, physician and community support providers on community collaborative efforts to improve transitions of care. Healthcare, 11(1), 100673. https://doi.org/10.1016/j.hjdsi.2022.100673
Keywords: innovation, transitional care, health tech
Acknowledgement
This study was conducted in the context of EVOLVE2CARE project, that has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement No 101158152
Morphological analysis of intraluminal thrombus of abdominal aortic aneurysm
E. Bei*, K. Politof*, K. Moirogiorgou*, M. Antonakakis* and M. Zervakis*
*Technical University of Crete/School of Electrical and Computer Engineering, University Campus, 73100 Chania, Hellas
Introduction: In cardiovascular physiology, the size, shape, and structure of the intraluminal thrombus (ILT) are particularly important since, in 70–80% of patients with abdominal aortic aneurysms (AAA), an ILT covers the vessel wall.1 However, the exact role of ILT in this context remains unclear and controversial. Building on previous studies that primarily examined the presence, size, and consistency of ILT,1,2 this study aims to clarify the dimensions of the two distinct layers of ILT: the luminal layer and the medial/abluminal layer. Additionally, it seeks to explore how these layers relate to the AAA maximal diameter and other relevant AAA parameters.
Methods: In a retrospective study involving 17 patients with AAA, STL files from 3D models of intraluminal thrombus were utilized to analyze its layers. The 3D ILT models used as input were generated from patient CT scans in the context of the SAFE-AORTA project. The structure of the intraluminal thrombus was classified into established layers, including the luminal layer and the medial/abluminal layer. To extract the luminal layer and calculate its exact surface, we employed various filters from the Paraview post-processing visualization engine, such as Surface Normals, Connectivity, and Threshold method. Paraview was also utilized to estimate the integration of ILT surface value and visualize the results of each patient.
Results: ILT was observed in nine out of 17 patients (age, mean ± SD: 70.23 ± 8.57, male: 94.12%). Representative distinct ILT layers are shown in Figure 1. Table 1 shows the results for the surface of both ILT layers in patients with AAA.
Representative figures of the distinct ILT layers. Up: both layers (luminal & medial/abluminal). Down: luminal layer.
Calculation of the surface of ILT layers in patients with AAA using the Paraview toolbox.
Discussion: Moderate negative correlations were observed between the entire ILT or its layers, and both age and neck diameter. Additionally, strong negative correlations were identified between the ratio of the medial/abluminal layer to the luminal layer and: (a) neck diameter (r = −0.73, p = 0.03) and (b) maximum AAA diameter (r = −0.65, p = 0.05). Future research involving a larger and more diverse population could further validate and expand these findings.
Conclusion: This approach may improve our understanding of ILT’s role in AAA progression and could benefit decision-making in pre-operative planning in endovascular repair.
References
1. A. Piechota-Polanczyk, et al. (2015) The Abdominal Aortic Aneurysm and Intraluminal Thrombus: Current Concepts of Development and Treatment. Front Cardiovasc Med., 2:19
2. VL. Nguyen, T. Leiner, FA. Hellenthal, et al. (2014) Abdominal aortic aneurysms with high thrombus signal intensity on magnetic resonance imaging are associated with high growth rate, Eur J Vasc Endovasc Surg., 48(6):676-84
This research was part of the “Flagship Actions in Interdisciplinary Scientific Fields” initiative, implemented by the National Recovery and Resilience Fund Greece 2.0 and funded by the EU’s NextGenerationEU program (Project ID: TAEDR-0535983).
Source-level EEG functional connectivity analysis in schizophrenia patients
G. Parisis*, P. Tsitsopoulos*,**, P. D. Bamidis*** and A. Athanasiou*,**,***
*MSc Neurosciences: Linking Basic Science with Clinical Application, School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki (AUTH), Thessaloniki, Greece
**2nd Department of Neurosurgery, Ippokratio General Hospital, School of Medicine, AUTH, Thessaloniki, Greece
***Lab of Medical Physics & Digital Innovation (iMedPhysLab), School of Medicine, Faculty of Health Sciences, AUTH, Thessaloniki, Greece
Introduction: Schizophrenia is a chronic psychiatric disorder characterized by disruptions in cognition, thought, and perception and is linked to regional brain disconnection.1 EEG signal analysis has offered unique insights into brain region interactions through neuronal oscillations and cross-frequency coupling with high temporal resolution. This study utilized source-level EEG analysis for localization of neural networks involved.
Methodology: The used dataset comprised of two groups (paranoid schizophrenia patients (F20.0) and healthy controls) of 14 subjects (7 m:7f each, age (27 ± 5 y.o.) and gender matched). Patients were unmedicated for a minimum of 7 days and were not very early stage (i.e., first psychotic episode) or suffered another organic or major neurological disorder. In this preliminary analysis, Multivariate Autoregressive (MVAR) modelling was performed on 2 recordings (15 min eye-closed resting state) from both groups. A 19-channel 10-20 EEG montage and sampling frequency 250Hz were used.
Preprocessing included mapping channel locations, down-sampling (100 Hz), re-referencing to average reference electrode and applying FIR filter at 1 Hz (shown to improve Independent Component Analysis (ICA) performance).2 Bad channels and/or time series were removed and after interpolating rejected channels, ICA and ICLabel were used to differentiate ICs that contribute linearly to produce the recorded signal and flag them as brain or artifact. Non-brain ICs were removed. Locations were co-registered on the MNI head model and dipoles were fitted for each of the ICs using DipFit (representing sources in 3D space). Connectivity metrics such as Dynamic Directed Transfer Function (dDTF) were calculated using the Source Information Flow Toolbox (SIFT), utilizing a MVAR model the order of which is specified using the Akaike Information Criterion.
Results: The EEG processing pipeline managed to reduce artifacts and localize neural sources using DipFit. Preliminary application of SIFT produced band-specific connectivity measures of metrics such as dDTF and Complex Spectral Density (S) as shown in the figure below (Figure 1).
Time-frequency grid of alpha activity. Frequency is represented on the y-axis and time on the x-axis. dDTF is plotted between sources (from row to column) and the diagonal highlighted row represents S.
Discussion: Data preprocessing was essential for isolating source-level neural activity and extracting reliable connectivity metrics. Future advancements will involve calculating these metrics across subjects for validation and integrating statistical tools to explore group-level differences.
Conclusions: This novel MVAR modeling approach provides a useful framework for studying functional connectivity for schizophrenia, as with other neurological conditions.3 By improving artifact rejection and source localization, we aim to enhance connectivity analysis reliability, in order to further advance functional neuroimaging schizophrenia research.
References
1. Schmitt A, Hasan A, Gruber O, Falkai P. Schizophrenia as a disorder of disconnectivity. Eur Arch Psychiatry Clin Neurosci. 2011; 261 Suppl 2:S150–4.
2. Winkler, I., Debener, S., Müller, K.-R., & Tangermann, M. (2015). On the influence of high-pass filtering on ICA-based artifact reduction in EEG-ERP. Conference Proceedings: IEEE Engineering in Medicine and Biology Society, 4101–4105.
3. Athanasiou A, et al. (2018) Functional Brain Connectivity during Multiple Motor Imagery Tasks in Spinal Cord Injury. Neural Plast. 2018:9354207
Introduction: Serious Games (SGs) have evolved since the 1970s from military training tools to applications in healthcare, research, and professional training.1 With advancements in technology, serious games have integrated new interaction methods, such as eye tracking, to enhance user experience and data collection.2 This study focuses on designing an ergonomically optimized UI and evaluating usability through eye movement efficiency, task performance, and user engagement. By combining eye-tracking data with user feedback, it aims to assess the role of eye tracking in rehabilitation-focused SGs and its impact on game design.
Methodology: The study follows a structured research protocol, including user interface design analysis, ergonomic assessments and best practices, and data collection on gaze behavior.3 To investigate the effectiveness of eye tracking in serious games, a rehabilitation-focused serious game, developed for the HEROES project 4. with optimized eye-based UI elements, incorporating the eye tracker Gazepoint GP3 Eye model for collection of fixations, gaze duration, and saccades. Data will be collected through eye movement patterns, user feedback, reaction time, UI navigation efficiency, user fatigue levels and task completion efficiency. Following the data collection, a statistical analysis will be conducted to measure the system’s effectiveness in enhancing engagement, reducing fatigue and the overall success rate, key factors for a successful UI design.
Discussion: The integration of eye tracking in serious games has demonstrated the proper UI methodologies that enhance user engagement and the user experience and reduce strain in the past while also providing crucial information that would have otherwise been hidden.2
Results: This research is expected to produce similar results, while also slightly translating them into this different category of serious games, namely for rehabilitation. A comparative analysis will be conducted between datasets collected before and after incorporating the eye tracking data to assess its contribution and the degree to which it impacts data quality and overall outcomes (Figure 1).
Experimental methodology diagram.
References
1. Giagloglou E., Panagiotis A., Macuzic I., Bamidis P. 2015. Improving Safety through gaming: A serious game’s application for risky professions. 49th ESReDA Seminar
2. K. Kiili, H. Ketamo, and M. D. Kickmeier-Rust, “Evaluating the usefulness of Eye Tracking in Game-based Learning,” Int. J. Serious Games, vol. 1, no. 2, Jun. 2014, doi: 10.17083/ijsg.v1i2.15.
3. Z. Chen, “Research on the application of ergonomics in UI interface design,” Appl. Math. Nonlinear Sci., vol. 9, no. 1, p. 20230787, Jan. 2024, doi: 10.2478/amns.2023.2.00787.
4. K. Mitsopoulos et al., “NeuroSuitUp: System Architecture and Validation of a Motor Rehabilitation Wearable Robotics and Serious Game Platform,” Sensors, vol. 23, no. 6, p. 3281, Mar. 2023, doi: 10.3390/s23063281.
Keywords: Eye Tracking Serious Games, User Interaction, Ergonomics in Gaming, Rehabilitation Technology
Acknowledgement
This research project was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) https://www.elidek.gr under the “2nd Call for H.F.R.I. Research Projects to support Faculty Members & Researchers” (Project Number: 4391). Special thanks to the members of the Biomedical Electronics, Robotics & Devices (BERD) group of the iMedPhysLab.
Brain network biomarkers in psychiatric diagnosis
V. Skopintsev*, H. Kondylakis*,**, G. N. Dimitrakopoulos*,*** and A. Athanasiou*,****
*Bioinformatics and Neuroinformatics, School of Science and Technology, Hellenic Open University, Patra, Greece
**Computer Science Department, University of Crete and Institute of Computer Science, FORTH, Heraklion, Crete
***Department of Informatics, Ionian University, Corfu, Greece
****Lab of Medical Physics and Digital Innovation (iMedPhysLab), School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki, Greece
Introduction: Mental disorders are conditions that affect a person’s thinking, emotions, behavior, and overall ability to function in daily life. These disorders can be temporary or chronic and may vary in severity. Electroencephalography (EEG) is a powerful tool for studying and diagnosing mental disorders because it records electrical activity in the brain in a non-invasive manner, requires a lower level of patient cooperation than functional magnetic resonance imaging, and is relatively inexpensive and easily available.1
Methods: The aim of this study is to conduct a narrative review examining the use of EEG and functional Magnetic Resonance Imaging (fMRI) in the diagnosis of mental disorders. The review focused on analyses that could provide useful information to clinicians by tracking neuroimaging biomarkers, and especially on functional brain connectivity, that can be detected in various psychiatric disorders. Searches were performed on Pubmed and Google Scholar repositories. During Jan-Feb 2025, 178 sources were examined, of which 84 were considered most suitable based on the keywords “SWNs”, “Functional connectivity”, “schizophrenia”, “bipolar disorder”, “ADHD”, “MDD”, “EEG”, “fMRI”. Sources dated before the year 2000 were excluded and emphasis was placed on those that had been published in international journals.
Discussion: The two basic principles on which the organization of the brain is based are functional segregation and functional integration.2 More intense activity is recorded in specific areas of the cortex that are not observed in normal conditions and alterations in small-world networks (SWNs) are present (Figure 1). The biomarkers detected in schizophrenia are EEG-microstates, the evoked potentials mismatch negativity (MMN) and P50. In bipolar disorder there is an increase in the waveforms Delta, Theta, Beta and Gamma and a decrease in Alpha. A characteristic biomarker in the case of major depressive disorder (MDD) is the evoked potential P300. Ιn the case of attention deficit-hyperactivity disorder (ADHD), the suggested biomarkers are the ratio of Theta to Beta waveforms and the evoked potentials P300 and contingent negative variation (CNV).
