Abstract
Background
At present, nonvirtual neurovascular training can be performed using either an angiographic suite under fluoroscopic guidance (entailing radiation exposure) or direct optical visualization with a camera-based system. The angiographic approach offers high-fidelity visualization and catheter control but is constrained by the limited availability of such specialized facilities, whereas the camera-based approach can be implemented virtually anywhere yet lacks comparable realism in key procedural aspects. The objective of this work is to develop and evaluate a novel camera-based angiography training system (CBATS) that generates artificial angiograms and roadmaps, thereby combining the advantages of both imaging techniques while eliminating radiation exposure.
Methods
Three distinct aneurysm models were integrated into a novel neurointerventional training platform, which was evaluated by neurointerventionalists across three training imaging modalities (conventional digital subtraction angiography [DSA], camera-only visualization, and CBATS) to simulate general endovascular procedures. Following the training sessions, a questionnaire-based evaluation was conducted to compare the performances of the camera-only and CBATS approaches with the conventional DSA, which served as the gold standard.
Results
Ninety percent of the raters agree that the visual realism of our CBATS training system is comparable to real angiography and offers a significant advantage over the camera-only variant. The absence of radiation, in particular, was rated as a major advantage by 89.3% of the raters.
Conclusions
In summary, our angiography-like training system is considered comparable to a real angiography system in terms of realism for the training scenarios addressed. However, it offers the advantage of being radiation-free and can be set up in almost any location.
Introduction
Endovascular therapy has become the standard of care for most cases of intracranial aneurysms (IAs). 1 This transformation of healthcare reality from microsurgical clipping toward endovascular therapy of IAs in recent years has been accompanied by the development of numerous new devices for IA therapy. 2 The establishment of a thrombectomy service has led to an increased demand for neurointerventionalists per department. Consequently, the number of aneurysm embolizations performed per trainee has decreased. Additionally, due to the rising number of endovascular treatments overall and the large variety of devices, there is an increasing need for training of future neurointerventionalists.3,4 Current training methods include synthetic models, anesthetized animals, human cadavers, and virtual reality (VR) simulation.5,6 However, each training method has its own strengths and weaknesses such as realistic appearance and behavior of material on the one hand or the need for using real digital subtraction angiography (DSA) machines with concomitant X-ray exposure or high costs for simulators on the other hand. Especially in the field of 3D printed models, relevant advances have been made in the last years, making these models more realistic. 7 The wider adoption of 3D printers and the capability of producing patient-specific models in a relevant number of centers has increased significantly over the last years. 8 However, the use of these models always goes along with X-ray exposure and the availability of a DSA-machine for training purposes. Our angiography training system, therefore, aims at producing DSA (especially roadmap) like images derived from standard optic camera images to mimic a real angiography without X-ray exposure. This study evaluates the feasibility of our novel training systems in a head-to-head comparison toward identical models being used in an angiography suite for neurointerventional training.
Methods
Hardware
The training scenarios to be covered include probing, embolization of aneurysms using intrasaccular devices or coils, and the implantation of stents and flow diverters. The central element of the camera-based angiography training system (CBATS) is the actual model of the cerebral vessels (see Figure 1). To ensure good monoplane visibility, cases with aneurysms and adjacent vessels lying in the same plane as far as possible were selected. The vessel structure was segmented on a threshold-base from a 3D DSA voxel volume and converted into a surface model in STL file format. The vessels were 3D printed as a negative in a block model with clear resin using a Form3 (Formlabs Inc., Somerville, USA) stereolithography printer, ensuring precise reproduction of the anatomical structures. 9 The block models were constructed as flat as possible with a suitable standard projection perspective directly from above. The model includes a single inflow, with the branching outgoing vessels converging within the block to form a single outlet. Tube connectors were printed on the exterior of the block to facilitate the inlet and outlet connections. To optimize visual clarity, the upper surface of the model was subjected to sanding and polishing. The block models were fabricated with consistent external dimensions, allowing them to be securely positioned on an acrylic glass plate to prevent displacement. Beneath this acrylic plate is an A4-format light panel (HSK LED Ltd, Shenzhen, China), which provides uniform and consistent illumination of the model. A document camera (Ziggi-HD Plus, IPEVO Inc., Sunnyvale, USA) mounted on a stand is positioned adjacent to the panel. The camera has a maximum resolution of 3264 × 2488 pixels at a frame rate of 15 frames per second and is connected to a computer via USB. The CBATS is operated using a PCsensor USB foot pedal (Shenzhen RDing Tech Co. Ltd, Shenzhen, China) to trigger the artificial X-ray, while an Elgato StreamDeck Classic (Corsair Gaming Inc., Fremont, USA) is used for controlling most other functions.
