Research Progress on Artificial Intelligence in the Management of Diabetic Foot Ulcers

Abstract

As a typical chronic complication of diabetes, the management of diabetic foot ulcers has long faced multiple challenges, including the highly subjective nature of wound assessment, patients‘limited capacity for self-care at home, and the difficulty in predicting healing outcomes. The emergence of artificial intelligence (AI) technology offers new approaches to addressing these challenges. Based on recent domestic and international literature, this paper systematically reviews the research progress of artificial intelligence in the care of diabetic foot ulcers, focusing on four key dimensions: intelligent wound assessment (eg deep learning-based image analysis models achieving an F1 score of 93.1% for infection identification and 92.8% accuracy for area assessment), patient self-health management (eg mobile health applications and wearable monitoring devices), the development and application of smart dressings (eg responsive biomaterials and sensor integration), and the prediction of healing trends (eg machine learning prediction models, with support vector machines achieving 85.3% accuracy and an AUC of 0.922 in predicting complete DFU healing). It summarizes key technological achievements and the current status of clinical translation, provides an in-depth analysis of existing shortcomings (including insufficient algorithm generalizability, lack of prospective clinical validation, and absence of data standardization), and outlines future development directions (encompassing multimodal integration, explainable artificial intelligence, and implementation science research), with the aim of offering guidance for the further application of artificial intelligence technology in the field of chronic wound care.

Introduction

Diabetic foot ulcers (DFUs) are one of the serious chronic complications faced by people with diabetes. Globally, the lifetime prevalence of DFUs ranges from 19% to 34%,¹ with an overall prevalence of approximately 6.3%.² Due to the complex healing mechanisms and high recurrence rate of DFUs, they have become the leading cause of non-traumatic amputations, significantly impacting patients‘quality of life whilst placing a substantial economic burden on healthcare systems worldwide.³ Study⁴ indicates that the 1-year postoperative recurrence rate of DFU patients is as high as 40%. Traditional DFU care relies heavily on the clinical experience and manual procedures of nursing staff. Processes such as wound assessment, dressing selection and prognosis evaluation are highly subjective and vary greatly between individuals, making it difficult to achieve precise and continuous care objectives.⁵ In recent years, rapid advancements in Artificial Intelligence (AI) technology have opened up new avenues for the development of DFU care. By integrating technologies such as deep learning, machine learning and smart sensing, AI is gradually being embedded throughout the entire DFU care process, driving the evolution of the nursing paradigm from “experience-oriented” to “data-driven”.³ In the healthcare sector, AI has been successfully applied to imaging-assisted diagnostic scenarios such as pulmonary nodule detection and diabetic retinopathy screening, and has also been preliminarily applied in the nursing field to automatic staging of pressure injuries and dynamic prediction of fall risk in hospitalized patients.⁶ These successful practices provide a technical foundation and methodological insights for the transfer of AI applications to DFU care. However, current research still faces key bottlenecks, including insufficient algorithm generalizability (with most studies relying on single-center, small-sample data and lacking multi-center, large-scale standardized DFU multimodal databases), the “black-box” nature of most AI models leading to low clinical trust, and poor integration of AI tools with clinical workflows, all of which urgently require systematic review and integration. This paper systematically reviews the current state of research on AI in the field of DFU care, analyses existing issues and challenges, with the aim of providing theoretical support and practical insights for the intelligent upgrading of DFU care.

Overview and Applications of Artificial Intelligence

Basic Concepts and Application Scenarios of Artificial Intelligence

Artificial intelligence is a discipline that studies and develops the relevant theories, methods, technologies and application systems used to simulate, extend and enhance human intelligence.⁷ In the healthcare sector, the core technological branches closely related to DFU care primarily cover the following four areas. Firstly, machine learning (ML) utilises algorithms to automatically learn patterns and regularities from data in order to achieve classification, recognition or outcome prediction; it is the foundational technology for AI applications in the medical field.⁷ Second, Deep Learning (DL), as a subset of Machine Learning, automatically extracts hierarchical features from data by constructing multi-layer neural networks. It performs particularly well in complex tasks such as image recognition and natural language processing; amongst these, Convolutional Neural Networks (CNNs) are widely used for the automatic segmentation of wound images and the identification of tissue types. Thirdly, computer vision endows computers with the ability to “understand” and “analyse” digital images, and is used for the automatic delineation of wound boundaries, area measurement, and the classification and identification of granulation tissue, slough and necrotic tissue.⁸ Fourthly, expert systems and clinical decision support systems (CDSS) formalise the knowledge of domain experts and evidence-based guidelines into a rule base that can be executed by computers, thereby assisting clinicians in making diagnostic and treatment decisions.⁶

