Sage Journals: Discover world-class research

Abstract

Introduction:

Early detection of diabetic foot complications is essential for effective management and prevention of complications. Artificial intelligence (AI) technology based on digital image analysis offers a promising noninvasive method for diabetic foot screening. This systematic review aims to identify a study on the development of an AI model for diabetic foot screening using digital image analysis.

Methodology:

The review scrutinized articles published between 2018 and 2023, sourced from PubMed, ProQuest, and ScienceDirect. The keyword-based search resulted in 2214 relevant articles and nine articles that met the inclusion criteria. The article quality assessment was done through Quality Assessment of Diagnostic Accuracy Studies (QUADAS). Data were extracted and analyzed using NVivo.

Results:

Thermal imagery or foot thermogram was the main data source, with plantar temperature distribution patterns as an important indicator. Deep learning methods, specifically artificial neural networks (ANNs) and convolutional neural networks (CNNs), are the most commonly used methods. The highest performance is demonstrated by the ANN model with MATLAB’s Image Processing Toolbox that is able to classify each type of macula with 97.5% accuracy. The findings show the great potential of AI in improving the accuracy and efficiency of diabetic foot screening.

Conclusion:

This research provides important insights into the development of AI in digital image–based diabetic foot screening. Future studies need to focus on evaluating clinical applicability, including ethical aspects and patient data security, as well as developing more comprehensive data sets.

Keywords

artificial intelligence diabetic foot deep learning machine learning digital image analysis

Backgrounds

The management and prevention of diabetic foot complications rely heavily on their early detection.^1,2 Early detection helps in the proper triage of patients or healthcare workers toward more measures that are effective yet conservative for the management of diabetic foot complications.^3,4 Moreover, as the health workers are aware of the risks that the patients possess, they can further give health education on feet care management to the patients, which would eventually help to prevent diabetic foot ulcers (DFUs).^5,6

One of the nursing interventions that can be explored in the assessment of the diabetic foot is noninvasive digital image processing technology and artificial intelligence (AI).^7,8 These algorithms learn to recognize complex hierarchy data at previously impossible speeds and from even the concepts devoid of any structuring.^9,10 Enhancements in image processing and analysis reduce the need for manual scrutiny that may overlook small differences and variation.^11,12 These adequacies allow AI to improve accuracy in convoluted cases and overcome human errors caused by exhausted and limited supervision.^13,14

Artificial intelligence technology based on digital image processing has been widely used for diabetic foot screening. In particular, a developing model for early detection that employs digital image decomposition uses thermograph images in order to perform the search for risk zones on the soles of people with diabetes mellitus (DM).^8,14,15 This was done by investigating the pattern of plantar temperature mapping that may correlate with the presence of ischemia in specific soft tissues.¹⁴ Also, the development of early detector using AI also made use of the changes associated with the skin in the patients with diabetes.^16,17

Although much research has been conducted on the use of AI and digital image processing in identifying complications of diabetic foot, few studies have been carried out in terms of systematic reviews on this topic. The research related to the systematic review does not discuss the use of digital image processing specifically.¹⁶ In addition, there is still a gap in understanding the features or characteristics of digital image data sets that contribute to the detection of diabetic feet by AI systems as well as the type of AI and its accuracy.

The research questions that this systematic review seeks to answer include the following: (1) What data sources and digital imagery features are most commonly used for diabetic foot detection by AI systems based on the synthesis of evidence from various studies? (2) What are the most commonly used AI subtypes that have the best performance for diabetic foot screening through digital image analysis?

The purpose of this research is to identify an AI model development study for diabetic foot screening based on digital image analysis. This research has implications for improving the accuracy and efficiency of diabetic foot screening using AI and digital imaging. The findings can drive the development and adoption of AI systems for early detection, lead to better complication prevention, as well as support their application in clinical settings by considering methodological and practical aspects.

