Diabetic Foot Ulcers Detection Model Using a Hybrid Convolutional Neural Networks—Vision Transformers DOI Creative Commons
Abdul Rahaman Wahab Sait,

R. Nagaraj

Diagnostics, Год журнала: 2025, Номер 15(6), С. 736 - 736

Опубликована: Март 15, 2025

Background: Diabetic foot ulcers (DFUs) are severe and common complications of diabetes. Early accurate DFUs classification is essential for effective treatment prevention complications. The existing methods have certain limitations, including limited performance, poor generalization, lack interpretability, restricting their use in clinical settings. Objectives: To overcome these this study proposes an innovative model to achieve robust interpretable classification. Methodology: proposed integrates MobileNet V3-SWIN, LeViT-Peformer, Tensor-based feature fusion, ensemble splines-based Kolmogorov–Arnold Networks (KANs) with Shapley Additive exPlanations (SHAP) values classify severities into ischemia infection classes. In order train generalize the model, authors utilized challenge (DFUC) 2021 2020 datasets. Findings: achieved state-of-the-art outperforming approaches by obtaining average accuracy 98.7%, precision 97.3%, recall 97.4%, F1-score 97.3% on DFUC 2021. On 2020, it maintained a generalization 96.9%, demonstrating superiority over standalone baseline models. findings significant implications research practice. offer platform scalable explainable automated management, improving patient outcomes practices.

Язык: Английский

Diabetic Foot Ulcers Detection Model Using a Hybrid Convolutional Neural Networks—Vision Transformers DOI Creative Commons
Abdul Rahaman Wahab Sait,

R. Nagaraj

Diagnostics, Год журнала: 2025, Номер 15(6), С. 736 - 736

Опубликована: Март 15, 2025

Background: Diabetic foot ulcers (DFUs) are severe and common complications of diabetes. Early accurate DFUs classification is essential for effective treatment prevention complications. The existing methods have certain limitations, including limited performance, poor generalization, lack interpretability, restricting their use in clinical settings. Objectives: To overcome these this study proposes an innovative model to achieve robust interpretable classification. Methodology: proposed integrates MobileNet V3-SWIN, LeViT-Peformer, Tensor-based feature fusion, ensemble splines-based Kolmogorov–Arnold Networks (KANs) with Shapley Additive exPlanations (SHAP) values classify severities into ischemia infection classes. In order train generalize the model, authors utilized challenge (DFUC) 2021 2020 datasets. Findings: achieved state-of-the-art outperforming approaches by obtaining average accuracy 98.7%, precision 97.3%, recall 97.4%, F1-score 97.3% on DFUC 2021. On 2020, it maintained a generalization 96.9%, demonstrating superiority over standalone baseline models. findings significant implications research practice. offer platform scalable explainable automated management, improving patient outcomes practices.

Язык: Английский

Процитировано

0