Scholarlink

This study presents a novel comparative analysis of five deep learning architectures for diabetic foot ulcer (DFU) detection and classification in medical imaging. The proposed framework integrates various state-of-the-art models: an LSTM-CNN hybrid for temporal feature extraction, a modified ResNet50 with Grad-CAM visualization, a CNN enhanced with LIME interpretability, a size-adaptive FCN for segmentation, and an attention-based EfficientNet model. The key innovation lies in the comprehensive integration of explainability techniques with each model, addressing the critical need for interpretable results in clinical settings. The combined approach demonstrates superior performance in DFU classification while providing clinicians with transparent decision-making tools through saliency maps, visual interpretability, and detailed severity assessments. This research advances the field by establishing a balanced framework between computational efficiency and clinical applicability, potentially transforming the landscape of automated DFU detection in medical imaging.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning for Diabetic Foot Ulcer Classification, Severity Assessment, and Explainable AI

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vashitva Bagga,
  • Shanaya Aggarwal,
  • Himanshi Sharma

摘要

This study presents a novel comparative analysis of five deep learning architectures for diabetic foot ulcer (DFU) detection and classification in medical imaging. The proposed framework integrates various state-of-the-art models: an LSTM-CNN hybrid for temporal feature extraction, a modified ResNet50 with Grad-CAM visualization, a CNN enhanced with LIME interpretability, a size-adaptive FCN for segmentation, and an attention-based EfficientNet model. The key innovation lies in the comprehensive integration of explainability techniques with each model, addressing the critical need for interpretable results in clinical settings. The combined approach demonstrates superior performance in DFU classification while providing clinicians with transparent decision-making tools through saliency maps, visual interpretability, and detailed severity assessments. This research advances the field by establishing a balanced framework between computational efficiency and clinical applicability, potentially transforming the landscape of automated DFU detection in medical imaging.