Method

Study Design

Researchers use systematic reviews in finding and synthesizing studies that deal with the development of AI models for diabetic foot screening based on digital image analysis. This systematic review is registered with INPLASY (No: INPLASY202510036).

Eligibility Criteria

Studies were included if they met the following criteria: (1) original articles published between 2018 and 2023; (2) research focus was specifically on primary research articles that developed or evaluated AI-based methods for diabetic foot screening, utilized digital image processing techniques, and included quantitative performance metrics of the AI system; and (3) were written in English. Studies were excluded if they were (1) including review articles (such as systematic reviews, meta-analyses, and narrative reviews), conference abstracts or proceedings, study protocols, case reports or case series, and letters to editor, commentaries, or opinion pieces; (2) sourced from books, encyclopedias, videos, or conference proceedings; (3) not accessible as full-text; (4) focused solely on general diabetic foot care without AI components, used AI for general diabetes management without specific foot screening components, or employed only traditional image processing without AI implementation; and (5) not have clear description of the AI methodology used, failed to report performance metrics, or did not specify sample size or data collection methods.

Information Sources

A comprehensive literature search was conducted using three electronic databases: PubMed, ProQuest, and ScienceDirect.

Search Strategy

The combination of keywords used in this research is “artificial intelligence AND image analysis AND diabetic foot prevention OR diabetic foot prediction OR diabetic foot screening”.

Selection Process

Study selection was performed in two stages. The first involved independent screening of titles and abstracts of all unique records identified by the search against the eligibility criteria by both reviewers (N.K.I.S.A. and H.K.). Excluded were those studies that clearly did not meet the criteria. The full-text retrieval and independent assessment of the remaining potentially relevant studies were done in the second stage by the two reviewers. Reasoning for exclusion of studies during the full-text phase was documented. Disagreements were resolved through discussion with a third reviewer (H.K.). The selection process was documented using the PRISMA 2020 flow diagram (Figure 1).

Figure 1.

Journal search and selection process flow chart using PRISMA 2020.

Data Extraction

Data from the included studies were extracted and summarized independently by the first author (N.K.I.S.A.) using NVivo. The coding scheme in NVivo was organized hierarchically, with parent nodes corresponding to major extraction categories and child nodes for specific data elements.¹⁸ In identifying and analyzing the content, the researcher performed systematic coding of information relevant to the research questions. The coding was organized into three main categories: introduction (authors, publication year, and research objectives), data sources and digital imaging features (the types of imaging used, number of respondents or samples, and data sources), and AI methodology (type of AI implemented, system architecture used, and system results or performance). All coding results were then tabulated and organized in spreadsheet format using Microsoft Excel to facilitate comprehensive analysis of the reviewed studies.

Article Quality Assessment

Reviewers (N.K.I.S.A. and H.K.) assessed each risk of bias for all studies included using the QUADAS-2 checklist as a quality evaluation method for diagnostic accuracy. Any disparities in scores were addressed through conversation. The risk of bias and applicability were evaluated across four domains: (1) patient selection, (2) index test, (3) reference standard, and (4) flow and timing. Each domain received a rating of low, uncertain, or high risk. Unclear risk was evaluated using inadequate study data to reach a judgment. Not applicable (N/A) was used when the QUADAS domain was not applicable owing to the study methodology.

Results

Study Selection

We found 2214 papers across four databases and originally eliminated 339 duplicates. The flowchart illustrating the selection process is presented in Figure 1. In summary, we removed 1856 references after screening titles and abstracts, leaving 19 for full-text examination. Finally, nine research studies were eligible in terms of the inclusion criteria.

Risk of Bias in Studies

According to Table 1, the QUADAS-2 study provides the following evaluations of four domains. Most of those studies present low risk of bias; however, there are certain differences. For patient selection, only one study¹⁹ was classified as high risk, while the other one was for reference standard with high risk.¹⁷ Applicability was also mentioned, including three of them^20
-22 in the patient selection high risk category, besides two studies^21,22 showing manifestations of unclear risk in index test. Thus, all the relevant comprehensive evaluation will provide an indication of the robustness and relevance of the studies included in the systematic review.