Complete setup of the camera-based angiographic training system.
A silicone tube was appended to the inlet port of the block model, facilitating access for endovascular devices via a hydrostatic valve. Two syringes were connected to this valve using a three-way stopcock, enabling the model to be perfused with either water or water dyed red using crepe paper as a surrogate contrast agent. An additional silicone tube was attached to the outlet port of the block model, terminating in a backflow valve located in a collection reservoir. All manually injected fluid exits via the backflow valve and is accumulated in the reservoir. To generate a roadmap, the surrogate contrast agent is injected under artificial X-ray in roadmap mode until it becomes visible in the image, similar to real angiography. Since there is no continuous flow through the model, the surrogate contrast agent is subsequently flushed out with water from the second syringe. The required hardware described here, including a CUDA (v12, NVIDIA Corporation, Santa Clara, USA) capable notebook or a PC with monitor, costs approximately $1200.
Software
We developed a software application in Python (v3.11), compatible with both Windows and Linux operating systems. OpenCV (v4.10) is utilized for capturing camera images, while Flask (v3.1) serves to deliver the processed images to an HTML output page, accessible via any modern web browser for display. The StreamDeck is controlled by Companion (v3.4.4, Bitfocus AS, Oslo, Norway), which communicates with the software through a REST API. Image processing is performed using the NumPy and SciPy derivatives of JAX (v0.4.38), enabling efficient handling of high-resolution camera images at full frame rate through CUDA, when available.
Artificial X-ray images are generated by converting all frames into grayscale images and subtracting the current frame from the first frame of the current X-ray sequence. This approach ensures that structures are only visible if they exhibit movement during imaging, replicating the process of real DSA. To generate a roadmap, the X-ray procedure is initiated, the surrogate contrast agent is injected, and the procedure is stopped once the contrast agent becomes clearly visible. The resulting subtraction image is then inverted and saved as a roadmap. For subsequent X-ray sequences, the image areas with changes in the subtraction are merged with the roadmap, and transition areas between the two are interpolated to enhance the visual quality (see Video 1). The software also features a zoom function and image translation in zoom mode, allowing users to examine specific regions in detail. In addition, calibratable length measurements can be carried out for intervention planning (see Figure 2) as well as the three-dimensional view of the anatomical target structures in a volume rendering (see Figure 3). An existing roadmap can be deactivated and reloaded as needed.
Aneurysm dimension measurement in camera-based angiography training system (CBATS).
Three-dimensional volume rendering of the anatomical target structures in camera-based angiography training system (CBATS).
Anatomies
The selected anatomies are shown in Figure 4.
The anatomies include (a) a carotid T-aneurysm and a media-bifurcation aneurysm, (b) an supraophthalmic ICA sidewall aneurysm and (c) a basilar tip aneurysm. The surrounding blocks of identical size, as well as the inlet and outlet connectors, are also visible.
