Scholarlink

Gastrointestinal (GI) cancers are a major global health concern, and early detection significantly improves patient outcomes. Current endoscopic methods, however, are hampered by inter-observer variability and high miss rates. Real-time artificial intelligence (AI) offers a promising solution but faces challenges in effective feature extraction and refinement for accurate GI abnormality detection. We propose a novel attention mechanism, V-CBAM, which combines volumetric attention with Convolutional Block Attention Module (CBAM) to enhance feature maps derived from GI images. V-CBAM preserves critical structural information while adaptively refining features through spatial and channel-wise attention maps. Integrated into various transfer learning architectures and evaluated on the largest available GI abnormality dataset, V-CBAM achieves an average improvement of over 7% in precision, recall, and F1 score, demonstrating a statistically significant enhancement in detection performance with the potential to transform clinical practice.

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

Dual Attention-Guided Deep Learning for Multi-class Gastrointestinal Disease Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Md Abrar Shahriar Kabir,
  • Sadia Binte Zahid,
  • Tareque Bashar Ovi,
  • Nomaiya Bashree,
  • Hussain Nyeem,
  • Md Abdul Wahed

摘要

Gastrointestinal (GI) cancers are a major global health concern, and early detection significantly improves patient outcomes. Current endoscopic methods, however, are hampered by inter-observer variability and high miss rates. Real-time artificial intelligence (AI) offers a promising solution but faces challenges in effective feature extraction and refinement for accurate GI abnormality detection. We propose a novel attention mechanism, V-CBAM, which combines volumetric attention with Convolutional Block Attention Module (CBAM) to enhance feature maps derived from GI images. V-CBAM preserves critical structural information while adaptively refining features through spatial and channel-wise attention maps. Integrated into various transfer learning architectures and evaluated on the largest available GI abnormality dataset, V-CBAM achieves an average improvement of over 7% in precision, recall, and F1 score, demonstrating a statistically significant enhancement in detection performance with the potential to transform clinical practice.