Table 1.

The Results of the Article Quality Assessment Using the QUADAS-2 Checklist.

Author (year)	Risk of bias				Applicability
Author (year)	Patient selection	Index test	Reference standard	Flow and timing	Patient selection	Index test	Reference standard
Z. Cao et al (2023)	Low	Low	N/A	Low	Low	Low	Low
Filipe et al (2022)	Low	Low	Low	Low	Low	Low	Low
Khandakar et al (2022)	Low	Low	Low	Low	Low	Low	Low
Dremin et al (2021)	Low	Low	High	Low	Low	Low	High
Toledo Peral et al (2018)	Low	Low	N/A	Unclear	Low	Low	N/A
Zhang et al (2022)	Low	Low	Low	Low	High	Low	Low
Arteaga-Marrero et al (2023)	High	Low	N/A	Low	Low	Low	N/A
L. Cao et al (2020)	Low	Low	N/A	Low	High	Unclear	Low
Mennes et al (2020)	Low	Low	Unclear	Low	High	Unclear	Low

Data Sources and Digital Imagery Feature

Based on the results of data extraction in Table 2, the data set used in these researches mainly (in four out of nine studies) consists of thermal images or foot thermometer images of people with diabetes and healthy subjects as controls.^8,19,23,24 Among nonthermal imaging studies, other study used hyperspectral imaging for skin complication detection,¹⁷ standard digital photography for macular analysis,²⁵ computed tomography (CT) angiography image,²⁰ plantar pressure image²¹ and laser speckle contrast imaging.²² Four studies (44.4%) used public data sets such as the WoundCareLog database,¹⁹ INAOE, and STANDUP,^8,19,24 while five other studies (55.5%) collected data specific to their research.^17,21
-23,25 The sample size varied from 19 patients with type II DM²⁵ to 416 patients that consist with people with diabetes and healthy volunteers.¹⁹

Table 2.

Data Extraction Results.