Evaluation
For the evaluation, a basilar artery aneurysm block model is employed using AXS Catalyst 5 and Trevo Trak 21 catheters (Stryker, Kalamazoo, USA). Intrasaccular embolization is simulated using a 9-mm Contour device and Target 360 Standard 7 × 20 mm coils (Stryker, Kalamazoo, USA). An Embotrap III 5.5-mm stent-retriever was used to simulate a stent-assisted procedure (Johnson & Johnson MedTech, CA, USA). The initial training is conducted in an angiography suite (ARTIS icono, Siemens Healthineers AG, Forchheim, Germany), utilizing a pulsatile water circuit. Roadmaps are generated using conventional iodine-based contrast agents. A similar block model is subsequently used for the same training procedures in the CBATS, incorporating artificial roadmapping, and under camera-only visualization without any roadmap. The training is carried out by 28 interventional radiologists or neuroradiologists and evaluated using a questionnaire with a 5-point Likert scale as defined in Table 1. This questionnaire was developed by two clinical experts, pretested during a separate neurointerventional training course, and subsequently refined. The participants in the final training have been interventionalists for an average of 7.5 (±5.3) years, with 19 participants classified as experienced, having at least 5 years of experience, and the remaining nine participants classified as inexperienced.
Results from a user evaluation survey comparing the performance of camera-based angiography training system (CBATS) and camera-only system to real DSA model training.
Participants rated their level of agreement on a 5-point Likert scale, from 1 = “Strongly Disagree” to 5 = “Strongly Agree” (n = 28).
Statistics
The mean values and standard deviations are computed for the ratings of each question. Similar calculations are conducted for the differences in ratings between CBATS and camera-only training. Significance testing for these differences is performed using the Wilcoxon signed-rank test (p < 0.05). To assess significant differences between the two rater groups for each question, Mann–Whitney U tests are employed (p < 0.05). All statistical analyses are conducted using the pandas library (v2.2.3) and SciPy (v1.15.2).
Results
An exemplary representation of each training scenario across all three training modalities is shown in Figure 5. All 28 raters completed all three training modalities. A detailed evaluation of the survey results is provided in Table 1. Mann–Whitney U testing determined no significant difference between the two rater groups (experienced and inexperienced).
Exemplary representation of (a) a coil, (b) a stent-retriever and (c) a contour device in a basilar artery head aneurysm model, each in (left) camera-only, (center) camera-based angiography training system (CBATS) and (right) subtracted angiographic imaging.
Discussion
The evaluation demonstrates that radiation-free CBATS platform can closely emulate conventional neurovascular training on real DSA systems, while overcoming many limitations of other methods. In key domains of realism and performance, CBATS was rated on par with actual angiographic training by neurointerventionalists, and it clearly outperformed a basic optical (camera-only) setup. For example, overall image realism of CBATS was judged very high, with 85.7% of participants agreeing that the angiographic view was similar to real DSA, compared to only 50% for the camera-only view. These results indicate that by using high-resolution optical imaging with digital subtraction and roadmap overlay, CBATS can produce angiographic images that are nearly indistinguishable from real fluoroscopy. Notably, 89.3% of the respondents indicated that CBATS provides equal or better spatial resolution than true DSA. This is attributable to the 3264 × 2488-pixel camera sensor, which yields sharp vessel detail without the noise or motion blur of X-ray imaging. Only minor optical refraction artifacts were noted, thanks to the solid resin phantom design, in contrast to transparent hollow-vessel phantoms that required submersion in water to avoid refractive distortion. 10
The surrogate contrast agent (colored water) produced a vivid representation of injected contrast within the vessels, effectively mimicking DSA's contrast visualization. While the contrast dynamics are not perfectly identical to iodinated contrast under fluoroscopy, users found the bolus appearance and dispersion realistic for training purposes. Indeed, the visual realism of CBATS approaches that of real angiography in image quality, contrast visibility, and anatomical detail, surpassing conventional camera-only models and rivaling high-end VR simulators.
