Scholarlink

<p>Skin cancer affects a growing number of more than 1.5 million people worldwide annually, highlighting the urgent need for accurate and autonomous detection methods. Among these, non-melanoma skin cancers account for approximately 1.2 million cases, while melanoma, the most aggressive form, contributes to over 325,000 new cases each year (World Health Organization in, Skin cancers, 2023. <a href="https://www.who.int/news-room/fact-sheets/detail/skin-cancers">https://www.who.int/news-room/fact-sheets/detail/skin-cancers</a>). Traditional medical practices, like examining skin tissue samples, lack precise results. Thus, there is a need for advanced computer vision detection systems. This research presents the BCB–CSPA network as an optimized deep learning system that merges bipartite convoluted blocks (BCB) and condensed semantic perceptual attention (CSPA) with pyramid multi-scale convolution for better skin lesion classification. The model strengthens feature extraction by using BCBs to refine elements while focusing on the most essential features at various perception scales. The BCB–CSPA network demonstrated its effectiveness on dermatological datasets by achieving 95.92% accuracy, 94.32% precision, 94.62% recall, 94.47% F1-score, 94.62% specificity, 95.23% Cohen's Kappa score, and 92.23% Jaccard similarity. Our system shows highly accurate predictions for all skin cancer groups, especially in AKIEC cancer, which reached 99.69% accuracy. Tests against present analysis models show that BCB–CSPA produces better diagnostic results, plus maintains an easily comprehensible design strategy. Explainability methods, such as Grad-CAM and visual attention, show how the model detects lesions and uses these details to make decisions on diagnosis. BCB–CSPA generates precise results to help patients and doctors trust the AI-supported skin examinations.</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

Comparing Bipartite Convoluted and Attention-Driven Methods for Skin Cancer Detection: A Review of Explainable AI and Transfer Learning Strategies

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Tathagat Banerjee

摘要

Skin cancer affects a growing number of more than 1.5 million people worldwide annually, highlighting the urgent need for accurate and autonomous detection methods. Among these, non-melanoma skin cancers account for approximately 1.2 million cases, while melanoma, the most aggressive form, contributes to over 325,000 new cases each year (World Health Organization in, Skin cancers, 2023. https://www.who.int/news-room/fact-sheets/detail/skin-cancers). Traditional medical practices, like examining skin tissue samples, lack precise results. Thus, there is a need for advanced computer vision detection systems. This research presents the BCB–CSPA network as an optimized deep learning system that merges bipartite convoluted blocks (BCB) and condensed semantic perceptual attention (CSPA) with pyramid multi-scale convolution for better skin lesion classification. The model strengthens feature extraction by using BCBs to refine elements while focusing on the most essential features at various perception scales. The BCB–CSPA network demonstrated its effectiveness on dermatological datasets by achieving 95.92% accuracy, 94.32% precision, 94.62% recall, 94.47% F1-score, 94.62% specificity, 95.23% Cohen's Kappa score, and 92.23% Jaccard similarity. Our system shows highly accurate predictions for all skin cancer groups, especially in AKIEC cancer, which reached 99.69% accuracy. Tests against present analysis models show that BCB–CSPA produces better diagnostic results, plus maintains an easily comprehensible design strategy. Explainability methods, such as Grad-CAM and visual attention, show how the model detects lesions and uses these details to make decisions on diagnosis. BCB–CSPA generates precise results to help patients and doctors trust the AI-supported skin examinations.