Scholarlink

<p>Skin diseases represent a significant burden in healthcare management due to delayed diagnosis, visual complexity, and variability in clinical decision-making. Manual methods are prone to misdiagnosis, time-consuming, and often lack consistency. With the rise of clinical dermatology cases, there is a critical need for automated, explainable systems to support reliable skin disease diagnosis. This study proposes an Explainable Artificial Intelligence (XAI)-driven Hybrid Multi-Class Skin Disease Classification (HMSC) framework designed for a clinical dermatology system. This research work utilized benchmark Dermnet dataset for the evaluation of HMSC framework. The framework integrates deep learning models VGG16, EfficientNet-B0, Inception-V3, and MobileNet-V3 for automatic feature extraction and fusion from dermoscopic images of five common skin conditions: acne, psoriasis, eczema, urticaria, and scabies. LIME XAI technique is incorporated to improve interpretability by visually highlighting the most influential image regions in each prediction. The HMSC proposed framework gives an accuracy of 98.70%, with 98% precision, 98.70% recall, 98.50% specificity, 99.30% AUC-ROC, 96% MCC, and 98.30% F1-score. The model exhibited excellent cross-class generalization and interpretability of skin diseases on test data. The reliability and stability of the proposed framework are further tested using k-fold cross-validation technique and testing it on another 2 independent datasets, HAM1000 and ISIC-2019. These results suggest that the HMSC framework is a reliable and transparent diagnostic framework for skin disease classification. HMSC scalability and explainability show the potential for its real-world deployment in healthcare systems after some ethical and diverse demographic skin tone considerations. This framework has the capacity to minimize the clinician workload with improved diagnostic efficiency and support consistent decision-making.</p>

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

Feature Fusion and Explainable Deep Learning Framework for Intelligent Skin Disease Classification Using Clinical Dermatology Images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月22日

查看全文
  • Muhammad Shafiq,
  • Najia Saher,
  • Bayan Alabdullah,
  • Faisal Shahzad,
  • Muhammad Umer,
  • Khaled Alnowaiser,
  • Yunyoung Nam,
  • Yongwon Cho

摘要

Skin diseases represent a significant burden in healthcare management due to delayed diagnosis, visual complexity, and variability in clinical decision-making. Manual methods are prone to misdiagnosis, time-consuming, and often lack consistency. With the rise of clinical dermatology cases, there is a critical need for automated, explainable systems to support reliable skin disease diagnosis. This study proposes an Explainable Artificial Intelligence (XAI)-driven Hybrid Multi-Class Skin Disease Classification (HMSC) framework designed for a clinical dermatology system. This research work utilized benchmark Dermnet dataset for the evaluation of HMSC framework. The framework integrates deep learning models VGG16, EfficientNet-B0, Inception-V3, and MobileNet-V3 for automatic feature extraction and fusion from dermoscopic images of five common skin conditions: acne, psoriasis, eczema, urticaria, and scabies. LIME XAI technique is incorporated to improve interpretability by visually highlighting the most influential image regions in each prediction. The HMSC proposed framework gives an accuracy of 98.70%, with 98% precision, 98.70% recall, 98.50% specificity, 99.30% AUC-ROC, 96% MCC, and 98.30% F1-score. The model exhibited excellent cross-class generalization and interpretability of skin diseases on test data. The reliability and stability of the proposed framework are further tested using k-fold cross-validation technique and testing it on another 2 independent datasets, HAM1000 and ISIC-2019. These results suggest that the HMSC framework is a reliable and transparent diagnostic framework for skin disease classification. HMSC scalability and explainability show the potential for its real-world deployment in healthcare systems after some ethical and diverse demographic skin tone considerations. This framework has the capacity to minimize the clinician workload with improved diagnostic efficiency and support consistent decision-making.