The elimination of X-rays was universally lauded—89.3% of participants strongly agreed that avoiding radiation is a significant benefit. This allows trainees to practice angiographic skills without exposure to ionizing radiation, an especially important factor for neurointerventionalists who face cumulative exposure in clinical cases. By contrast, training on real angiography units (whether with phantoms, animals, or cadavers) unavoidably involves radiation to staff and must adhere to time/dose constraints. Moreover, because CBATS does not require a dedicated angio suite, it can be used in ordinary training labs, conference rooms, or simulation centers. The entire apparatus—a camera, light panel, and phantom block—is lightweight and portable, in stark distinction to a fixed angiography system. This portability was highlighted during the study, as the CBATS could be assembled in any location with minimal infrastructure, aligning with our goal of an “angiography-like” trainer usable in almost any setting.
Virtual reality simulators are likewise radiation-free and come with a huge amount of different training scenarios. They also offer mobility (some modern simulators are laptop-based), but they still require specialized hardware and technical support. In terms of cost, CBATS represents a highly cost-effective solution relative to other training modalities. The core components (high-definition document camera, 3D-printed models, basic computing, and controls hardware), which cost approximately $1200, are orders of magnitude cheaper than a clinical angiography machine or a high-fidelity VR simulator. Commercial endovascular VR simulators typically cost about $100,000–$200,000 for the unit, with additional annual service fees. 5
Cadaveric and live animal models remain the gold standard for haptics, as they involve real vessel tissue and blood flow. 11 Cadaveric vessels and live models offer authentic responses (e.g., vessel wall apposition, risk of perforation), but these models are inherently single-use and may not reproduce specific pathologies consistently. In addition, cadaver labs and animal training are expensive; a single cadaver may cost $1000–$3000 (plus facility and handling fees and veterinary support). 5 In our setup, multiple runs are feasible on the same printed model (until a device permanently occludes a target), and new identical models can be produced on demand for repeated practice. Virtual reality systems allow unlimited repeats as well, but again lack actual device behavior—for instance, a simulator cannot yet perfectly emulate the friction of deploying a complex coil or the feel of opening a flow diverter. This makes CBATS a valuable intermediate, giving trainees experience with the real tools and deployment techniques in a controlled setting.
Regarding their preferences, 46.4% of the participants conveyed a preference for CBATS over DSA training, with 28.6% strongly agreeing. Only a minority of 25.0% participants disagreed without any strong disagreement at all. Contrastingly, the camera-only training received more mixed responses. While 32.1% preferred this method over angiography, only half as much participants (14.3%) strongly agreed. In contrast, twice as much (32.1%) expressed their disagreement, including one person who even strongly disagreed. The p-value (p = 0.047) also indicates a significant advantage of CBATS over camera-only when comparing them to real DSA model training. While no further improvements are to be expected in camera-only training, the advantages of CBATS could be further enhanced by future developments.
Limitations
Despite its advantages, CBATS does have technical limitations that warrant discussion. As mentioned, the visibility of device markers through the delivery catheter is suboptimal due to the optical imaging medium. While the tip of aspiration catheters often offers a good transparency, this is not the case for tightly braided microcatheters and was regularly mentioned as the biggest disadvantage by the participants. Nevertheless, this is not a CBATS-specific problem and affects all purely visual training methods. Trainees must adapt by slower movements and tactile feedback to position devices, which could actually be seen as a useful skill exercise, though it is different from standard fluoroscopic visualization. Especially microcatheters with a transparent sheath and less braiding could help here.
Previous studies, such as, 12 have indicated that flexible models offer superior transparency compared to rigid models. However, we addressed this limitation by utilizing block models and refining them through polishing. Despite this, our models still exhibit lower compliance than human arteries due to the absence of flexibility. Participants also reported a diminished tactile sensation when deploying intrasaccular devices in the rigid models. Since this issue is not exclusive to CBATS compared to DSA training, it does not affect the current numerical findings. For instance, previous research like Oishi et al. 13 has examined the influence of model material on the forces generated and catheter behavior during coiling procedures, although no similar studies have been conducted for intrasaccular web devices. With regard to model visibility and device performance, we hypothesize that cast silicone blocks could potentially offer superior tactile feedback while improving transparency. However, the production of such silicone blocks is more complex and costly, preventing their in-house fabrication at this stage.