Conclusion: In conclusion, we can say that there are certain stable biomarkers that can be used for the diagnosis of schizophrenia, bipolar disorder and major depressive disorder, while in the case of ADHD the results are not yet clear.3
References
1. Freeman D, Garety PA (2003) Connecting neurosis and psychosis: the direct influence of emotion on delusions and hallucinations. Behav Res Ther.41:923-47
2. Khaleghi Ali et al, (2019). Abnormalities of Alpha Activity in Frontocentral Region of the Brain as a Biomarker to Diagnose Adolescents With Bipolar Disorder. Clin EEG Neurosci. 50(5):311-318.
3. Luo Y et al. (2020) Biomarkers for Prediction of Schizophrenia: Insights From Resting-State EEG Microstates. IEEE Access. 8:213078-213093
Introduction: Pediatric CT scans require careful attention due to children higher sensitivity to ionizing radiation. Balancing image quality and radiation dose is essential.1,2 This study evaluates the effect of image acquisition parameters (tube voltage and tube current) on image quality and radiation dose, aiming to define optimized protocols (that ensure diagnostic accuracy at the lowest possible dose) tailored to clinical indication.
Materials and methods: A Mini CT QC acrylic phantom (diameter: 15.25 cm), that incorporates structures of varying electron density (bone and soft-tissue equivalent materials), was used to simulate pediatric head. CT image acquisition was performed using a 128-slice scanner (Somatom go.Top, Siemens). Vendor-suggested pediatric head protocol (100 kVp, IQ = 350) was considered as a reference for comparison purposes. Additional exposure protocols included tube voltage ranging from 70 to 110 kVp (step 10 kVp) and a range of tube current values (mA) defined by user selected “IQ level” (100 to 400, step 50). Default iterative reconstruction parameters were adopted (strength S2 and reconstruction kernel for bone and soft tissue).
For each tissue-equivalent material, image quality was assessed quantitatively [Contrast-to-Noise Ratio (CNR), Signal-to-Noise Ratio (SNR), and noise] on reconstructed images corresponding to varying exposure settings. The relationship between tissue-specific image quality and radiation dose (CTDIvol) was investigated towards optimization.
Results and discussion: For all materials analyzed, at specific x-ray energy spectrum (kVp constant), CNR was improved with tube current increase (attributed to noise reduction), at the expense of increased radiation dose (CTDIvol was increased). Image quality indices (CNR and SNR) are differentiated with respect to material composition.
Further considering x-ray energy dependence, the relationship between image quality and radiation dose is material composition dependent (Figure 1, Figure 2), suggesting the importance of clinical indication-specific optimization.
Diagram illustrating the relationship between scan parameters, radiation dose, and image quality in case of bone equivalent material. The “reference protocol” (100 kVp) serves as a standard benchmark for comparison.
Diagram illustrating the relationship between scan parameters, radiation dose, and image quality in case of polyethylene material.
Conclusions: Optimizing image acquisition parameters in pediatric CT is essential for maintaining diagnostic image quality while minimizing radiation exposure. Optimized protocols tailored to tissue/lesion properties and clinical requirements ensure patient safety. Future research should focus on refining dose reduction strategies (further exploiting optimized image reconstruction parameters) and validating optimized protocols in clinical settings.
References
1. E. Sagy, S. Tschauner, C. Schramek and E. Sorantin (2023) Paediatric CT made easy, Pediatr Radiol, 53: 581-588
2. C. Granata, E. Sorantin, R. Seuri and C.M. Owens (2019) European Society of Paediatric Radiology Computed Tomography and Dose Task Force: European guidelines on diagnostic reference levels for paediatric imaging, Pediatr Radiol, 49(5):702-705
Study of the self-absorption effect in the gamma spectroscopic analysis of environmental samples
N. Merkos*, N. Salpadimos*, K. Karfopoulos* and K. Potiriadis*
*Greek Atomic Energy Commission (EEAE)/Environmental Radioactivity Monitoring Unit, Athens, Greece
nikosmerkos.mp@gmail.com
Introduction: One of the main challenges encountered in gamma spectrometry, particularly in the low-energy region, is the self-absorption effect. Self-absorption refers to the fact that photons emitted by the sample, are absorbed by the sample itself. This study examines aspects of this phenomenon in the gamma spectroscopic analysis of environmental samples. The objective was to develop a correction method for this phenomenon with direct implementation in measurements performed by the Environmental Radioactivity Monitoring Unit (ERMU) of the Greek Atomic Energy Commission (EEAE).
Materials and methods: Experimental and computational calibration of the detection system used for the measurements was performed, and a correction method was developed, based on estimating a corrective factor of the efficiency as the calibration source differs from the sample material leading to different self-absorption intensity. The extraction of the efficiency reduction factor values involves a double integral calculation.1 The parameters involved in this calculation were experimentally studied. One of them is the linear attenuation coefficient for which there is no complete database for environmental samples due to their variation in density and chemical composition. Within the framework of this study, an experimental setup and a method of determination were developed and tested for their accuracy.
Results: The contributing parameters were successfully determined, and the correction factors were calculated through numerical integration. Experimental values of the linear attenuation coefficient were determined for the calibration source and the samples of interest.
Discussion: An interactive user-interface application was created for the sufficient utilization of the correction method, which achieves the extraction of tabulated self-absorption correction factor values for any detection system, provided the user inputs a dataset of the contributing parameters. This can be employed for systematical correction for the self-absorption effect, which is a requirement of the ISO 20042:2019 standard2 implemented by the ERMU of EEAE.
Conclusions: The precise assessment of radioactivity in environmental samples is of high importance within the environmental radioactivity monitoring field to ensure the radiation protection of workers and the environment, in compliance with the international basic safety standards.
References
1. K. Debertin and R.G. Helmer, Gamma- and X-ray Spectrometry with Semiconductor Detectors. Netherlands, North-Holland, 1988.
Study of event-related potentials in EEG data in major depressive disorder
M. Spyrou*, P. Tsitsopoulos*,**, P. D. Bamidis*** and A. Athanasiou*,**,***
*MSc Neurosciences: Linking Basic Science with Clinical Application, School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki (AUTH), Thessaloniki, Greece
**2nd Department of Neurosurgery, Ippokratio General Hospital, School of Medicine, AUTH, Thessaloniki, Greece
***Lab of Medical Physics & Digital Innovation (iMedPhysLab), School of Medicine, Faculty of Health Sciences, AUTH, Thessaloniki, Greece
Introduction: The aim of this study was to investigate the differences in electroencephalographic (EEG) responses between patients diagnosed with major depressive disorder (MDD) and healthy controls.1,3,4 The dataset was provided by the MODMA organization6 and included EEG measurements from 24 MDD patients and 29 healthy participants, aged between 16 and 52 years. Such a comparison could benefit advancements in early diagnosis and improve the understanding of MDD.
Methods: Event-Related Potentials (ERPs) (Figure 2) were extracted from the dataset’s 128-channel EEG recordings. Data processing was performed on the EEGLAB toolbox in MATLAB.2 Sampling rate was 125 Hz, band-pass filtering was performed at 1-20 Hz, and, through extensions background noise, artifacts, and bad channels were recognized and eliminated, ensuring the removal of any unwanted data.
Results: Initial analyses revealed distinct differences in EEG patterns between the MDD group and the healthy control group (Figure 1). Statistical comparisons of event-related potential (ERP) amplitudes and latencies showed no significant deviations between the two groups, with p-values exceeding 0.05. These comparisons were conducted using permutation statistics with False Discovery Rate (FDR) correction to account for multiple comparisons.
Avarage erp of healtyh (left) and MDD (right) participants.
Erp of a single participant.
Discussion: Despite extensive analysis, our findings suggest no clear electrophysiological differences between individuals with MDD and healthy controls. The lack of significant ERP amplitude and latency differences may indicate that the specific ERP components examined in this study are not sensitive markers of MDD under the current experimental conditions. Further studies with larger sample sizes, different experimental paradigms, or additional neurophysiological measures may be required to better capture potential group differences.
Conclusion: These results highlight the complexity of identifying reliable biomarkers for MDD in EEG studies and underscore the importance of refining both methodological and conceptual approaches to studying depression using neurophysiological techniques.
Keywords: Major Depressive Disorder, Event-Related Potentials, ERP Analysis
Acknowledgements
We sincerely thank the Gansu Provincial Key Laboratory of Wearable Computing, Lanzhou University, China, for providing the MODMA Dataset.
References
1. Bruder G et al. (2002). Cognitive ERPs in Depressive and Anxiety Disorders during Tonal and Phonetic Oddball Tasks. Clinical EEG, 33. 119-24
2. The MathWorks Inc. (2022). MATLAB version: 9.13.0 (R2022b), Natick, Massachusetts: The MathWorks Inc
3. Nandrino J et al. (1996). Endogenous evoked potentials assessment in depression: a review. Eur Psychiatry. 11:357-68
4. Proudfit GH, et al. (2015). Depression and Event-related Potentials: Emotional disengagement and reward insensitivity. Curr Opin Psychol. 1; 110-113.
5. Sur S, et al. (2009 Event-related potential: An overview. Ind Psychiatry J. 18(1):70-3
6. Zhang, Y., et al. (2002) MODMA Dataset: A Multi-modal Open Dataset for Mental-disorder Analysis.
Wayfinding in hospitals - A narrative review of existing solutions and upcoming technological advancements
G. Sotiroudis*, P. D. Bamidis** and A. Athanasiou*,**
*2nd Department of Neurosurgery, Ippokratio General Hospital, School of Medicine, AUTH, Thessaloniki, Greece
**Lab of Medical Physics and Digital Innovation (iMedPhysLab), School of Medicine, Faculty of Health Sciences, AUTH, Thessaloniki, Greece
Introduction: Modern hospitals are vast and often complex environments. Their intricate layouts can make navigation challenging for patients and caregivers, especially those unfamiliar with the hospital’s internal organization. Wayfinding solutions seek to enhance the efficiency and quality of intra-hospital navigation for all visitors. Five key stages are involved in such an effort: (1) review existing literature on hospital navigation aids to understand current solutions and gaps, (2) conduct targeted surveys to assess the specific needs of different communities and the unique requirements of the hospital in question, (3) develop a unified action plan with input from all stakeholders, addressing legal, logistical, and technological considerations, (4) implement the navigation aid infrastructure, ensuring accessibility, reliability, and integration with existing systems and lastly, (5) monitor system performance continuously, analyzing user feedback to guide iterative improvements. This extended abstract focuses on building a foundational understanding that will inform subsequent stages of the initiative.
Materials and methods: A narrative review of relevant literature was conducted using the PubMed and Scopus online databases. Search terms included “hospital site navigation” and “hospital wayfinding.” Books relevant with site navigation were also utilized. In addition to academic sources, a search for already established hospital navigation service providers was performed using standard web search engines. The technologies and techniques employed by these providers are summarized in the following section.
Results: Hospital visitors’ wayfinding behaviour is influenced by the reason for their visit, as well as by space and time constraints. Visual cues that allow them to confirm time and distance are essential in emergencies.1 People usually follow the main route unless they already know a shortcut. Providing pre-visit information, such as directions to the hospital, parking availability, and customized floor maps with step-by-step guidance to the intended destination can significantly reduce wayfinding time and stress for patients and visitors.2 A variety of technological aids have been implemented to improve hospital navigation for visitors, spanning from traditional installments (stationary digital info kiosks) to solutions that integrate hospital infrastructure with data provided through visitors’ smartphone applications. These solutions provide turn-by-turn instructions using maps, combining audiovisual cues with multi-layered interactive maps, and can even implement augmented reality. The wireless technologies supporting these systems include Wi-Fi, Radio-frequency Identification (RFID), Global Positioning System (GPS), and Bluetooth Low Energy (BLE).3 An alternative smartphone-based approach, which avoids the need for continuous wireless infrastructure maintenance, involves the use of QR codes placed throughout the hospital. These can be scanned to instantly convey the user’s location to a navigation app.4 Modifications of the above solutions, such as embedding the wireless beacons in indoor flooring, or using specialized smartphones with haptic feedback, have been proposed to assist visually impaired visitors in hospital navigation. More forward-looking ideas include autonomous electronic wheelchairs or assistant robots.5 Existing hospital navigation service providers are listed in Table 1.
