Scholarlink

<p>Skin cancer remains one of the most common causes of cancer death globally, and it is rising steadily. Early detection is essential for improving survival rates; however, timely diagnosis is often limited due to restricted access to dermatologists, high costs, and delays in clinical procedures. The objective of this study is to design an efficient and user-friendly mobile-based system for the early detection of skin cancer that can be used by general users without the need for complex medical facilities. We propose a hybrid deep learning approach that combines an ensemble of three pre-trained convolutional neural networks (CNNs) (InceptionV3, MobileNetV2, DenseNet169) with a Convolutional Block Attention Module (CBAM) to highlight significant regions of the lesion. A fuzzy rank-based method is proposed for fusing model predictions by nonlinearly transforming rank scores, thereby reducing the impact of less accurate predictions on the final decision. The solution is implemented on a mobile-cloud platform, allowing for lightweight processing on resource-constrained devices. The proposed system is tested on the ISIC dataset for binary classification, achieving an accuracy of 89.09%, which is higher than the individual models and demonstrates greater robustness. Furthermore, explainable AI using Grad-CAM is applied to highlight relevant areas of the skin lesions. In conclusion, the proposed system offers an effective, interpretable, and efficient approach to real-time skin cancer detection, making it a feasible integration into mobile healthcare applications. The relevant source code for this proposed model is publicly available on <a href="https://github.com/sandip-mondal-0248/SkinGlance-ModelAPI">GitHub</a>.</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

Explainable fuzzy fusion of attention-enhanced deep CNNs for mobile-assisted skin cancer detection

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Sandip Mondal,
  • Krishnendu Mahata,
  • Saikat Basu,
  • Koushik Majumder,
  • Mihir Sing,
  • Santanu Chatterjee

摘要

Skin cancer remains one of the most common causes of cancer death globally, and it is rising steadily. Early detection is essential for improving survival rates; however, timely diagnosis is often limited due to restricted access to dermatologists, high costs, and delays in clinical procedures. The objective of this study is to design an efficient and user-friendly mobile-based system for the early detection of skin cancer that can be used by general users without the need for complex medical facilities. We propose a hybrid deep learning approach that combines an ensemble of three pre-trained convolutional neural networks (CNNs) (InceptionV3, MobileNetV2, DenseNet169) with a Convolutional Block Attention Module (CBAM) to highlight significant regions of the lesion. A fuzzy rank-based method is proposed for fusing model predictions by nonlinearly transforming rank scores, thereby reducing the impact of less accurate predictions on the final decision. The solution is implemented on a mobile-cloud platform, allowing for lightweight processing on resource-constrained devices. The proposed system is tested on the ISIC dataset for binary classification, achieving an accuracy of 89.09%, which is higher than the individual models and demonstrates greater robustness. Furthermore, explainable AI using Grad-CAM is applied to highlight relevant areas of the skin lesions. In conclusion, the proposed system offers an effective, interpretable, and efficient approach to real-time skin cancer detection, making it a feasible integration into mobile healthcare applications. The relevant source code for this proposed model is publicly available on GitHub.