Another limitation is the current lack of continuous blood flow and pulsatility in the model. We found that using a recirculating pump was impractical because repeated contrast injections progressively dye the entire fluid volume, eliminating the visual contrast between the injected bolus and background fluid. Thus, our training runs were done with static flow conditions, and the contrast agent was washed out by subsequent water injections. This means hemodynamic effects (like pulsatile movement of catheters or contrast timing in relation to cardiac cycle) are not fully replicated. Observing the inflow behavior into an aneurysm after implantation of an intrasaccular device has also been difficult for this reason. In future designs, a continuous-flow system with active flushing or dye filtration could be explored to enable longer runs and simulate heart-pulsed flow.
Currently, CBATS only offers monoplane imaging. In contrast, clinical neuroangiography often uses biplane projections for simultaneous orthogonal views. While monoplane imaging is adequate for the current training scenarios—largely due to the simplicity of the selected anatomical structures—more advanced scenarios, such as stent-assisted coiling in anatomically complex, nonplanar configurations, would benefit from biplane imaging. However, the integration of a second camera has already been implemented on the technical level. The vessel models would have to be adapted accordingly in order to obtain approximately orthogonal views and to be able to orient the block surfaces in such a way that both cameras can be aligned perpendicularly to one of the surfaces.
Finally, while the visual quality of CBATS angiograms is high, the grayscale contrast intensity is based on visible color density rather than X-ray attenuation. This has minimal impact on the trainee's experience, but certain subtleties (for instance, distinguishing overlapping vessels by depth or seeing device shadowing) may differ slightly from true DSA. Users did not report this as a significant issue, as the overall image impression was maintained, but it remains a point for continuous refinement.
Conclusions
Camera-based angiography training system offers a realistic and cost-effective alternative for neurovascular training that combines realistic catheter handling with a near-real angiographic view. It provides an authentic experience that outperforms traditional radiation-free methods of physical model training in terms of image quality and utility, while being much less expensive than high-fidelity VR simulators. With further improvements, CBATS could become an integral part of neurovascular training, providing safe, accessible, and cost-effective skills development.
The planned next development steps include the integration of a pump that generates a continuous flow both manually and automatically at the moment of roadmap acquisition, thereby enabling automatic flushing of the artificial contrast agent as well as the assessment of potential stasis. In addition, the integration of a second camera for biplane imaging is planned, along with automatic camera calibration using defined, fixed markers within the field of view, allowing for reliable size measurements even when the cameras are not positioned perpendicularly.
Supplemental Material
Footnotes
Ethical considerations
Ethics approval and consent to participate: Ethical approval was obtained as required according to the guidelines of the local ethics committees (Ethics Committee, Medical Faculty of Otto-von-Guericke-University Magdeburg, Magdeburg, Germany). Regarding the aneurysm image data, this anonymous retrospective study, which was conducted in accordance with the Declaration of Helsinki, fulfills the guidelines of the federal state of Saxony-Anhalt. The data have been legally collected in accordance with §15 paragraph 5 (University Hospital Magdeburg).
Author contributions
S.K. and R.S. are equal contributors to this work and are designated as cofirst authors. Conception/design of work: S.K. and R.S.; Software development: S.K.; Hardware development: S.F.; Data collection: F.S. and S.K.; Data analysis and interpretation: S.K. and R.S.; Drafting the article: S.K.; Critical revision of the article: D.B.; All authors approved the final version of the manuscript.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: R.S. has received honoraria as a speaker and consulting fees for Infinity Neuro, Acandis, Stryker and Vesalio. D.B. has received honoraria as a speaker and consulting fees of Stryker, Acandis, Balt, Phenox and Vesalio. No disclosures or competing interests were declared by the remaining authors.
Supplemental material
Supplemental Material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