Navigation, wandering patient and healthcare staff safety system, asset tracking
Conclusion: Improving wayfinding in hospitals requires a multi-step, systematic approach. Recent technological aids offer a stress-free hospital visit, tailored to users’ custom needs.
References
1. Carpman JR, Grant MA (2016). Design that cares: Planning health facilities for patients and visitors (3rd ed.). Jossey-Bass.
2. Mollerup P (2016) The challenge of wayfinding in health care Environments. In: Hunter R, Anderson L, Belza B (eds) Community wayfinding: pathways to understanding. Springer
3. Lee E, et al. (2020). Journal of perianesthesia nursing: official journal of the American Society of PeriAnesthesia Nurses, 35:250-54
4. Tim HC, et al. (2016). World hospitals and health services: the official journal of the International Hospital Federation, 52, 7-9.
Introduction: Stereotactic radiosurgery (SRS) can be utilized not only for treating cranial tumors, but also for functional disorders including Parkinson’s disease (PD) tremor and trigeminal neuralgia (TGN).1 For this to be achieved, a high amount of beam energy must be focused on a very small target volume (∼mm)2,3 using conical collimators (physical cones) adapted on the Gantry Head of a Linear Accelerator (LINAC).4
Due to many limitations associated with physical cones,5 Popple et al.6 developed an alternative method known as Virtual Cone Technique. In this approach, the central two leaves of the multileaf collimator are positioned in a small gap, and non-coplanar arcs at different table positions and with collimator angles of 45 and 135 degrees are employed. The dose provided by the LINAC is proportional to the sine of the gantry angle, delivering a spherical dose distribution at the target volume, equal to physical cones. The aim of the current work was the evaluation of this technique.
Materials and methods: In this study, treatment plans were developed using the VC technique in Eclipse treatment planning system (TPS), utilizing AAA Version 15.6.06 calculation model on an anthropomorphic head phantom, with a Dosimetric Leaf Gap (DLG) value of 0.36 mm and a 2.1 mm leaf gap (LG) of the two central leaves of the MLC. The plans were recalculated and delivered on the OCTAVIUS 4D phantom, equipped with 1600SRS (PTW) ion chamber array, using a 10 MV (2400 MU/min) FFF beam of an Edge LINAC (Varian Medical Systems), equipped with a 120 leaf HD MLC. Thereafter, the measured dose distributions were compared to the ones calculated by the TPS, applying a gamma analysis for plan verification and deliverability of the technique. The gamma criteria utilized were 1 mm distance to agreement and 3% and 5% dose difference.
Results: The gamma analysis results showed a gamma passing rate of 90.1% & 95.1% (gamma criteria 1 mm/5%) and 88.3% & 84.8% (gamma criteria 1 mm/3%), for 50% and 10% threshold, respectively.
Discussion: Comparing our results with the results of the gamma analysis of Brown et al.5 after EBT3 film irradiation, they showed a higher gamma passing rate of 99.4% (gamma criteria 1 mm/2%) with 10% threshold. The difference between Brown’s study and ours could originate from their use of films for verification, which offers superior spatial resolution compared to a detector array. Our future steps will involve the implementation of film dosimetry in plan verification in combination with refining various model parameters for the optimal outcome.
Conclusions: Virtual Cone technique could successfully replace the use of physical cones in clinical practice for SRS treatment of functional disorders. Further refinement is required for the implementation of the technique.
References
1. Hynes PR DJ. Stereotactic Radiosurgery (SRS) and Stereotactic Body Radiotherapy (SBRT): Treasure Island (FL): StatPearls Publishing; [Updated 2023 Jul 25]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK542166/
3. Yasmeh J MC, Stea B Repeat LINAC-based Stereotactic Radiosurgery for Recurrent Trigeminal Neuralgia. Cureus. 14.
4. Luxton G, et al. Stereotactic radiosurgery: principles and comparison of treatment methods. Neurosurgery. 1993;32:241-59; discussion 59.
5. Brown TAD, Ayers RG, Popple RA. Commissioning a multileaf collimator virtual cone for the stereotactic radiosurgery of trigeminal neuralgia. J Appl Clin Med Phys. 2022;23:e13562.
6. Popple RA, Wu X, Brezovich IA, Markert JM, Guthrie BL, Thomas EM, et al. The virtual cone: A novel technique to generate spherical dose distributions using a multileaf collimator and standardized control-point sequence for small target radiation surgery. Adv Radiat Oncol. 2018;3:421-30.
SPECT/CT whole-body scintigraphy with 131 iodine in the assessment of recurrence of patients with differentiated thyroid cancer
A. Tsagkalidou* and S. Koukouraki**
*Bioinformatics and Applied Genomics Unit, Hellenic Pasteur Institute
**Department of Nuclear Medicine, Medical School, University of Crete, Greece
a.tsagkalidou@pasteur.gr
Introduction: Planar scintigraphic imaging (PS) with Ι-131 remains the gold standard in monitoring patients with well differentiated thyroid cancer (DTC) after initial surgery. However, interpretation of PS is often limited by non-specific radioiodine uptake, which may lead to false-positive findings due to physiological or benign uptake in normal tissues.1-4 Hybrid imaging systems, such as SPECT/CT, integrating functional data with anatomical localization from low-dose CT. This study assessed the added additional diagnostic value of post-therapeutic SPECT/CT+WBS in terms of assessing recurrence in patients with DTC.
Materials and methods: This retrospective study included 205 patients with DTC who underwent post-therapeutic WBS and SPECT/CT with I-131 at the Nuclear Medicine Department of the University General Hospital of Heraklion. SPECT/CT was performed immediately following planar WBS to further evaluate equivocal or suspicious findings. Diagnostic performance, lesion characterization, and clinical impact were evaluated. Imaging datasets and reports were manually reviewed and interpreted by nuclear medicine physicians. Statistical analysis was performed using IBM SPSS Statistics 25.0.
Results: SPECT/CT detected all lesions (N = 261, 100%) of iodine uptake observed on PS, and identified 82 additional foci (23.9%). Combining WBS with SPECT/CT provided higher diagnostic value in terms of interpretation of findings, as it clarified 94 equivocal findings. The incremental diagnostic value of metastases is also undisputed, as 29 distant foci detected by PS were identified and classified. Overall, SPECT/CT classified 1 equivocal PS, confirmed 3 negative scintigraphic studies, detected and confirmed PS findings by improving anatomic localization of sites of radioiodine uptake observed on planar WBS in 98 patients, reclassified abnormal findings on planar imaging by attributing them to normal uptake in 44 patients, while in 11 patients with positive planar scintigraphy, SPECT/CT was negative. Moreover, in 27 scans the hybrid method detected more foci of abnormal uptake, while in 21 patients it detected more foci of abnormal and benign uptake.
Conclusions: SPECT/CT improved the diagnostic performance by enhancing lesion detection and reducing false-positive findings. In post-ablation scans, SPECT/CT demonstrated clear incremental value over planar imaging by accurately characterizing equivocal uptake sites. Ιt identified unexpected sites of either neck lymph node or distant metastases and clarified the presence of residual thyroid tissue, thyroglossal duct remnants and the existence of a thymus gland. Τhe combination of WBS+SPECT/CT offers a powerful diagnostic tool for the detection and classification of regional and distant metastases, providing better anatomical localization and interpretation of radioiodine uptake. These findings are in consistent with recent systematic reviews and meta-analyses demonstrating that the addition of SPECT/CT to WBS significantly improves diagnostic confidence, enables more accurate staging and risk classification, and influences therapeutic decision-making in a substantial proportion of patients.4,5
References
1. Wong, Ka Kit, et al. “Incremental value of diagnostic 131I SPECT/CT fusion imaging in the evaluation of differentiated thyroid carcinoma.” American journal of roentgenology 191.6 (2008): 1785-1794.
2. Shinto, Ajit S., et al. “Incremental value of 131I SPECT-CT versus planar whole body imaging in patients with differentiated thyroid carcinoma.” Thyroid Research and Practice 12.1 (2015): 8-13.
3. Oh, Jong-Ryool, and Byeong-Cheol Ahn. “False-positive uptake on radioiodine whole-body scintigraphy: physiologic and pathologic variants unrelated to thyroid cancer.” American journal of nuclear medicine and molecular imaging 2.3 (2012): 362.
4. Chong, Ari, et al. “Clinical implications of adding SPECT/CT to radioiodine whole-body scan in patients with differentiated thyroid cancer: a systematic review and meta-analysis.” Clinical Nuclear Medicine (2022): 10-1097.
5. Xue, Yan-Li, et al. “Value of 131 I SPECT/CT for the evaluation of differentiated thyroid cancer: a systematic review of the literature.” European journal of nuclear medicine and molecular imaging 40 (2013): 768-778.
Blockchain-enabled federated learning models for secure and open medical data sharing: A comparative review
D. Bamidis*, E. Konstantinidis* and N. Athanasopoulos*
*Medical Physics & Digital Innovation Laboratory, School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki, Thessaloniki, Greece
Introduction: Secure and efficient sharing of medical data across institutions is vital for advancing healthcare diagnostics, research, and treatment outcomes. However, concerns over patient privacy, data ownership, and regulatory compliance have obstructed open data exchange. Federated learning (FL) has emerged as a promising machine learning paradigm that allows institutions to collaboratively train models without sharing raw data. At the same time, blockchain technologies provide immutable, decentralized frameworks to enhance trust, traceability, and transparency. This abstract summarizes findings from a comparative review of recent approaches combining FL and blockchain in the context of health data sharing.
Materials and methods: The review included peer-reviewed journal articles, IEEE conference papers, and case studies published between 2021 and 2025 that were identified through searches conducted on Google Scholar, PubMed and ScholarGPT, using a combination of keywords. A total of eleven papers were analyzed, including comparative models and three practical use case implementations. The rejected papers did not implement federated environment and focused on pre-trained models. The six retained covered the federated learning setting and applied blockchain technology to enhance transparency.
Results: The models varied in terms of blockchain architecture (public vs. permissioned), federated learning designs (asynchronous vs. FedAvg), and privacy mechanisms. Key findings include:
1. Security Enhancements: Models like T-BFL1 employed two-dimensional trust mechanisms, while another research work2 used parameter masking with asynchronous learning. Another framework utilized resilience metrics against cyberattacks demonstrating significant robustness.3
2. Data Storage and Validation: Some frameworks4 utilized IPFS and blockchain for decentralized storage; others used smart contracts oracles for trust enforcement.3
3. Use Case Effectiveness: A digital twin model that eliminates identity and raw data was proposed,5 offering privacy-preserving simulation-based collaboration.
Discussion: The integration of blockchain into federated learning provides several advantages. However, significant trade-offs exist. For instance, cryptographic techniques (e.g., homomorphic encryption) increase computational complexity. Moreover, consensus mechanisms like PBFT may not scale effectively in public blockchain environments. A notable advancement is the shift toward “identity-less” and “data-less” training via digital twins,5 which introduces new paradigms for privacy in medical AI. Distinct differences across the models lie in their trust architecture (reputation vs. smart contracts vs. oracles), deployment models (cloud vs. edge-based) and data contribution strategies (quality scoring, gradient auditing, masking).
Conclusions: The use of blockchain and federated learning is promising in forming a secure and privacy-aware ecosystem for medical data sharing. The reviewed models demonstrate varying strengths in scalability, security and interoperability. Future work should emphasize cross-domain operability and real-time use case validation in clinical settings.
References
1. Jiang, R., Zhang, et al., (2025). T-BFL model based on two-dimensional trust and blockchain-federated learning for medical data sharing. The Journal of Supercomputing, 81(2), 378.
2. Kumar, R., Bernard, C. M., et al., (2024). Privacy-preserving blockchain-based federated learning for brain tumor segmentation. Computers in Biology and Medicine, 177, 108646.