Typical Applications of Artificial Intelligence in Healthcare

The application of AI in healthcare has expanded from medical imaging-assisted diagnosis to a wide range of clinical scenarios. In the field of imaging diagnosis, AI-assisted detection of pulmonary nodules, screening for diabetic retinopathy, and interpretation of mammograms have achieved accuracy levels that match or even exceed those of specialist physicians. In chronic disease management, the combination of wearable devices and AI algorithms has enabled real-time monitoring of physiological parameters such as heart rate, blood pressure and blood glucose, as well as early warning of abnormalities, thereby enhancing patients‘self-management capabilities.⁹ In the nursing sector, AI has been preliminarily applied to the automatic staging of pressure injuries, dynamic prediction of fall risk in hospitalised patients, and nutritional status assessment.⁶ These successful practices provide a technical foundation and methodological insights for the transfer of AI applications to DFU care, whilst also accumulating valuable experience for further exploration of new AI-driven wound care models.

The Application of Artificial Intelligence in the Care of Diabetic Foot Ulcers

AI-Based Local Assessment of Diabetic Foot Ulcers and Patient Self-Management

Traditional local assessment of diabetic foot ulcers (DFUs) primarily relies on nursing staff to visually inspect the wound and manually measure its size using a ruler, whilst relying on clinical experience to determine tissue types and signs of infection. This method suffers from issues such as inconsistent assessment criteria, poor inter-observer agreement, and time-consuming procedures. The limitations of traditional local assessment of DFUs are evident in the fact that manual assessment makes it difficult to accurately define the boundaries of the wound (particularly when the wound margins exhibit skin deformity, hyperkeratosis or blurred borders), and the differentiation between different tissue types (granulation tissue, slough and necrotic tissue) is highly dependent on individual experience, resulting in low inter-observer reliability.¹⁰ AI-driven wound assessment systems offer an effective solution to these challenges. Automated wound analysis technology based on deep learning models can rapidly and accurately perform wound region segmentation, tissue type identification (granulation tissue, slough, necrotic tissue), and quantitative measurement of key parameters.¹¹ A typical AI assessment workflow is: patients capture wound images via mobile devices → image preprocessing (denoising, illumination correction) → deep learning models automatically perform wound region segmentation and tissue type identification → output classification results and related quantitative parameters (area, depth, etc) → results are synchronized to the healthcare provider’s terminal to assist clinical staff in making judgments. Studies have shown that AI-driven DFU detection and classification models can achieve sensitivity and specificity of over 90%.¹² Mostafavi et al¹³ developed a mobile application based on the lightweight MobileNetV3 neural network architecture. This application can automatically assess each component of the SINBAD scoring system for DFUs, achieving an F1 score of 93.1% (recall 96.2%) for bacterial infection identification and an F1 score of 89.8% (precision 92.8%) for area assessment. This lightweight design enables its deployment in resource-limited primary care settings and community environments, helping to bridge the gap in DFU assessment capabilities across different regions.