Author (year)	Aims	Data set				Type/AI method	Findings
Author (year)	Aims	Type	Source	Number of patients	Number of images	Type/AI method	Findings
Cao et al²³	To test the performance of the Plantar Foot Segmentation Network (PFSNet) to better extract foot contours in thermal imagery from active thermography compared with other deep learning approaches	Thermogram	Own	10 patients with diabetes and 10 healthy volunteers	1800	Deep Learning (PFSNet)Compared with: Unet, Unet++, U2Net, AttentionUNet, dan TransUNet	PFSNet achieved 97.3% (Intersection over Union) and 95.4% (Dice Similarity Coefficient) accuracy on a specially collected cold stress active thermography data set, superior to the other five deep learning models
Filipe et al²⁴	Evaluate the machine learning classification algorithm for two diabetes thermogram classification models	Thermogram	Public database (public database from the General Hospital of the North, the General Hospital of the South, the BIOCARE clinic, and the National Institute of Astrophysics, Optics and Electronics [INAOE])	122 patients with diabetes, 45 healthy volunteers	167 pairs	ML (Logistic Regression [LR], Support Vector Machine [SVM] [with two variants], K-Nearest Neighbor [k-NN] [with two variants])	Model 2 (binary classification followed by multiclass classification) outperforms model 1 (direct multiclass classification), with an average F-score of almost 10% higher and Accuracy 5% higher
Khandakar et al⁸	Classify diabetic foot thermometers using image-enhancing techniques and Convolutional Neural Network (CNN)	Thermogram	Public database (public database from the General Hospital of the North, the General Hospital of the South, the BIOCARE clinic, and the National Institute of Astrophysics, Optics and Electronics [INAOE])	122 patients with diabetes and 45 healthy volunteers	167 pairs	CNN (ResNet18, ResNet50, ResNet100, DenseNet201, InceptionV3, VGG19, and MobileNetV2)	Machine learning classifiers have comparable performance to CNN’s two-dimensional performance using image enhancement. CNN’s VGG 19 model showed accuracy, precision, sensitivity, F1-score, and specificity of 95.08%, 95.08%, 95.09%, 95.08%, and 97.2% in severity stratification, respectively
Dremin et al¹⁷	Developing a method for early detection of skin complications in DM patients through Hyperspectral Imaging (HSI) and machine learning (ML)	Hyperspectral Image	Own	Preliminary study (patients with diabetes)Clinical trial (20 patients with diabetes type 2 and 20 healthy volunteers)	Not mentioned	ANN	The proposed technology achieves sensitivity and specificity of 95 and 85%, respectively, in classifying DM and control groups (healthy volunteers)
Toledo Peral et al²⁵	Building an application for skin macular characterization based on a three-stage segmentation and characterization algorithm used to classify vascular macula, petekie, tropic changes, and trauma from digital photographs of the lower limbs presented	Photo of macular skin of the legs	Own	19 patients with diabetes type 2 (without DFU)	82 photos	ANN with MATLAB Image Processing Toolbox	The application can classify each type of macula with 97.5% accuracy
Zhang et al²⁰	Developing an artificial neural network (ANN) model to predict DFU based on clinical data and lower extremity CTAs	CT angiography image	Public database (the WoundCareLog database)	203 patients with DFU	Not mentioned	The artificial neural network model was created using SPSS 26.0 statistical software with a multilayer perceptron (MLP) algorithm	The proposed model performs very well on DFU predictions with an accuracy of 91.6%. Evaluated with holdout samples, model accuracy, sensitivity, specificity, PPV, and NPV were 88.9%, 90.0%, 88.5%, 75.0%, and 95.8%, respectively
Arteaga-Marrero et al¹⁹	Compare the relevant and robust features of the INAOE data set with those extracted from the expanded data set	Thermograms	Public database (STANDUP, INAOE, and Local Dataset)	STANDUP (145 patients with diabetes and 82 healthy); INAOE (122 patients with diabetes and 45 patients without diabetes);Local dataset (22 healthy volunteers)	Not mentioned	(ML) Lasso, Random Forest, Dropout	The lasso approach provides the best performance metrics for optimized SVMs, with an F1-score of 0.7956 ± 0.0291 for the all data set
Cao et al²¹	Evaluate the proposed fusion system to collect plantar pressure imagery data from patients with diabetes type 2	Plantar pressure images	Own	10 patients with diabetes type 2	Not mentioned	Several types of wavelet base functions are used for comparative analysis	This study determined that the transformation of the Haar wavelet with five levels of decomposition in the plantar pressure image achieved superior performance
Mennes et al²²	Developing an algorithm to assess Region of Interest (ROI) in Laser Speckle Contrast Imaging (LSCI) recordings in patients with diabetic foot complications	LSCI recordings	Own	33 patients with DFU	66 recordings	The algorithm is developed with custom code written in MATLAB using the Iterative Closest Point (ICP) algorithm	Intraclass correlation coefficients (ICCs) between the two raters and the algorithm were good to excellent (ICC = 0.790-0.978) and higher than the ICC between the two human raters (ICC = 0.628-0.988)

AI Subtypes

Based on Table 2, it can be observed that Deep Learning methods constituted the overwhelming majority of analytical type having been used in six out of nine studies, representing 66.7%. In this case, three of the included studies applied CNNs, while another three studies applied ANN approaches.^17,20,25 Traditional machine learning methods have been utilized in three out of the nine studies, which comprises 33.3%. These studies would be Filipe et al,²⁴ who applied the Logistic Regression, Support Vector Machine, and K-Nearest Neighbor techniques; Arteaga-Marrero et al,¹⁹ who used Lasso, Random Forest, and Dropout methods; and Mennes et al,²² who employed the Iterative Closest Point algorithm, will also be included in this traditional machine learning methods. Some studies like L. Cao et al²¹ focused specifically on analytical techniques such as wavelet base functions as opposed to conventional machine learning or deep learning approaches.