3. Myrzashova, R., Alsamhi, S. H., et al., (2024). Safeguarding patient data-sharing: Blockchain-enabled federated learning in medical diagnostics.
4. Gupta, M., Kumar, M., & Gupta, Y. (2024). A blockchain-empowered federated learning-based framework for data privacy in lung disease detection system. Computers in Human Behavior, 158, 108302.
5. Wickramasinghe, N., & Ulapane, N. (2025). A Solution for the Health Data Sharing Dilemma: Data-Less and Identity-Less Model Sharing Through Federated Learning and Digital Twin-Assisted Clinical Decision
Keywords: Blockchain, Federated learning, Health data sharing, Open data, Privacy-preserving AI
Acknowledgements
This work is partially supported by EC funded project RAISE (Project 101058479)
Estimation of dose in neonates under 1 kg in incubator from chest X-rays
M. Patsioti*, I. Antonakos*, M. Zachou*, G. Christopoulos** and E. P. Efstathopoulos*
*Department of Applied Medical Physics, Medical School, National and Kapodistrian University of Athens, Athens, Greece
**2nd Department of Radiology, Medical School, National and Kapodistrian University of Athens, Athens, Greece
Purpose: The aim of this study is to estimate the typical diagnostic reference levels (DRL) of radiation exposure values for chest radiographs in neonates under 1 kg in mobile imaging at a University Hospital in Greece and to compare these values with the existing DRL values from the literature.
Materials and methods: Patient and dosimetry data, including sex, age, weight, tube voltage (kV), tube current (mA), exposure time (s), exposure index of a digital detector (S) and dose area product (DAP) were collected from a total of 80 chest radiography examinations performed on neonates (<1 kg and <30 days old). All examinations were performed in a single X-ray system and all data (demographic and dosimetry data) were collected from the PACS of the hospital. Typical radiation exposure values were determined as the median value of DAP and ESD distribution. Afterwards, these typical values were compared with DRL values from other countries. Three radiologists reviewed the images to evaluate image quality for dose optimization in neonatal chest radiography.
Results: The mean value and standard deviation of DAP, from all examinations, was 0.13 ± 0.11 dGy·cm2 (range, 0.01 – 0.46 dGy·cm2) and ESD was measured at 11.55 ± 4.96 μGy (range, 4.01-30.4 μGy). The typical values in terms of DAP and ESD were estimated to be 0.08 dGy·cm2 and 9.87 μGy, respectively. The results show that the DAP value decreases as the exposure index increases. This study’s typical values were lower than the DRLs reported in the literature because our population had lower weight and age. From the subjective evaluation of image quality, it was revealed that most of the radiographs (over 80%) met the criteria for being diagnostic as they received an excellent rating in terms of noise levels, contrast, and sharpness.
Conclusion: This study contributes to the determination of typical dose values in a rare and sensitive category of patients (neonates weighing <1 kg) as well as information on the image quality of chest X-rays that were performed in this group (Table 1).
Comparison of typical values (median value of DAP and ESD) with literature.
Reference
Age/weight category
DAP (dGy·cm2)
ESD (μGy)
This study
<1 kg
0.08
9.87
R. Gilley et al.
<1 kg
0.03
-
T. J. M. Minkels et al.
600-1000 g
0.02
-
K. Alzyoud et al.
0–1 y
-
130
L. Hora et al.
<5 kg
0.09
-
G. Compagnone
0 y
0.14
-
National Diagnostic Reference Levels in Japan (2020)
0-1 y
-
200
A. Schegerer et al.
<3 kg
0.03
-
RADIATION PROTECTION No 185
<5 kg
0.15
-
B. Mohsenzadeh et al.
<1 y
-
60
A. Bouaoun et al.
<4 kg
0.23
55.2
H. Kim et al.
0 y
0.5
-
References
1. European Commision. Radiation Protection No 185: European Guidelines on Diagnostic Reference Levels for Paediatric Imaging; Luxembourg Publications Office of the European Union: Luxembourg, 2018; pp. 1–122.
2. ICRP, International Commission on Radiological Protection. Diagnostic Reference Levels in Medical Imaging; ICRP Publication: Ottawa, ON, Canada, 2017.
Reinforcement learning for walking assistance control of a lower limb exoskeleton
D. Markoglou* and K. Ampountolas*
*Department of Mechanical Engineering, University of Thessaly, Volos, Greece
dmarkoglou01@gmail.com, k.ampountolas@mie.uth.gr
Introduction: Spinal Cord Injury (SCI) often leads to paraplegia, affecting both physical and mental health. Lower limb exoskeletons can aid rehabilitation and improve mobility, enhancing quality of life. Traditional models require crutches for balance, limiting upper limb freedom. Heavier, crutch-free devices offer slower mobility with minimal advantages over wheelchairs.1 This paper presents a Deep Reinforcement Learning (DRL) approach to ensure walking balance for the Hermes exoskeleton, eliminating the need for additional external support. Developed by the HERMES Team at the University of Thessaly, this project aims to create a robotic exoskeleton that promotes hands-free, independent mobility and supports activities of daily living (ADL) for individuals with (SCI).
Materials and methods: The proposed control system replaces the Hermes controller with an RL agent, whose policy is designed to find the optimal set of actions that maximize the cumulative reward, specifically aiming to make the exoskeleton walk in a straight line with minimal control effort. The simulation environment for this system is built using Simscape Multibody in MATLAB, which models the exoskeleton’s dynamics, including the flexion and extension of the hip, knee, and ankle joints. For training the agent, both Deterministic Policy Gradient (DDPG) and Twin-Delayed Deep Deterministic Policy Gradient (TD3) are utilized.2 These are model-free, off-policy RL methods that employ an actor-critic architecture. The actor learns the optimal policy by interacting with the environment, while the critic evaluates the actions taken by the actor.
Results: The agent underwent a total of 20 simulations, 10 for each algorithm. Table 1 presents the mean reward received by the agent during the entire training period with a 95% confidence interval for each of the two algorithms. The TD3 algorithm achieved a higher mean reward (177.42 ± 32.7) compared to DDPG (158.87 ± 14.35), indicating superior performance. Notably, TD3 achieved higher rewards with fewer steps, demonstrating a more efficient gait pattern. In addition to the quantitative analysis, eleven gait parameters were assessed. Each parameter was rated on a scale from 1 to 10 as shown in Figure 1. A t-test was performed to confirm statistical significance, revealing that TD3 outperformed DDPG in parameters such as Speed (GS), Walking Pattern (WP), and Stability (St), while DDPG performed better in Ground Contact (GC), Knee Angle (KA), and Frontal Inclination (FI). Despite differences in individual parameters, TD3 (7.99 ± 1.17) outperformed DDPG (6.34 ± 1.33), showcasing its superior overall performance.
Comparison of DDPG and TD3.
Algorithm
Mean reward
Mean steps
TD3
177.42 ± 32.7
156 ± 24
DDPG
158.87 ± 14.35
185 ± 10
Gait parameters evaluation.
Conclusion: DRL has shown promise in enhancing Hermes exoskeleton walking balance, enabling crutch-free locomotion and more natural, independent movement. The TD3 algorithm outperformed DDPG, demonstrating its capability to enhance exoskeleton control for more practical and autonomous rehabilitation use.
References
1. S. Luo, G. Androwis, S. Adamovich, E. Nunez, H. Su, and X. Zhou (2023) Robust walking control of a lower limb rehabilitation exoskeleton coupled with a musculoskeletal model via deep reinforcement learning, Journal of NeuroEngineering and Rehabilitation, 20:34.
2. S. Fujimoto, H. van Hoof, and D. Meger (2018) Addressing Function Approximation Error in Actor-Critic Methods, arXiv:1802.09477.
Designing a wearable lower body distributed sensor system using data fusion from SEMG and MARG
V. Triantafyllidis*, K. Mitsopoulos*, V. Fiska*, G. Stamboulis*, V. Mantiou*, A. Moraitopoulos*, L. J. Hadjileontiadis**, D. Kugiumtzis**, P. D. Bamidis* and A. Athanasiou*
*Lab of Medical Physics & Digital Innovation (iMedPhysLab), School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki (AUTH), Thessaloniki, Greece
**School of Electrical & Computer Engineering, AUTH, Thessaloniki, Greece
Introduction: A large part of the human population is experiencing health complications in regards to movement that are caused by damage of the nerve tissue. Neuroplasticity can be induced with the help of wearable robotics, in order to aid in rehabilitation.2 This work focuses on the kinematics and dynamics of lower limb motion, and how its analysis1 can provide a greater understanding of the impact of the pathology to the user’s movement. Through the real-time acquisition of surface Electromyography (sEMG) and Inertial sensor (IMU) measurements, in combination with a musculoskeletal kinematic model in OpenSim, we aim to optimise rehabilitation outcome through a detailed evaluation of the user’s kinematic status.
Materials and methods: The system consists of a wearable sensor modality of sEMG and MARG sensors, which gather kinematic and muscle activation data from stroke patients. This data is collected and processed by a Robot Operating System (ROS) package, enabling efficient sensor data management and distribution. The combined data is then forwarded to OpenSim, a software platform for biomechanics simulation which generates detailed musculoskeletal models and can describe the error factor in lower body movement. The proposed system, as well as a serious game application made in Unity, are combined to provide real-time biofeedback to the end user, and refine their use which results in increased immersion and engagement by the patients (Figure 1).2
The ROS - OpenSim system architecture.
Discussion: It is important to consider the challenges of optimizing sensor placement and calibration methods to minimize interference while ensuring high data quality and real-time transfer without sacrificing accuracy. Looking ahead, research should focus on enhancing sensor positioning, improving fusion algorithms, and validating performance using established motion capture systems. These advancements should contribute to enhance the system for effective use in neurorehabilitation and its applications in real-world scenarios.
References
1. Margaritis, F., Mitsopoulos, K., et al. (2024). Kinematic and Dynamic Analysis of Lower Limb Movement: Towards the Design of a Wearable Rehabilitation Assistant Device . Global Clinical Engineering Journal, 6(SI6), 62–67.
2. Mitsopoulos, K., Fiska, V., et al. (2023). NeuroSuitUp: System Architecture and Validation of a Motor Rehabilitation Wearable Robotics and Serious Game Platform. Sensors, 23(6), 3281.
3. Athanasiou A, Mitsopoulos K, et al. Neurorehabilitation Through Synergistic Man-Machine Interfaces Promoting Dormant Neuroplasticity in Spinal Cord Injury: Protocol for a Nonrandomized Controlled Trial JMIR Res Protoc 2022;11(9):e41152
Keywords: sEMG, MARG, data fusion, ROS, OpenSim, lower body movement, kinematics chain, burden of disease, wearable sensors, sensor system
Acknowledgement
This research project was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) https://www.elidek.gr under the “2nd Call for H.F.R.I. Research Projects to support Faculty Members & Researchers” (Project Number: 4391). Special thanks to the members of the Biomedical Electronics, Robotics & Devices (BERD) group of the iMedPhysLab.
Energy harvesting systems for self-powered prostheses
K. Birmpas*,**, K. Balntoukas*, A. Dermitzakis* and K. Moustakas*
*Department of Electrical and Computer Engineering, University of Patras, Patras, Greece
**Faculty of Biological Sciences, University of Leeds, Leeds, UK
Introduction: Powered prosthetic limbs enhance autonomy for individuals with limb loss, yet battery limitations remain a key barrier. Frequent recharging, weight, and bulkiness reduce usability and comfort. Energy harvesting systems—specifically Triboelectric Nanogenerators (TENGs)—offer a promising route to achieve self-powered prostheses1,3 by reclaiming biomechanical energy otherwise lost during movement.
State of the art: Many energy harvesting techniques have been developed to date, taking advantage mainly of the inherent biomechanical energy produced by biological systems:
Pyroelectric nanogenerators (PYENGs) work by utilizing the pyroelectric effect
Piezoelectric nanogenerators (PENGs) work by converting mechanical stress or strain into electrical energy through the piezoelectric
TENGs work by harnessing the principle of triboelectricity, where two dissimilar materials come into contact and separate, causing the transfer of electrons between them and generating electrical energy as a result of the resulting imbalance in charge.2
Among those, TENGs appear as the most promising alternative,2,4 being able to produce higher power in lower operational frequencies, while also harvesting energy not only from normal forces, but also from shear stresses which dominate the forces exerted at the joints.