AI-Based Self-Management and Remote Monitoring for DFU Patients

Home care for DFU patients following discharge has consistently been a weak link in clinical practice. Due to limited specialist knowledge, patients often find it difficult to independently identify signs of wound deterioration and lack the ability to recognise complications such as infection at an early stage. With the aid of mobile health applications utilising AI technology, patients need only take a photograph of the wound to receive an immediate, automated assessment; the relevant data and images can also be synchronised and uploaded to healthcare professionals“ devices. In terms of user interface design, such applications need to fully consider the differences in patients” digital literacy, and enhance usability and patient adherence through measures such as simplified operational procedures (eg, photo-based assessment), voice-guided features, visualised healing trend charts, and personalized feedback.^14,15 Evidence from existing systematic reviews indicates that self-management interventions for DFU using mobile health applications can effectively improve patients‘self-efficacy, glycaemic control, and adherence to preventive foot care.¹⁵ Concurrently, wearable temperature sensors and pressure-sensitive insoles, combined with AI algorithm analysis, enable continuous monitoring of inflammatory responses in the foot (where localised increases in skin temperature often indicate infection or active inflammation) and abnormal pressure distribution on the sole, issuing timely alerts to patients and healthcare professionals before ulcers form or recur. Such wearable devices are primarily designed for daily use by patients in the home environment, achieving “proactive prevention” through continuous monitoring of the foot microenvironment. This marks a paradigm shift in DFU management from “reactive treatment” to “proactive prevention”, which is of significant clinical importance; however, the current cost of such devices remains relatively high, which may limit widespread adoption in resource-limited regions.¹⁶ In the remote monitoring system, AI tools play an auxiliary and bridging role, with patients receiving real-time assessments and alerts through the application, whilst maintaining ongoing contact with clinicians to ensure professional guidance and timely intervention.

AI-Assisted Dressing Preparation and the Application of Smart Dressings

The integration of AI technology into the field of DFU dressings is primarily manifested in two dimensions. Firstly, AI-assisted screening and design of dressing materials. Traditional dressing R&D faces bottlenecks such as lengthy development cycles and high trial-and-error costs. By analysing the intrinsic relationships between the physicochemical properties and microstructure of biomaterials and wound healing outcomes, machine learning models can accelerate the selection and structural optimisation of novel dressing materials, thereby effectively shortening development cycles and reducing costs.¹⁷ Secondly, the development of smart dressing systems. Smart dressings can be classified into the following two categories based on their functions: (1) Passive monitoring dressings — researchers have integrated flexible biosensors and microprocessors with traditional dressing substrates to develop novel dressings capable of real-time detection of changes in the wound microenvironment. These smart dressings can monitor various wound microenvironment parameters, such as pH, temperature, humidity and inflammatory factors, and transmit data to external receiving devices via built-in near-field communication (NFC) or Bluetooth modules.¹⁸ (2) Active closed-loop therapeutic systems — namely, the construction of a closed-loop smart dressing system that integrates “sensing, decision-making and intervention”. This system continuously monitors the wound condition via biosensors within the dressing, whilst edge AI chips perform real-time analysis of the sensory data. Upon detecting specific signals—such as elevated pH indicating infection or abnormally high local glucose concentrations—it automatically triggers the on-demand release of loaded anti-inflammatory drugs or growth factors.¹⁹ For example, researchers have drawn inspiration from the biological properties of snail mucus, utilising the glycosaminoglycans and gelatin it contains to construct biomimetic gel patches. Through dynamic boronate bond technology, the tea polyphenol epigallocatechin gallate (EGCG) is “anchored” within the gel network. When local glucose concentrations at the wound site rise, the dynamic bonds within the gel network break in response, accelerating drug release and thereby enabling a smart ’on-demand‘delivery mechanism.²⁰ The precise operation of such systems relies on AI algorithms to accurately control the timing and dosage of drug release, representing a key technological pathway for achieving personalised wound care for DFUs.