Discussion

Thermal imaging or foot thermometers tend to be the most relevant data in a number of studies.^8,19,23,24 Thermographic imaging allows noninvasive, real-time visualization of patterns of plantar temperatures directly corresponding to underlying physiological changes in diabetic foot.^8,23,24 Beside of that, through thermography feature, temperature asymmetries distribution detect the earliest signs of sites of inflammation and potential ulceration before they manifest any visible clinical signs, making it a very valuable tool for preventive health care.^14,15,23 Thermal imaging systems also have become widely available and inexpensive compared with other imaging modalities such as CT and magnetic resonance imaging while still showing a great deal of accuracy in diagnosis.^19,23 Furthermore, thermographic data can be easily standardized and digitized for automated analysis making this well-suited for AI-based interpretation systems.^8,24

The review has produced a significant finding indicating progressive advances in AI techniques to diabetic foot screening, in which deep learning approaches, mainly CNNs and ANNs, emerged as the leading techniques. The superiority of deep learning methods is based on their automatic capability to learn the hierarchical features from raw image data in an end-to-end fashion without requiring manual feature engineering.^8,23 This becomes particularly obvious in the performance metrics—CNN-based models have attained accuracy rates up to 95.08% in severity stratification,⁸ while ANN implementations reached 97.5% accuracy in macular classification.²⁵ All these developments attributed to deep learning will come from their capability to deal with large dimensional image data and complex data structures, and recognize imperceptive features by humans regardless. Furthermore, the flexibility of these models in all kinds of imaging modalities, from thermal images to hyperspectral data, shows the robustness of the clinical applications.^17,20,25 In general, the trend is toward using advanced deep learning techniques for diabetic foot image analysis.

This systematic review reveals innovative contributions to the implementation of AI in diabetic foot literature. While the previous systematic reviews^13,16 look further toward untangling diabetic foot technologies, the present one discovers how such methodologies using AI as virtual image analysis approach lend themselves to diabetic foot screening. Important technological advancement in favor of the sophistication of deep learning architectures and implementation is revealed in the findings. Among these advances are improved model architectures, improved feature extraction capacities, and higher classification performance over former approaches. This systematic analysis attests to a clear path in evolving AI applications—from direct through basic machine learning language to deep learning frameworks that are quite complex and need to be optimized for medical image analysis.

For nurses, this review will provide a foundation as they seek to integrate this technology in their primary health care, especially for patients with diabetes. Such a system will enable nurses to help in the preparation of diabetic foot complication—related images at an earlier stage and within a shorter time; hence, the appropriate care can be given beforehand so as to avoid the escalation of more serious concerns.^23,24 This technology can help nurses be more effective in their workload as they can care for more patients and concentrate on care that is needed only to cases in special attention.^26
-28 Such technology can also enhance patients’ self-care of diabetes, integrating better insights and understanding of their feet’s status.²⁹

In this context, methodological limitations should be borne in mind when the results are interpreted. The relatively small size of the studies included (n = 9) reduces generalizability. Such differences in methods, evaluation metrics, and even characteristics of the data set included in the systematic review make it very difficult to carry out direct comparisons of AI methodologies.

Future research should prioritize developing smartphone-based diabetic foot screening methods using standard digital photograph, as this approach is more accessible and convenient for regular home monitoring compared with specialized imaging modalities.^8,19,21,24 In addition, future studies should focus on creating comprehensive, multi-institutional data sets encompassing diverse patient groups and evaluating the clinical implementation of AI-enabled screening tools in terms of workflow integration, cost-benefit analysis, and real-world performance metrics. Addressing ethical considerations, such as data privacy and algorithm transparency, is also crucial for the successful integration of AI technologies into diabetic foot care.