Materials and methods: We propose a design incorporating TENGs at key locations in lower-limb prostheses (Figure 1), namely the foot (for ground reaction forces), ankle, and knee. The mechanical dynamics at these sites were simulated using OpenSim (gait2392_simbody.osim) and SCONE, modelling both generic and amputee gait.
Erp of a single participant.
Results: The loads at the TENG installation sites (joint reaction forces and ground reaction forces) were extracted through inverse dynamics, static optimization and joint reaction analysis algorithms.
The maximum forces reach up to 3000 N, while the shear element reaches hundreds of N (the shear part could not be effectively harnessed if PENGs were used)
This approach however does not take into consideration the fact that our model should simulate the gait of a prosthetic-bearing amputee. For this reason, the H0914M.osim3.osim model was modified accordingly in two ways and a forward simulation was performed using the SCONE software.
1. Stiff leg approach:
The knee joint was “locked” simulating the first prosthetics
2. No-muscles approach:
To further validate the results the simulation was performed erasing the muscles moving the shank and foot, turning it into a prosthetic.
We appreciate that the more accurate the simulation, the higher the exerted forces that will be transformed to useful energy, as shown in Figure 1 leading up to 18.45 W produced.
Furthermore, we suggest incorporating TENGs at more sites, such as the wheelchair the amputee uses when charging the prosthetic (Figure 2).
TENGs incorporated into wheelchairs.
Discussion: This study presents a framework for integrating TENGs into lower-limb prostheses, demonstrating the potential to reclaim considerable biomechanical energy during movement. While our in silico approach provides valuable insights, it does not fully capture user-specific variations in gait or the physiological implications of energy harvesting. Notably, even slight increases in energy demand may influence user comfort and the practical efficiency of the prosthesis. As such, future work should aim to refine these designs. Experimental validation and user-centred optimization will be key to translating this approach into clinically viable applications.
References
1. Wang Z (2013). Triboelectric Nanogenerators as New Energy Technology. ACS Nano
2. Ahmed A. et al. (2020). Triboelectric vs Piezoelectric Generator Comparison. iScience, 23(7)
3. Qu X. et al. (2021). Assistive Devices Powered by TENGs. J. Phys. Mater., 4(034015)
4. Zhang Z., Cai J. (2021). High Output TENGs for Human Motion. Current Applied Physics, 22
Keywords: Energy harvesting, Triboelectric Nanogenerators, Prosthetics, Self-powered systems, Gait simulation
Literature review of short-arm human centrifuge rehabilitation protocols
C. Plomariti*, P. E. Kartsidis*, G. Kioselaki*, I. Machairas*, C. Kourtidou-Papadeli* and P. D. Bamidis*
*Laboratory of Medical Physics and Digital Innovation, School of Medicine, Aristotle University of Thessaloniki, Thessaloniki, Greece
Introduction: The absence of gravity during long-term spaceflights, as well as its reduced effect on Earth due to a sedentary lifestyle, has been shown to cause similar negative effects on the human body and cause pathologies in various physiological systems.1 The application of artificial gravity via the Short-Arm Human Centrifuge (SAHC) has been studied as a possible countermeasure against spaceflight-related dysregulation. However, hyper-gravity protocols implemented using the SAHC have also been used to treat various pathologies. We conducted a literature review to study employed protocols for mitigating related symptoms.
Materials and methods: A literature review was conducted using the PRISMA2 framework of literature published in English since 1990 on Scopus, PubMed, and IEEE Xplore. The search was performed using “artificial gravity”, “gravity therapy”, “microgravity,” and “rehabilitation” as keywords to identify relevant studies. Article selection and characterization were performed by two independent reviewers using pretested forms.
Results: The search identified 2445 manuscripts published from 1990 to January 2025. 2008 articles were initially screened. 60 articles met the inclusion criteria and were full-text screened, while 6 articles were not retrieved. From these articles, 5 SAHC rehabilitation studies were identified (Figure 1).
Article selection flowchart.
The selected articles varied in terms of purpose, methodology, and detail of reporting. 2 of the selected studies investigated SAHC as a countermeasure to simulated weightlessness, 1 used healthy participants, 2 articles presented case reports where SAHC was used on a stroke and an MS patient. Study completion times varied from 3 days in one of the simulated weightlessness articles to 3 months in the case of the other 4. One of the studies employed a continuous protocol, while the other 4 (80%) used an intermittent centrifugation protocol. The magnitude of the +Gz accelerations ranged between 0.5 and 2.7. The frequency of the centrifugation varied between daily (1 article, 20%) and a three-week exposure (4 articles, 80%). Finally, while most articles (80%) recorded one session/day, one study suggested 2-3 runs/day of exposure.
Discussion: Our search for rehabilitation protocols involving SAHC in the published literature aimed to be comprehensive while balancing practicality and available resources. It was not within the remit of this literature review to assess the methodological quality of individual studies included in the analysis. Based on the characteristics, range of methodologies, and reported challenges in the included articles, we have identified a lack of comprehensive studies exploring the use of the SAHC on patients as a rehabilitation tool. Although the results from the usage of SAHC towards the mitigation of the detrimental effects induced either by disability or simulated microgravity appear promising, more research involving large cohorts, strict protocols, and a variety of outcome measures is required.
Conclusions: This literature review of SAHC-implementing research described the various protocols available in the relevant literature. We have identified that the number of protocols used for rehabilitation is limited, with only 5 being described in the literature. Furthermore, the geographic distribution of SAHC in use is limited to 3 centers. Further research in the field will unlock the benefits of SAHC in mitigating the symptoms and accelerating the rehabilitation of patients.
References
1. Sandler, H., and Vernikos, J. (1986). Inactivity: Physiological Effects. Orlando, FL: Academic Press Inc
2. Haddaway, N. R., Page, M. J., Pritchard, C. C., & McGuinness, L. A. (2022). PRISMA2020: An R package and Shiny app for producing PRISMA 2020-compliant flow diagrams, with interactivity for optimised digital transparency and Open Synthesis Campbell Systematic Reviews, 18, e1230.
Keywords: Rehabilitation, Human centrifuge, Review
Linking semantic medical vocabularies to 3D assets for a ChatGPT powered, virtual reality based, medical education environment
P.E. Antoniou*, K. Tagaras*, A. Athanasiou* and P. D. Bamidis*
*Medical Physics and Digital Innovation Laboratory, School of Medicine, Faculty of Health Sciences, Aristotle University of Thessaloniki, Thessaloniki, Greece
Introduction: Virtual and augmented reality (collectively referred to as eXtended Reality - XR) have both been identified as widely available three-dimensional (3D) immersive technologies for anatomy education envisioning spatial relationships between them, providing a deeper educational outcome.1 AI successes provide promising indicators for AI as aids for medical students.2 Integration of ChatGPT powered virtual assistants to VR faces the challenge of content relevance and discoverability. There, medical semantic vocabularies such as the Unified Medical Language System (UMLS),3 can provide annotated information about biomedical and health related concepts. This work presents a service and messaging data flow implementation for annotating and discovering 3d assets based on the Unified Medical Language System metathesaurus.
Methods: This implementation enhances asset annotation and discoverability in a Unity3D-based VR medical education platform. A Node.js based interface queries the UMLS Search API.
Results: The user asks a question in the Unity-based VR environment, and the LLM AI response extracts key nouns. These nouns are sent to the UMLS Search API, which returns valid medical terms with Concept Unique Identifiers (CUIs). Matching terms are compared to an annotated 3D models library, ranking models by relevance. The most relevant 3D models are displayed to support the chatbot’s educational response. The overall architecture supporting this solution is presented in Figure 1.
TENG output characteristics for different prosthetic leg configurations.
Discussion and conclusion: This implementation is currently in the prototype stage. Future evaluation plans include teacher and student quantitative and qualitative feedback. A key limitation of its performance is its implicit dependence on the robustness of the linked LLM API. A key challenge for medical education technology resources is that of content relevance and recovery. Previous endeavors have identified the importance of semantic annotation for discoverability and repurposing of resources such as virtual patients.4 Freeform, queries to an LLM chatbot cannot be supported, even with an encyclopedic VR asset base if these assets are not discoverable through medical terms annotation. Robust communication between term repositories (e.g., UMLS) and VR education assets is the necessary intermediate step towards discoverable and repurposable, versatile medical education technology resources. Developing procedural educational content and self-directed learning supporting educational platforms is the natural extension and use of this work.
References
1. Pickering JD, Antoniou PE, Ntakakis G, Athanassiou A, Babatsikos E, Bamidis PD. Assessing the difference in learning gain between a mixed reality application and drawing screencasts in neuroanatomy. Anat Sci Educ [Internet]. 2022 May 6;15(3):628–35.
2. Antoniou P.E. Tagaras K., Bamidis P.D. (2024) Enhancing Anatomy Education through an AI-Powered Virtual Reality Auditorium In S. Konstantinidis, P.D. Bamidis (eds) Conference Programme & Book of Abstracts of the 5th International Conference on Medical Education Informatics (pp. 14) ISBN: 978-960-243-750-6
4. Dafli, E., Antoniou, P., Ioannidis, L., Dombros, N., Topps, D., & Bamidis, P. D. (2015). Virtual patients on the semantic Web: a proof-of-application study. Journal of medical Internet research, 17(1), e16.
Keywords: Medical Education, Virtual Reality, Large Language Models, ChatGPT, Semantic Annotation.
Monte Carlo determination of correction factors for BeO-based optically stimulated luminescence detectors in gamma knife radiosurgery small fields
V. Margaroni*, E. P. Pappas*, A. Drakopoulou* and P. Karaiskos*
*Medical Physics Laboratory, Medical School, National and Kapodistrian University of Athens, Athens, Greece
Introduction: Optically Stimulated Luminescence (OSL) dosimeters have been introduced in Quality Assurance (QA) programs for clinical dosimetry in radiotherapy applications.1 The aim of this study is to implement a Monte Carlo- (MC-) based framework for the determination of volume averaging effects and the output correction factors in the context of IAEA TRS-483 dosimetry code-of-practice,2 for commercially available BeO-based OSL radiation detectors, suitable for Gamma Knife (GK) dosimetry procedures.
Materials and methods: MC simulations were performed using the EGSnrc V2019 MC software package. The commercially available myOSLchipTM dosimeter (RadPro International GmbH, Germany) was modelled based on blueprints provided by the corresponding manufacturer. The detector consists of a small plastic case (external dimensions 10 x 10 x 2 mm3) which houses the sensitive volume made of BeO (square disk of 4.65 x 4.65 x 0.5 mm3 and density ρ = 2.85 g/cm3). The active volume of the OSL dosimeter was positioned at the center of a spherical water phantom (diameter: 16 cm), which coincided with the Radiation Focus Point (RFP) of the GK system. Phase space files for three collimator sizes (4 mm, 8 mm and 16 mm) of a GK irradiation unit were used as the source models.3 The 4 mm and 8 mm collimators were used as the clinical fields (fclin), while the nominal 16 mm collimator was regarded as the machine-specific reference field (fmsr). For the determination of output correction factors for relative dosimetry,2 the absorbed dose in a small water cavity with a radius of 0.25 mm was calculated. The volume averaging effect for the available collimator sizes was also quantified. All calculations were performed for the three cardinal orientations of the detector: (i) axial, (ii) coronal, (iii) sagittal (Figure 1).
(a) The water spherical phantom with the OSL dosimeter in three nominal cardinal orientations: (b) axial, (c) coronal (d) sagittal.
Results: The MC calculated correction factors are presented in Table 1. Volume averaging effect was found to have a significant impact on dose measurements, reaching up to 26% for the 4 mm collimator.
MC calculated correction factors 2. in a water phantom for the GK-PFX irradiation unit. The 16 mm collimator corresponds to fmsr. Corresponding overall combined uncertainties at the confidence 68% level are shown in the parentheses.