AI-Based Prediction of Diabetic Foot Ulcer Healing Trends

Traditional healing prediction models are predominantly based on logistic regression analysis, incorporating a limited number of predictor variables; consequently, the models’ discriminatory power and calibration are often less than ideal. Accurately predicting the healing trends of diabetic foot ulcers (DFUs) is of great significance for optimising treatment strategies, rationally allocating medical resources, and facilitating communication between healthcare providers and patients.²¹ Machine learning and deep learning models have significantly improved the accuracy of healing predictions by integrating multi-dimensional, multi-modal data (including wound image features, laboratory test results, patient demographic information, and medical history). In a study comparing seven machine learning models (including Support Vector Machines, Random Forests, XGBoost, K-Nearest Neighbours, logistic regression, Naive Bayes, and decision trees) for predicting complete DFU healing, the Support Vector Machine (SVM) model achieved a prediction accuracy of 85.3% and an Area Under the Curve (AUC) of 0.922, with good model calibration, demonstrating high clinical utility.²² Another study utilised a three-layer Long Short-Term Memory (LSTM) recurrent neural network, based on multi-time-point follow-up data from 1766 DFU patients, achieving an 80% accuracy rate in healing prediction and an AUC of 0.85. The study further analysed key factors influencing healing and found that wound depth and area at the initial visit, along with the percentage of granulation tissue at the second follow-up, were the indicators with the highest predictive value.²³ Furthermore, AI-based healing trend prediction can be extended to non-clinical scenarios—patients regularly upload wound images and self-monitoring data via mobile applications, and AI models perform real-time analysis to deliver healing trend assessments and alerts to patients, enabling them to obtain professional healing progress judgments at home and facilitating timely identification of poor healing risk and proactive help-seeking.¹⁹ These studies demonstrate that AI is not only capable of “predicting outcomes” but can also assist clinicians in identifying which factors are key targets driving the healing process, thereby providing data support for the development of personalised interventions.

AI-Based Clinical Decision Support System for Chronic Wound Care

The seamless integration of the aforementioned assessment, monitoring and prediction functions into an actionable and interpretable clinical decision support system represents the ultimate form of AI-enabled DFU care. This is manifested in the deep integration of evidence-based guidelines, patient-specific data and AI predictive models, enabling the provision of personalised and interpretable decision recommendations to clinicians at key clinical junctures—such as the fitting of pressure-relieving footwear, dressing selection, setting follow-up intervals, and determining the timing for initiating antimicrobial therapy.²⁴ A complete DFU-CDSS workflow can be summarised as: multimodal data collection (wound images processed via computer vision, laboratory indicators, demographic information, medical history, etc) → AI model analysis (deep learning for image segmentation and classification, machine learning for risk prediction) → decision engine integration (evidence-based guidelines + patient-specific data + AI prediction models) → output of interpretable decision recommendations to clinicians. Taking the prescription of DFU pressure-relieving footwear as an example, researchers have constructed a five-tier CDSS architecture: the first tier defines the minimum viable dataset and standardises collection criteria; the second tier employs a hybrid decision-making architecture, integrating a rule-based reasoning engine, optimisation algorithms and interpretable machine learning models; the third tier outputs structured features to ensure that the basis for decisions is clear and traceable; the fourth tier establishes mechanisms for continuous validation and dynamic updates; the fifth layer enables seamless integration with clinical workflows and telemedicine systems.²⁵ However, no mature CDSS system has yet been genuinely applied in the clinical prescription practice of DFU pressure-relieving footwear. The main obstacles lie in the fact that relevant knowledge is scattered across clinical guidelines, sensor research and machine learning prototypes, resulting in a lack of a unified decision-making logic framework; most AI models exhibit “black-box” characteristics, making it difficult for clinicians to understand the rationale behind recommendations, which in turn affects trust and willingness to use the system; and there is a lack of large-scale prospective validation in real clinical settings, with evidence regarding the system’s effectiveness and safety requiring further substantiation.²⁴

Issues and Outlook

Although AI holds great promise in the field of DFU care, the translation of research findings into clinical practice still faces numerous bottlenecks: ① In terms of data and algorithms, existing studies largely rely on single-centre, small-sample data, and there is a lack of multi-centre, large-scale, standardised multimodal DFU databases; conditions for wound image acquisition (light intensity, shooting angle, subject-camera distance, and device model) vary considerably, severely limiting the robustness and generalisability of algorithms; the decision-making logic of most models is obscure, making it difficult for clinical staff to understand and accept the basis for AI judgements;²⁶ ② Clinical validation and system integration: Rigorously designed, large-sample prospective clinical trials remain scarce, and there is a lack of systematic evaluation of the clinical efficacy and health economic value of AI systems; most AI tools have poor interoperability with hospital information systems and electronic health records, requiring additional manual data entry and system switching during use, which conversely increases the clinical workload;²⁷ ③ Ethics and accessibility: Responsibility for AI-assisted decision-making remains unclear; there is a lack of consensus on liability when AI recommendations conflict with clinical judgement or when the system malfunctions; stricter technical and institutional safeguards are urgently needed to protect the privacy and information security of patients‘wound images and health data; current AI-DFU research is predominantly concentrated in high-income countries, and the digital divide may exacerbate imbalances in the distribution of healthcare resources between different regions and populations of varying economic status.²⁸