Conclusion

Based on the analysis, it can be concluded that AI based on digital image analysis has great potential in improving the accuracy and efficiency of diabetic foot screening. Thermal imagery was the most commonly used data sources, with plantar temperature distribution patterns as an important indicator. Deep learning methods (ANN and CNN) were the most widely used methods. The highest performance is achieved using an ANN model implemented in MATLAB’s Image Processing Toolbox, which can classify each type of macula with an accuracy of 97.5% success rate. Future research needs to focus on evaluating clinical applicability and developing more comprehensive data sets.

Footnotes

Acknowledgements

The authors gratefully acknowledge the AI Center, Brawijaya University for providing research facilities and funding support that enabled the successful completion of this study.

Abbreviations

AI, artificial intelligence; ANNs, artificial neural networks; CNNs, convolutional neural networks; DM, diabetes mellitus; DFU, diabetic foot ulcer; PFSNet, plantar foot segmentation network; QUADAS, Quality Assessment of Diagnostic Accuracy Studies.

Author Contributions

N.K.I.S.A., H.K., D.H., Y.Y., P.L.T.I., R.R., and R.E.K. contributed to concept and design; N.K.I.S.A. and A.D.P. contributed to drafting of the manuscript; H.K., D.H., and R.E.K. contributed to supervision; N.K.I.S.A. and A.D.P. contributed to acquisition, analysis, or interpretation of data; H.K., D.H., Y.Y., P.L.T.I., R.R., and R.E.K. critically reviewed the manuscript for important intellectual content.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research is funded by Brawijaya University through Artificial Intelligence Applied Research and Development Program—Fiscal Year 2024; number of contract: 00067.09/UN10.A0507/B/KS/2024

ORCID iDs

Ni Kadek Indah Sunar Anggreni

Heri Kristianto

Dian Handayani

Yuyun Yueniwati

Paulus Lucky Tirma Irawan

Rulli Rosandi

Rinik Eko Kapti

Avief Destian Purnama

References

Reeves

Cassidy

Abbott

Yap

. Novel technologies for detection and prevention of diabetic foot ulcers. The Science, Etiology and Mechanobiology of Diabetes and its Complications. Amsterdam, Netherlands: Elsevier; 2021:107–122.

Sangam

Hema Priyadarshini

Anand

Prathibha

Purohit

Nalamitha

. Early detection of diabetic foot ulcer. J Phys Conf Ser. 2021;1937(1):012049.

Anthony

Femino

Miller

, et al. Diabetic foot surveillance using mobile phones and automated software messaging, a randomized observational trial. Iowa Orthop J. 2020;40(1):35-42.

Rismayanti

IDA

Nursalam

Farida

, et al. Early detection to prevent foot ulceration among type 2 diabetes mellitus patient: a multi-intervention review. J Public Health Res. 2022;11(2):2752.

Vansteenland

Forss

. What are the current diabetic foot assessment methods in private podiatry practices in Flanders, Belgium: an exploratory mixed method study. J Foot Ankle Res. 2023;16(1):17.

Zhao

Zhou

, et al. Screening behaviors for diabetic foot risk and their influencing factors among general practitioners: a cross-sectional study in Changsha, China. BMC Prim Care. 2023;24(1):68.

Gosak

Svensek

Lorber

Stiglic

. Artificial intelligence based prediction of diabetic foot risk in patients with diabetes: a literature review. Appl Sci. 2023;13(5):2823.

Khandakar

Chowdhury

MEH

Reaz

MBI

, et al. A novel machine learning approach for severity classification of diabetic foot complications using thermogram images. Sensors. 2022;22(11):4249.

Alowais

Alghamdi

Alsuhebany

, et al. Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC Med Educ. 2023;23(1):689.

10.

Najjar

. Redefining radiology: a review of artificial intelligence integration in medical imaging. Diagnostics. 2023;13(17):2760.