OSLD orientation
correction factor2
fclin: 8 mm collimator
fclin: 4 mm collimator
Axial
1.008 (5)
1.134 (6)
Coronal
1.011 (5)
1.186 (6)
Sagittal
1.010 (5)
1.184 (6)
Conclusion: correction factors for the BeO-based OSL detectors and all cardinal orientations were determined, contributing to data availability for relative dosimetry in GK radiosurgery.
References
1. Kry S F et al 2020, Med. Phys. 47 e19–51.
2. Palmans et al 2017, IAEA TRS-483, Vienna
3. Pappas E P et al 2016, Phys Med. Biol. 61 1182–203
Keywords: dosimetry, OSL, small field, correction factors, Gamma Knife, radiosurgery
Acknowledgements
RadPro International GmbH are acknowledged for providing detailed schematics of the corresponding OSL dosimeters. This work was supported by computational time granted from the Greek Research and Technology Network (GRNET) in the National HPC facility – ARIS – under project ID pr017028.
Agile methodology for user requirements elicitation in the raise platform
K. Tsimpita*, I. Makridou*, T. Dimos**, J. M. S. Grau**, A. Karakoltzidis***, A. Gotti***, S. Nifakos****, I. Cejudo*****, H. Gharrad, P. Bamidis* and E. Konstantinidis*
*Aristotle University of Thessaloniki, School of Health Sciences, Department of Medicine, Lab of Medical Physics, Thessaloniki, Greece
**Hellenic Institute of Transport (HIT), Centre for Research & Technology Hellas (CERTH)
***Environmental Engineering Laboratory, Department of Chemical Engineering, Aristotle University of Thessaloniki
****Department of Learning, Informatics, Management and Ethics, Karolinska Institutet
Introduction: The RAISE platform supports open, transparent, and reproducible biomedical research by allowing computational scripts to be executed on protected datasets, with only results returned and blockchain-registered.1 It aligns with FAIR (Findable, Accessible, Interoperable, Reusable) principles while maintaining data security. Agile and Living Lab methodologies2 were employed to adaptively develop a user-centered environment.
RAISE use case diagram.
Materials and methods: A mixed-methods Agile approach was applied:
1. Structured user engagement: Interviews, surveys, and usability testing identified key system needs.
2. UX and performance evaluation: Task-based assessments and participatory design ensured intuitive data navigation and reproducibility.
3. Incremental development: Bi-monthly iterations integrated validated user feedback. Researchers interacted with RAISE (Figure 1) by searching, viewing metadata, and processing datasets securely. Quantitative and qualitative analyses identified usability trends and optimization points. Continuous two-month cycles ensured progressive adaptation.
Results: Agile iterations led to refinements including improved search functionalities, interface enhancements, and real- time monitoring for computational traceability. Modular system design ensured scalability and FAIR compliance, facilitating integration with external infrastructures.
Discussion: Agile methods enabled rapid responsiveness to user feedback, addressing workflow fragmentation and integration issues. Co-creation with pilot users shaped RAISE’s evolution. While resource-intensive, this iterative process minimized usability gaps and strengthened platform effectiveness.
Conclusions: Combining Agile development, real-world testing, and structured UX evaluations, RAISE delivers a scalable, FAIR-compliant platform that supports open, reproducible biomedical research. Continuous engagement ensures adaptability to future needs.
References
1. Wilkinson, M. D., et al. (2016). The FAIR Guiding Principles for Scientific Data Management and Stewardship. Scientific Data, 3, 160018.
2. Dell’Era, C., & Landoni, P. (2019). Living Lab: A Methodology Between User-Centered Design and Participatory Design. Creativity and Innovation Management, 28(3), 380–395.
Keywords: FAIR principles, Biomedical data processing, Research accreditation, Data security, Open Science
Acknowledgement
This study was conducted as part of the RAISE Science Horizon Project, funded by the European Union, GA Number 10105847
The impact of computed tomography iterative reconstruction in radiation dose reduction
F. Spanou*, S. Skiadopoulos*, A. Karahaliou* and G. Panayiotakis*
*Department of Medical Physics, School of Medicine, University of Patras, Greece
Introduction: Iterative Reconstruction (IR) techniques in Computed Tomography (CT) imaging have demonstrated potential in obtaining high quality images with reduced radiation dose1-3 IR algorithms such as Sinogram Affirmed Iterative Reconstruction (SAFIRE) and Adaptive Statistical Iterative Reconstruction (ASIR) have been tested in clinical-based, as well as in phantom-based studies. The current phantom-based study investigates the impact of IR parameters in image quality and radiation dose in head CT.
Materials and methods: To simulate patient head, a Mini CT QC acrylic phantom (diameter: 15.25 cm) was utilized, that includes tissue equivalent inserts (bone-equivalent and soft-tissue materials) and a high resolution insert. Image acquisition was performed using a 128 slice CT scanner (Somatom go.Top, Siemens). Vendor default exposure settings for head CT considered tube voltage of 100 kVp and tube current of 212 mA. A range of tube current values was further exploited (180-61 mA) for radiation dose reduction (CTDIvol). Reconstruction parameters considered varying SAFIRE strength values (S1, S2, S3, S4, S5) and two reconstruction kernels (Hr40f and Hr60f). Image quality was assessed quantitatively [Noise, Contrast-to-Noise Ratio (CNR), Signal-to-Noise Ratio (SNR)] and qualitatively (spatial resolution).
Results and discussion: Increasing SAFIRE strength from S1 to S5, noise was reduced, while CNR (Figure 1) and SNR (Table 1) were increased. Images reconstructed by high SAFIRE strength (S4 and S5) were smoother, without degradation of spatial resolution. Sharper reconstruction kernel (Hr60f) provided improved spatial resolution as compared to smoother kernel (Hr40f). Increased value of radiation dose (CTDIvol) resulted in improved SNR and CNR values. Low dose images (current = 151 mA, CTDIvol = 32.36 mGy) reconstructed with SAFIRE strength S4 demonstrated similar image quality (SNR and CNR without compromising spatial resolution) as compared to the reference protocol (current = 212 mA, CTDIvol = 45.50 mGy).
4D diagram illustrating CNR and noise for bone-equivalent material with respect to SAFIRE Strength (S1-S5) and radiation dose (CTDIvol) at 100 kVp.
Signal-to-noise ratio (SNR) for the bone-equivalent material with respect to iterative reconstruction (IR) strength for two reconstruction kernels (exposure conditions: tube voltage 100 kVp and tube current 121 mA).
IR Strength
Kernel Hr40f
Kernel Hr60f
S1
88.67
23.02
S2
90.58
25.75
S3
92.59
29.05
S4
94.55
33.66
S5
97.02
39.66
Conclusions: SAFIRE iterative reconstruction allows for reduced radiation dose, maintaining CT image quality and can be used for pediatric imaging and low-dose protocols.
References
1. M.J. Willemink and P.B. Nöel (2019) The evolution of image reconstruction for CT – from filtered back projection to artificial intelligence, Eur Radiol, 29:2185-2195
2. J. Greffier, F. Macri, A. Larbi et.al. (2015) Dose reduction with iterative reconstruction: Optimization of CT protocols in clinical practice, Diagn Interv Imag, 96:477-486
3. W. Stiller (2018) Basics of iterative reconstruction methods in computed tomography: A vendor independent overview, Eur J Radiol, 109:147-154.
Introduction: FLASH radiotherapy (FLASH-RT) is a new paradigm of radiation therapy, featuring ultra-high Dose rate (UHDR) irradiation of tumours, with Dose rate of 40 Gy s-1 or higher. While maintaining its anti-tumour effect, FLASH-RT is characterised by a transient hypoxic state of the healthy tissue, which results in its extended sparing, when compared to conventional radiotherapy methods.1
Materials and methods: A comprehensive review of literature on FLASH-RT has been conducted, focusing on technological advancements in UHDR-radiation delivery in various modalities. Special attention has been paid to novel methods and proposed structures dedicated for FLASH.
Results: Numerous technological advancements have been made in the past decade to accommodate the production and transmission of various modalities of UHDR radiation. The implementation of existing electron LINACs has been shown to enable FLASH-RT for superficial tumours, while there have also been various studies regarding the employment of Bremsstrahlung-target- and synchrotron- produced X-Rays for FLASH.2,3
Furthermore, focused Very-High-Energy Electron beams have been shown to enable the production of a Spread-Out Electron Peak, resembling the Dose distribution of the Spread-Out Bragg Peak of hadrons4 Last, but not least, existing hadron radiotherapy systems have been shown to be, in principle, compatible with FLASH-RT, with cyclotrons and synchrotrons presenting both strengths and weaknesses on the production and delivery of UHDR beams.5
Discussion: Despite its promising potential, several technological challenges remain to be addressed before additional pre-clinical and clinical trials, and widespread adoption. Technological advancements should also facilitate biological experiments and aid in the understanding of the, currently unclear, conditions for the induction of the FLASH effect, which should, in turn, facilitate the devisement of treatment plans. Additionally, development of real-time dosimetry methods and of beam diagnostics equipment to ensure dose and dose rate optimisation is crucial for the accurately monitored delivery of UHDR radiation.
Conclusions: Advancements in accelerator technology, beam control and real time dosimetry are vital in the successful implementation of FLASH radiotherapy. Overcoming these challenges requires further research and is expected to significantly optimise cancer treatment methods.
References
1. C. Limoli and M.-C. Vozenin (2023) Reinventing Radiobiology in the Light of FLASH Radiotherapy, Annual Review of Cancer Biology, 7:1-21
2. M.-C. Vozenin, et al. (2024) FLASH: New intersection of physics, chemistry, biology and cancer medicine, Reviews of Modern Physics, 96:035002
3. P. G. Maxim, S. G. Tantawi and B. W. Loo Jr. (2019) PHASER: A platform for clinical translation of FLASH cancer radiotherapy, Radiotherapy and Oncology, 139:28-33
4. J. Fischer, et al. (2024) Very-high energy electrons as radiotherapy opportunity, EPJ Plus, 139:728
Rehabotics: An integrated rehabilitation system for assessing and treating upper limb spasticity after stroke
V. Potsika*, N. Tachos*, V. Tsakanikas*, G. Papagiannis**, A. Triantafyllou** and D. Fotiadis*,***
*Unit of Medical Technology and Intelligent Information Systems, Department of Materials Science and Engineering, University of Ioannina, GR 45110 Ioannina
**Biomechanics Laboratory, Physiotherapy Department, University of the Peloponnese, Sparta and Physioloft, Physiotherapy Center
***Biomedical Research Institute, Foundation for Research and Technology-Hellas
Introduction: Limb spasticity resulting from stroke, traumatic brain injury, multiple sclerosis, or various central nervous system disorders such as brain tumors, leads to loss of hand function, joint stiffness, and severe pain. According to,1 stroke incidents worldwide have been recorded at 12.2 million, with 6.55 million deaths from stroke, and 143 million disability-adjusted life years, while upper limb impairment is the most common disability affecting 77.4% of patients. In clinical practice, the primary rehabilitation method involves a stretching program and the patients must visit their physiotherapist daily. This poses significant challenges for both the patients and their families considering their reduced motor function. Also, traditional methods for assessing upper limb spasticity are susceptible to variations influenced by the clinician’s expertise.
Materials and methods: The automation of traditional evaluation methods presents an emerging area, with robot-aided systems being one of the most promising approaches. This paper introduces the REHABOTICS integrated rehabilitation system to deliver highly personalized assessment and treatment for upper limb function in patients with spasticity following a stroke. The REHABOTICS exoskeletal system consists of two functional elements: (i) the passive (soft) exoskeletal aid (PEA), and (ii) the active exoskeletal aid (AEA). PEA refers to the part of the system responsible for measuring and acquiring the data through a glove with specialized sensors. The data monitoring and storing takes place in real time, extracting useful metrics and conclusions. The system includes an interactive virtual environment using machine vision and augmented reality technologies able to acquire, interpret and evaluate the spatial information of the user’s hand from either the PEA or a computer camera or a combination of the above. The AEA is an exoskeleton for managing finger movements by utilizing the rotational motion of servo motors (Dynamixel AX-12A, Corona, CA, USA). An articulated structure has been developed with specially designed rings which guide a wire in order to transfer the motion of the servo motor to the finger components converting rotational motion into linear.