To address the aforementioned bottlenecks, future research should focus on the following areas: Firstly, establishing a multi-centre, large-scale, standardised multimodal DFU database, with unified data collection protocols and annotation standards; Secondly, developing explainable AI technologies to enhance the comprehensibility and traceability of model outputs, thereby increasing clinicians‘trust and willingness to adopt such systems; Thirdly, advancing “human-in-the-loop” hybrid intelligent systems that organically integrate AI’s predictive capabilities with clinical experts‘judgement, thereby establishing a human-machine collaborative decision-making model; Fourthly, conducting implementation research to systematically evaluate the integration pathways, implementation barriers, user acceptance and health economic value of AI tools in real-world clinical settings, thereby providing evidence-based support for the sustainable application of AI in DFU care.^17,29–31

Summary

In summary, artificial intelligence is reshaping the care model for diabetic foot ulcers across multiple dimensions, ranging from local wound assessment and patient self-management to the development and application of smart dressings, the prediction of healing trends, and clinical decision support.³² The application of deep learning and computer vision technologies has transformed wound assessment from a subjective, experience-based judgement into a precise quantitative analysis;³³ smart dressings and wearable monitoring devices have extended care services from healthcare facilities into patients’ home environments;³⁴ whilst machine learning predictive models provide data support for formulating personalised treatment plans and optimising the allocation of healthcare resources.^35,36

At the clinical practice level, AI-assisted image analysis tools have demonstrated high accuracy in identifying infections and assessing wound area,¹³ indicating potential for wider adoption in primary care settings; the integration of mobile health applications with wearable devices¹⁶ offers a viable approach to home management and early warning; machine learning predictive models (such as support vector machines and long short-term memory networks) have demonstrated clinical utility in screening high-risk populations for poor wound healing;^22,23 whilst the application of artificial intelligence clinical decision support systems (AI-CDSS) in standardised decision-making still requires prospective validation.²⁴ At the academic research level, current translational efforts still face numerous challenges, such as insufficient data standardisation, a lack of model interpretability, weak clinical validation, and a scarcity of health economic evidence.^26–28 Future work should focus on developing interpretable artificial intelligence technologies,³⁰ advancing hybrid intelligence systems based on human–machine collaboration,³¹ and conducting implementation studies to evaluate the effectiveness and cost-effectiveness of integration in real-world clinical settings.²⁹ Furthermore, future research might consider presenting the correspondence between AI applications and methods in tabular form, thereby providing a clearer overview of the technological landscape and offering readers a systematic framework for reference and comparison. At present, a gap persists between “technical feasibility” and “clinical usability”; data standardisation, model interpretability, prospective clinical validation and equitable access remain critical bottlenecks that require urgent resolution. Research should be guided by clinical needs, strengthening interdisciplinary collaboration between medicine and engineering to advance AI technology from the laboratory to clinical practice, ultimately enabling patients with diabetic foot ulcers (DFUs) to truly benefit from new AI-driven intelligent care models.³¹

Funding Statement

No funding was received for this study.

Data Sharing Statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Author Contributions

Jinxia Du: Conceptualization; Data curation Methodology Writing – original draft; Writing – review & editing; Visualization; Supervision; Project administration.

Xiaoli Liu: Conceptualization; Methodology; Supervision; Project administration; Writing – review & editing.

Guihui Wu: Conceptualization; Formal Analysis; Writing – review & editing.

Yin Huang: Data curation Writing – review & editing.

Yan Zhang: Data curation Writing – review & editing.

Xiang Long: Data curation; Writing – review & editing.

Xiaoyi Yang: Data curation Writing – review & editing.

Yan Zhou: Data curation; Writing – review & editing.

All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.