Scholarlink

Early and accurate classification of skin diseases is crucial for improving patient outcomes, particularly for melanoma. This study presents a deep learning-based system leveraging transfer learning to classify dermoscopic images into multiple skin disease categories. We trained and evaluated four convolutional neural networks—ResNet50, EfficientNetB4, InceptionResNetV2, and DenseNet121—on the HAM10000 and ISIC 2019 datasets, applying preprocessing and data augmentation to address class imbalance. ResNet50 and EfficientNetB4 achieved the highest accuracies on HAM10000 (89.5%) and ISIC 2019 (85%), respectively, demonstrating strong generalization across datasets. Unlike prior works focusing on a single dataset or ensemble models, our study provides a systematic cross-dataset evaluation under consistent settings. To extend practical applicability, we developed a web-based tool that allows users to upload dermoscopic images and receive instant predictions with confidence scores, supporting both dermatologists and non-specialists in preliminary skin assessment.

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

Skin Disease Classification Using Transfer Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hadrou Adnan,
  • Bengoud Kenza,
  • Elhadrachi Fatima-ezzahra,
  • Elrherari Asmaa,
  • Elharch Rania

摘要

Early and accurate classification of skin diseases is crucial for improving patient outcomes, particularly for melanoma. This study presents a deep learning-based system leveraging transfer learning to classify dermoscopic images into multiple skin disease categories. We trained and evaluated four convolutional neural networks—ResNet50, EfficientNetB4, InceptionResNetV2, and DenseNet121—on the HAM10000 and ISIC 2019 datasets, applying preprocessing and data augmentation to address class imbalance. ResNet50 and EfficientNetB4 achieved the highest accuracies on HAM10000 (89.5%) and ISIC 2019 (85%), respectively, demonstrating strong generalization across datasets. Unlike prior works focusing on a single dataset or ensemble models, our study provides a systematic cross-dataset evaluation under consistent settings. To extend practical applicability, we developed a web-based tool that allows users to upload dermoscopic images and receive instant predictions with confidence scores, supporting both dermatologists and non-specialists in preliminary skin assessment.