Results: The assessment tests which are performed using the PEA are based on: (i) the Ashworth Scale (AS), which is the most universally accepted clinical tool used to measure the increase of muscle tone, (ii) the Passive Range of Motion (PROM), (iii) the Active Range of Motion (AROM), and (iv) Free Session (Figure 1a). The AEA offers an interactive interface which allows users (patients and physiotherapists) to adjust parameters like finger extension speed, set the maximum extension angle, and control the speed of return to the starting position (Figure 1b). Using an intuitive navigation menu, the selected exercise is executed, prompting the servo motors to begin moving the fingers. The movement angles for each servo motor are displayed in degrees on the graphical interface.
The REHABOTICS: (a) PEA, (b) AEA.
Discussion: The REHABOTICS exoskeletal system based on the PEA and AEA components is a novel robot-aided system for the assessment and treatment of upper limb spasticity after stroke. Although further clinical research is required, the REHABOTICS solution is designed to: (i) provide an affordable and easy-to-use system for both patients and clinicians, (ii) deliver quantitative data to clinicians through specialized assessments (e.g. Box and Block test, AS scale, PROM, AROM), and (iii) motivate patients to enhance their motor skills.
Reference
1. V. Feigin et al., Global, regional, and national burden of stroke and its risk factors, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019, The Lancet Neurology 20 (10), 795 – 820, 2021.
This work was co-financed by the European Regional Development Fund of the European Union and Greek national funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH – CREATE - INNOVATE as part of the Rehabotics project (project code: Τ2ΕDΚ-04333).
CPR assist: A wearable and mobile app system for real-time CPR feedback and training
E. Rouka*, V. Fiska*, V. Mantiou*, K. Mitsopoulos*, P. Bamidis* and A. Athanasiou*
*Lab of Medical Physics and Digital Innovation (iMedPhysLab) School of Medicine, Faculty of Health Sciences Aristotle University of Thessaloniki (AUTH), Thessaloniki, Greece
Introduction: Out-of-hospital cardiac arrest (OHCA) has a survival rate below 10%, with high-quality chest compressions being critical for improving outcomes.1 However, even trained professionals struggle to maintain proper cardiopulmonary resuscitation (CPR) technique due to fatigue, incorrect hand placement, and lack of real-time feedback.2 Traditional CPR training relies on sensor- equipped mannequins, which, while effective, are bulky, expensive, and not practical for continuous practice.2 More recently, wearable CPR assistance devices have been developed to offer real-time feedback in a more portable form. Despite these advancements, many still suffer from rigidity, discomfort, and lack of emergency response integration, limiting their effectiveness in both training and real-world scenarios.3,4
CPR assist glove and app: The CPR Assist Glove is a wearable device designed to improve CPR technique by providing real-time feedback and emergency response support. It is equipped with sensors to measure compression depth, frequency, and other metrics, ensuring users follow proper guidelines. It connects wirelessly to a mobile application, which acts as the main interface for monitoring performance and tracking CPR quality. The app has two main modes: Emergency Mode – When activated, the system prompts the user with an option to call emergency services (112). The app includes an AED Locator which shows the nearest defibrillator and provides real-time navigation directions. Additionally, it provides live compression feedback by analyzing sensor data in real-time and issuing corrective prompts to guide effective CPR. Training Mode – This mode helps users improve their CPR skills by offering real-time feedback on compression depth, frequency, and effectiveness. After each session, the app stores performance data, allowing users to review previous sessions, track progress over time, and analyze detailed insights for skill improvement. This system assists in real-time emergencies while also acting as a CPR training tool. By integrating wearable technology and real-time analytics, it ensures precise CPR execution and supports continuous skill improvement in both clinical and public settings.
Conclusion: This work presents the CPR Assist Glove, a wearable solution aimed at improving CPR quality through real- time feedback, emergency response integration, and structured training. Future work will focus on enhancing sensor accuracy, improving glove comfort for extended use, and conducting real-world testing to evaluate its effectiveness in both training and emergency situations (Figure 1).
RAISE use case diagram.
References
1. Li S, et al. Survival after out-of-hospital cardiac arrest before and after legislation for bystander CPR. JAMA Netw Open. 2024 Apr 26;7(4):e247909.
2. Xie J, Wu Q. Design and evaluation of CPR emergency equipment for non-professionals. Sensors. 2023 Jun 27;23(13):5948.
3. Guridi S, et al. A proof-of-concept study on smart gloves for real-time chest compression performance monitoring. IEEE Access. 2024; 12: 22331–44
4. Zhang J, et al. Wearable inertial and pressure sensors-based chest compression quality assessment to improve accuracy and robustness. IEEE Sensors J. 2024 Feb 1;24(3):3774–87.
This research project was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) https://www.elidek.gr under the “2nd Call for H.F.R.I. Research Projects to support Faculty Members & Researchers” (Project Number: 4391).
Searching for biomarkers in cerebral palsy: An in-silico analysis
K. Baldoukas*,**,1, K. Birmpas*,***,1, K. Risvas* and K. Moustakas*
*Department of Electrical and Computer Engineering, University of Patras, Patras, Greece
**National Technical University of Athens, Athens, Greece
***Faculty of Biological Sciences, University of Leeds, Leeds, UK
Introduction: Cerebral palsy (CP) comprises a group of non-progressive neurodevelopmental disorders that impact posture, coordination, and motor control. While spasticity, hypertonia, and muscle weakness are hallmark symptoms, therapeutic decisions—particularly the initiation and timing of botulinum toxin (BoNT-A) injections—are largely empirical, lacking quantitative guidelines. This project investigates the use of digital twins and predictive modelling to derive kinematic biomarkers that can guide therapeutic intervention, offering a framework for objective and personalised clinical decision-making.
Materials and methods: To simulate CP-related motor dysfunction, a planar 9-degree-of-freedom musculoskeletal model was constructed using OpenSim and SCONE, incorporating 18 lower-limb muscles. Literature-derived alterations in reflex delay, constant baseline activation (C0), and feedback gain (K_F) were used to simulate both full-blown and intermediate stages of the disease. A model of BoNT-A-treated CP was created by applying a 62.5% attenuation to C0 and K_F and eliminating reflex delays. We developed five classes of digital patients: healthy control, untreated CP, post-BoNT treatment, and intermediate states representing 50–80% disease progression and 40–80% treatment efficacy decay. Forward dynamic simulations were run to assess balance under each condition, and joint kinematic profiles (ankle, knee, and hip angles) were extracted for analysis. Two primary phases of motion were studied: the dynamic phase leading to balance and the postural steady-state phase. Metrics included absolute and normalized joint angle differences, standard deviation, interquartile range, and—most importantly—the average joint angle after balance. These were evaluated as potential biomarkers for disease state classification. A linear discriminant analysis (LDA) classifier was trained to detect the threshold at which patients transition from “normal” to “diseased,” informing optimal treatment initiation and reinjection timing.
Results: Simulated balance tasks revealed unique joint angle trajectories for each clinical state. Notably, the average hip and ankle joint angles post-balance robustly discriminated between normal, CP, and post-treatment cases. The classifier identified ∼70% disease progression as the threshold for initiating BoNT-A therapy, and ∼60% efficacy loss as the point for reinjection. The knee angle was found to be supportive but insufficient as a standalone biomarker due to non-unique mappings across conditions. Characteristic regression curves were fitted for each joint across disease and treatment progression stages, yielding clear visual decision boundaries. While the ankle and hip angles were monotonic and suitable for direct clinical use, the knee angle showed ambiguity in intermediate states. When used in combination, however, these joint-based metrics offered high diagnostic and prognostic resolution.
Discussion and conclusions: This study demonstrates the feasibility of using digital patient simulations to derive non-invasive, easily measurable kinematic biomarkers for CP. These biomarkers—average joint angles after balance—can be obtained from basic video recordings and computed with minimal post-processing, aligning with standard clinical workflows. Their application spans diagnostic screening, treatment planning, and therapy monitoring. We propose a novel methodology for managing cerebral palsy using joint angle-based biomarkers derived from validated in silico models. Hip and ankle angles post-balance provide reliable diagnostic and prognostic information and can inform the initiation and repetition of BoNT-A therapy. These findings support a shift toward quantitative, simulation-driven, and patient-specific medicine in pediatric neurorehabilitation. Experimental validation and calibration with wearable motion capture systems are essential next steps. Scaling the simulation to 3D models, incorporating contractures and tendon transfer scenarios, and refining the classifier with larger datasets will further enhance its translational potential.
References
1. Granata KP et al., Electromechanical Delay in Spastic CP, 2000
2. Lidbeck C et al., Muscle Strength and Standing Ability, 2015
3. Kainz H et al., Muscle Force Scaling in CP Models, 2018
4. Multani I et al., BoNT in CP, 2019
5. Chen YT et al., BoNT and Motor Performance in Stroke, 2020
Introduction: Graph theory is examined as a powerful tool for understanding and analyzing Spinal Muscular Atrophy (SMA), a genetic disorder that affects spinal cord neurons and causes progressive muscular weakness and atrophy.1 By modeling neurological mechanisms and gene interactions as graphs, researchers can obtain a computational and visual framework of the disease,2 enhancing understanding and enabling therapeutic interventions.
Disease: Spinal Muscular Atrophy (SMA) is a neurodegenerative and neuromuscular disease, genetically determined, that impacts the central and peripheral nervous system, as well as voluntary muscles. It is classified as a motor neuron disease and is primarily caused by mutations in the SMN1 gene.1 The severity of SMA varies, with forms ranging from mild to life-threatening.
Graph theory and modeling: Graphs, consisting of nodes and edges, are widely used to represent complex biological systems.2 In SMA, genes are modeled as nodes and their interactions as edges, forming protein networks and gene expression graphs. These models can simulate motor neuron apoptosis and synaptic disruptions, contributing to the understanding of impaired neural function.
Expected outcomes: Recent advances in genetic screening and gene therapy have improved outcomes for SMA patients. However, instability of therapeutic proteins remains a challenge. Through sub-graphs and network motifs, gene interactions can be better visualized, allowing prediction of novel biological pathways and regulatory relationships.4
Limitations: Despite its promise, graph-based modeling has limitations. Biological systems are highly dynamic and context-dependent, which simple graph structures may not fully capture. Moreover, the accuracy of graph inference depends on data quality and completeness, and large-scale graphs may require significant computational resources (Figure 1).3
End-to-end framework gene regulatory graph neural network (GRGNN), proposed by Wang et al. (2020).4
Conclusions: Graph theory offers valuable insights into SMA at molecular and genetic levels. Further development of graph-based algorithms may support early diagnosis, individualized therapies, as well as prediction of SMA’s progression.
4. Wang J., Ma A., Ma Q., Xu D., Joshi T. Inductive inference of gene regulatory network using supervised and semi-supervised graph neural networks. Computational and Structural Biotechnology Journal, Elsevier 2020, https://doi.org/10.1016/j.csbj.2020.10.022
Light-based nanomedicine approaches in ocular cancer management: Future perspectives
M. Tsoplaktsoglou*, E. Spyratou** and E. P. Efstathopoulos**
*Medical School, National and Kapodistrian University of Athens, 11527 Athens, Greece
**Department of Applied Medical Physics, Medical School, Attikon University Hospital, National and Kapodistrian University of Athens, 11527 Athens, Greece
Introduction: Current treatments for ocular cancers often face limitations such as suboptimal delivery of pharmaceutical ingredients, limited specificity and severe side effects. Nanomedicine aims to address these challenges by enhancing conventional methods, including light-based therapies, namely photothermal therapy (PTT) and photodynamic therapy (PDT).
Methods: A comprehensive literature review was conducted to evaluate recent advancements in nanoparticle-based therapies for the most common ocular cancers, retinoblastoma (RB) and uveal melanoma (UM), with a focus in PTT/PDT. Relevant preclinical and clinical studies were systematically analyzed.
Results/Discussion: PTT uses light-sensitive heating agents which, after being delivered to the tumour site, are irradiated by a NIR laser and convert light energy into heat to induce cell death via hyperthermia. PDT on the other hand employs photosensitizers that, upon activation with specific light wavelengths produce reactive oxygen species (ROS) to induce tumor cell apoptosis or necrosis.
