Scholarlink

<p>Coronary Artery Disease (CAD) is a severe global health problem caused by atherosclerotic contraction of the artery lumen. Exact and early detection of stenosis or blockage from coronary angiography images is vital for the identification and treatment of people with suspected cardiovascular illness. Recently, Vision Transformer (ViT) networks have emerged as a potential substitute for CNN models. The self-attention mechanism of ViT allows the classifier to analyze the importance of different pixels and dynamically optimize their influence on the output.The key novelty of this work is the development of an explainable ViT architecture enhanced with S-CAM to provide transparent and fine-grained localization of stenotic regions during classification. In this study, we propose an Explainable Vision Transformer (E-ViT) model to classify patients with healthy (no stenosis) and stenosis coronary arteries. The proposed model employs a gradient-free interpretation technique, Score-based Class Activation Mapping (S-CAM), to recognize abnormal regions through linearly weighted activation maps. The S-CAM mechanism enables the ViT model to focus on discriminative regions and provides a deeper understanding of the network’s decision process. The effectiveness of the E-ViT network is evaluated on the open-access ARCADE Phase 1 dataset and compared against state-of-the-art classifiers. The E-ViT model achieves superior CAD classification performance with 99.2% accuracy, 91.2% sensitivity, 99.2% specificity, 97.5% precision, 95.5% recall, 92.3% F1-score, 2.0% false positive rate, and 2.1% false negative rate. These results demonstrate that the proposed E-ViT model achieves competitive performance and is suitable for coronary angiography image classification.</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 vision transformer (E-VIT) based accurate diagnosis of stenosis in X-ray coronary angiography images (XCA)

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • R. Velvizhi,
  • B. Ankayarkanni

摘要

Coronary Artery Disease (CAD) is a severe global health problem caused by atherosclerotic contraction of the artery lumen. Exact and early detection of stenosis or blockage from coronary angiography images is vital for the identification and treatment of people with suspected cardiovascular illness. Recently, Vision Transformer (ViT) networks have emerged as a potential substitute for CNN models. The self-attention mechanism of ViT allows the classifier to analyze the importance of different pixels and dynamically optimize their influence on the output.The key novelty of this work is the development of an explainable ViT architecture enhanced with S-CAM to provide transparent and fine-grained localization of stenotic regions during classification. In this study, we propose an Explainable Vision Transformer (E-ViT) model to classify patients with healthy (no stenosis) and stenosis coronary arteries. The proposed model employs a gradient-free interpretation technique, Score-based Class Activation Mapping (S-CAM), to recognize abnormal regions through linearly weighted activation maps. The S-CAM mechanism enables the ViT model to focus on discriminative regions and provides a deeper understanding of the network’s decision process. The effectiveness of the E-ViT network is evaluated on the open-access ARCADE Phase 1 dataset and compared against state-of-the-art classifiers. The E-ViT model achieves superior CAD classification performance with 99.2% accuracy, 91.2% sensitivity, 99.2% specificity, 97.5% precision, 95.5% recall, 92.3% F1-score, 2.0% false positive rate, and 2.1% false negative rate. These results demonstrate that the proposed E-ViT model achieves competitive performance and is suitable for coronary angiography image classification.