11.

Khalifa

Albadawy

. AI in diagnostic imaging: revolutionising accuracy and efficiency. Comput Methods Programs Biomed Update. 2024;5:100146.

12.

Pinto-Coelho

. How artificial intelligence is shaping medical imaging technology: a survey of innovations and applications. Bioengineering (Basel). 2023;10(12):1435.

13.

Hazenberg

CEVB

Aan de Stegge

Van Baal

Moll

Bus

. Telehealth and telemedicine applications for the diabetic foot: a systematic review. Diabetes Metab Res Rev. 2020;36(3):e3247.

14.

Maldonado

Bayareh

Torres

Vera

Gutiérrez

Leija

. Automatic detection of risk zones in diabetic foot soles by processing thermographic images taken in an uncontrolled environment. Infrared Phys Technol. 2020;105:103187.

15.

Hernandez-Contreras

Peregrina-Barreto

Rangel-Magdaleno

Renero-Carrillo

. Plantar thermogram database for the study of diabetic foot complications. IEEE Access. 2019;7:161296-161307.

16.

Chemello

Salvatori

Morettini

Tura

, et al. Artificial intelligence methodologies applied to technologies for screening, diagnosis and care of the diabetic foot: a narrative review. Biosensors. 2022;12(11):985.

17.

Dremin

Marcinkevics

Zherebtsov

, et al. Skin complications of diabetes mellitus revealed by polarized hyperspectral imaging and machine learning. IEEE Trans Med Imaging. 2021;40(4):1207-1216.

18.

Mortelmans

. Doing a literature review with NVivo. In: Mortelmans

, ed. Doing Qualitative Data Analysis with NVivo. Cham: Springer Nature Switzerland; 2025:259–278.

19.

Arteaga-Marrero

Hernández-Guedes

Ortega-Rodríguez

Ruiz-Alzola

. State-of-the-art features for early-stage detection of diabetic foot ulcers based on thermograms. Biomedicines. 2023;11(12):3209.

20.

Zhang

Dong

Guan

, et al. CT-angiography-based outcome prediction on diabetic foot ulcer patients: a statistical learning approach. Diagnostics. 2022;12(5):1076.

21.

Cao

Dey

Ashour

, et al. Diabetic plantar pressure analysis using image fusion. Multimed Tools Appl. 2020;79(15–16):11213-11236.

22.

Mennes

Selles

Van Netten

Van Baal

Steenbergen

Slart

RHJA

. Semi-automatic tracking of laser speckle contrast images of microcirculation in diabetic foot ulcers. Diagnostics. 2020;10(12):1054.

23.

Cao

Zeng

Xie

, et al. Diabetic plantar foot segmentation in active thermography using a two-stage adaptive gamma transform and a deep neural network. Sensors. 2023;23(20):8511.

24.

Filipe

Teixeira

. Automatic classification of foot thermograms using machine learning techniques. Algorithms. 2022;15(7):236.

25.

Toledo Peral

Ramos Becerril

Vega Martínez

Vera Hernández

Leija Salas

Gutiérrez Martínez

. An application for skin macules characterization based on a 3-stage image-processing algorithm for patients with diabetes. J Healthc Eng. 2018;2018:9397105.

26.

Glauberman

Ito-Fujita

Katz

Callahan

. Artificial intelligence in nursing education: opportunities and challenges. Hawaii J Health Soc Welf. 2023;82(12):302-305.

27.

Rony

MKK

Alrazeeni

Akter

, et al. The role of artificial intelligence in enhancing nurses’ work-life balance. J Med Surg Public Health. 2024;3:100135.

28.

McGrow

. Artificial intelligence: essentials for nursing. Nursing. 2019;49(9):46-49.

29.

Alhalafi

Alqahtani

Alqarni

, et al. Utilizing artificial intelligence among patients with diabetes: a systematic review and meta-analysis. Cureus. 2024;16(4):e58713.