Nanoparticles can be employed directly as photothermal agents, as in the case of gold nanoparticles, which efficiently convert light energy into heat due to their optical properties, or iron-oxide nanoparticles, which can be used for magnetic hyperthermia. Similarly, TiO2 NPs have demonstrated effectiveness as photosensitizers in PDT. A different approach is to conjugate NPs with conventional photothermal agents (e.g., ICG) or photosensitisers (e.g., porphyrins), in order to create formulations with improved pharmacokinetics and bioavailability. Recent advances include thermoresponsive nanogels that allow controlled drug release upon irradiation. Surface functionalization with targeting ligands, such as EpCAM antibodies or FA, further enhances specificity and uptake by cancer cells. A major achievement of nanomedicine is the lipid nanoformulation of the photosensitiser verteporfin, Visudyne®, which is approved for PDT in excudative AMD. It is currently being investigated in clinical trials for both RB and UM, with particularly promising results for UM. Another noteworthy nanoformulation is AU-011, a virus-like nanoparticle conjugated with a fluorescent dye, that targets the heparan-sulfate proteoglycans overexpressed in cancer cells. When light activated, it induces immunogenic death, and it has advanced in phase 2 trials for UM.
A revolutionary aspect of nanotechnology in photo-based therapies is its potential for multimodal applications. For instance, AuNPs can act as enhancers for OCT, fluorescent or PA imaging, allowing in vivo tracking and image-guided PTT, while they have also been used as dose enhancers in radiotherapy. Another treatment often combined with PTT or PDT is chemotherapy, leveraging the increased susceptibility of treated cells to chemotherapeutic drugs. PTT and PDT can also benefit from each other. For example, PTT increases local blood flow and oxygenation, mitigating the oxygen dependency limitations of PDT. Overall, NPs enable the co-deliver of a variety of molecules (light-activated, imaging-enhancing, chemotherapeutic), allowing several combinations between chemotherapy, PDT, PTT and radiotherapy, along with imaging, to maximise their synergetic effects.
Conclusion: The integration of nanotechnology in photo-based therapies can offer treatment precision and optimised administration, with some nanoformulations already in clinical trials and several others in preclinical studies. Given the ability to combine them with other therapeutic methods, such as radiotherapy and chemotherapy, as well as multiple imaging modalities, nanomedicine holds the potential to transform the ocular oncology landscape.
References
1. Lara-Vega, I.; Vega-López, A. Combinational Photodynamic and Photothermal - Based Therapies for Melanoma in Mouse Models. Photodiagnosis Photodyn. Ther. 2023, 43, 103596.
Introduction: Intraoperative use of fluorophores aims in better visualization of the tumoral tissue.1 Its use in combination with microsurgical technique is well established in achieving greater resection with lower neurological complications. Endoscopes capable of inducing fluorescence have been gaining traction in the recent years hoping to maximize the extent of resection in the margins of a tumor where visualization only by microscope has technical limitations.
Materials and methods: A literature review in PubMed database of the last 5 years was conducted in March of 2025, using the search term “fluorescent use in endoscopic neurosurgery” retrieving 62 results. Only articles about nonvascular intraparenchymal tumors were included. Case reports were also excluded. Eight articles were reviewed comparing florescence-enhanced endoscopic and endoscope-assisted resection of intraparenchymal brain tumors against microscope only resection.
Results: The Fluorescent agents used by the researchers were 5-aminolevulinic acid (5ALA), Indocyanine Green (ICG) and Fluorescein (FNa). When compared with microscope, the visualization of the tumoral tissue by endoscope under fluorescent lighting was more intense and in some of the cases where the administration of the agent was characterized “unhelpful” under microscope view the endoscope achieved to show sufficient contrast.2 Furthermore, after mass excision under microscope when inspecting the cavity with an endoscope it was possible to identify residual malignant tissue due to better viewing angles and closer proximity to the surgical field. This advantage is reinforced by using endoscopes with 30° cameras.3 Better and more extensive visualization is also supported by a greater percentage of gross total resection achieved when the use of endoscope is implemented.4,5
Discussion: The treatment of brain tumors remains a challenge, with complete excision of lesions being a center point of the therapeutic course. The implementation of fluorescent agents during surgery has improved patients’ outcome and the use of endoscopes seems to further enhance the extent of resection. Specialized gearing is necessary both for microscopes and endoscopes capable of capturing the wave lengths emitted by fluorophores (ICG:820-860 nm, 5-ALA:640-710 nm, FNa:540-690 nm), with the latter not being widely available yet due to extra cost and additional training required.
Conclusion: Data so far is promising for using florescence-enhanced endoscopy treating intraparenchymal tumors. Additional high-quality studies are needed to clarify the superiority of combined endoscope/microscope technique against their use separately. More research is needed in the field of fluorophores, development of novel substances and exploring multiple agents use during surgery, analyzing technical challenges, interference and toxicity.
References
1. Xu Y, et al. (2025) Fluorescence Endoscopy with Second Window Indocyanine Green for Surgical Resection of Malignant Brain Tumors. World Neurosurg. 6;196:123766
2. Bettag C, et al. (2022) Endoscope-assisted visualization of 5-aminolevulinic acid fluorescence in surgery for brain metastases. J Neurosurg. 29;137(6):1650-1655.
3. Ma R, et al. (2024) Endoscopic 5-Aminolevulinic Acid-Induced Fluorescence-Guided Intraparenchymal Brain Tumor Resection-Can the Endoscope Detect More Fluorescence Than the Microscope? World Neurosurg. 185: e1268-e1279.
4. Ruzevick J, et al. (2022) From white to blue light: evolution of endoscope-assisted intracranial tumor neurosurgery and expansion to intraaxial tumors. J Neurosurg. 11;139(1):59-64.
5. Bettag C, et al. (2023) Endoscope-enhanced fluorescence-guided microsurgery increases survival in patients with glioblastoma. Acta Neurochir (Wien). 165(12):4221-4226.
Keywords: 5-ALA, Neuroendoscopy, Fluorescence-guided surgery, Endoscope, Fluorescein, Indocyanine Green
Knowledge-based treatment planning models in radiotherapy for prostate patients
K. Michailidis*,**, M. Chalkia***, K. Zourari***, N. Kollaros***, G. Patatoukas***, A. Kougioumtzopoulou****, V. Kouloulias****, D. Emfietzoglou** and P. Platoni***
*Department of Physics, University of Ioannina, Ioannina, Greece
**Medical Physics Laboratory, Department of Medicine, University of Ioannina, Ioannina, Greece
***Department of Applied Medical Physics, Attikon University Hospital, Medical School, National and Kapodistrian University of Athens, Athens, Greece
****Laboratory of Clinical Radiation Oncology, Attikon University Hospital, Medical School, National and Kapodistrian University of Athens, Athens, Greece
Background: Knowledge-Based Planning (KBP) in radiotherapy is used to improve treatment plan quality and to reduce both plan variability among users and the treatment planning time. A commercial KBP software which is used to create estimates of dose-volume histograms (DVHs) for new patient’s organs at risk (OARs). This study aims at the creation of three KBP models for prostate cancer patients by expanding the pre-existing models.
Materials and methods: Three RapidPlan™ (Varian Medical Systems, Palo Alto, CA) prostate models were created using 143 high quality, clinically acceptable VMAT plans. The models created were based on the dose prescribed to the Planning Target Volume (PTV) for the successive prostate irradiation phases; namely low-risk, intermediate and high-risk phase. During the verification of each model, the DVH’s (Figure 1-Left) and the regression and the residual plots of each structure were examined for potential outliers (Figure 1-Right, Figure 2-Left, Figure 2-Right). The statistical parameters of the new models were analyzed.
(Left) Dose volume histogram from a prostate treatment plan, including OARs and PTV. (The y-axis represents the ratio of total volume (%) and the x-axis represents the dose (Gy)). (Right) The influential point (blue crosshair) drives the regression plot of femoral head. The y-axis represents the DVH principal component score 1, and the x-axis represents the geometric distribution principal component 1.
(Left) Regression plot of the small bowel; the treatment plan in blue is a potential geometric outlier. Each crosshair in the graph corresponds to a different treatment plan included in training set. (Right) Rectum: Negative dosimetric outlier (clinical DVH is worse than the predicted DVH estimate from RapidPlan).
Results: Training and verification results showed improved results for the three new models. Specifically, the low-risk irradiation phase 2 out of 6 trained OARs achieved a coefficient of determination above 0.7, denoting a coherent model. Likewise, in the intermediate and high-risk irradiation phase, 3 out of 6 trained OARs achieved a coefficient of determination above 0.7.
Conclusion: The training and the verification of the three models showed favorable results and set the correct basis for the final step of models’ validation.
References
1. Anna-Maria Fanou, 2022, Study and introduction of knowledge-based planning (KBP) models for VMAT treatment plans into clinical use
2. W. Schlegel, P. Kneschaurek, 1999 May, Inverse radiotherapy planning, 175(5):197-207.
3. A. Fogliata, L. Cozzi et al., 30 October 2019, RapidPlan knowledge-based planning: iterative learning process and model ability to steer planning strategies, volume 14: article number 187.
Personalized real-time dosimetry for anaesthesiologists: Assessing radiation exposure across fluoroscopic guided surgical procedures
A. Papacharisi*, M.A. Kouri**,***, T. M. Axakali*, P. Georgakis****, V. Tanou****, E. Bournaki****, E. Kounadi***, C. Michail*, I. Valais* and G. Fountos*
*Department of Biomedical Engineering, Radiation Physics, Materials Technology and Biomedical Imaging Laboratory, AKTYBA, University of West Attica, Greece
**2nd Department of Radiology, Medical Physics Unit, Medical School, National and Kapodistrian University of Athens, Greece
***Medical Physics, General Hospital GHA Korgialeneio Mpenakeio-Hellenic Red Cross, Athens, Greece
****Department of Αnaesthesiology, General Hospital GHA Korgialeneio Mpenakeio-Hellenic Red Cross, Athens, Greece
Introduction: Ionizing radiation is a key component of many surgical procedures across specialties such as vascular surgery, orthopedics, and urology.1 Anaesthesiologists play a crucial role in these interventions, often requiring their continuous presence throughout the procedure.2 Consequently, their occupational exposure to radiation can be significant and varies with the type of procedure, highlighting the need for systematic monitoring. Each surgical procedure presents distinct exposure patterns, necessitating a tailored approach to radiation protection.3 This study aims to quantify the effective radiation doses received by anaesthesiologists during fluoroscopy-guided procedures, emphasizing the importance of procedure-specific exposure assessment to enhance personalized real-time dosimetry and improve radiation safety for healthcare staff.3,4
Materials and methods: ALMAR+ HERADO real-time, active personal dosimeters that recorded exposure every minute were employed for dose monitoring. A statistical analysis was performed to assess the impact of procedural variables, including surgical procedure, anaesthesiologist positioning, irradiation duration, and patient Dose Area Product (DAP). Additionally, the data were utilized to evaluate compliance with radiation protection guidelines and established exposure limits.
Results: The findings of this study reveal that radiation dose varies based on procedure type, irradiation time, and patient DAP. Neurointerventional procedures showed the highest effective dose, with prolonged irradiation and elevated DAP. ERCP procedures had the lowest exposure, featuring short irradiation and minimal DAP. Urological procedures showed high DAP despite shorter irradiation time, suggesting that the anaesthesiologist’s position in the operating room may significantly influence exposure, regardless of patient DAP.
Discussion: Radiation exposure among anaesthesiologists varies by procedure emphasizing the need for personalized dosimetric monitoring. Real-time dosimetry enables tailored protection strategies, optimizing shielding, positioning and exposure management to enhance occupational safety.
Conclusions: The comparison of surgical procedures reveals substantial variations in anaesthesiologists’ radiation exposure, largely dependent on irradiation time, and positioning. The findings highlight the necessity for procedure-specific monitoring and adaptive radiation protection strategies to ensure optimal occupational safety as well as the benefits of real time dosimeters use.
3. A.M. Sailer, et al. (2017) Real-Time Patient and Staff Radiation Dose Monitoring in IR Practice, Cardiovascular and Interventional Radiology, 40: 421-429. Available from: https://pubmed.ncbi.nlm.nih.gov/27942927/