Hybrid Deep Learning Framework for Diabetic Wound Assessment: Integrating Segmentation, Classification, and Explainable AI in Cloud-Based Telemedicine Systems

Abstract

Diabetic foot ulcers (DFUs) are among the most serious complications of diabetes mellitus and frequently lead to infection, hospitalization, and lower-limb amputation. Early and accurate assessment of wound severity is therefore essential for preventing complications and supporting clinical decision-making, particularly in telemedicine settings. This study proposes a hybrid deep learning framework for automated DFU evaluation that integrates image segmentation, ordinal-aware classification, and visual interpretability. The framework employs a U-Net-based segmentation module to isolate ulcer regions, followed by an EfficientNet-B3 classifier trained with consistent rank logits to grade wound severity according to the Meggitt–Wagner scale. Gradient-weighted Class Activation Mapping (Grad-CAM) visualization is incorporated to highlight discriminative regions that influence model predictions, thereby improving clinical transparency. Experimental evaluation on a dataset of 6,500 annotated DFU images achieved an Intersection over Union of 0.861, a Dice coefficient of 0.924, a classification accuracy of 92.3%, and a macro-F1 score of 0.904, with an average inference time of under 2 s per image. These results demonstrate that the proposed hybrid pipeline enables precise, interpretable, and computationally efficient wound assessment suitable for real-time telemedicine and mobile health applications. The framework provides a scalable and explainable artificial intelligence approach that supports consistent grading and early detection in diabetic foot management.