Scholarlink

<p>Accurate medical image classification is crucial for effective clinical decision support, improving patient outcomes and reducing healthcare costs. However, developing expert Computer-Aided Diagnosis systems for accurate medical image classification remains a challenging task. Recent advancements in attention mechanisms have revolutionized deep learning-based approaches, leading to improved performance even in applications with limited labeled data. Despite these advances, challenges such as overfitting and poor generalization persist. This work introduces a novel deep learning-based model that incorporates Adaptive Fusion Attention to enhance medical image analysis. The proposed attention module employs a hierarchical fusion of spatial and temporal attention mechanisms, complemented by adaptive refinement. This enables the model to focus on the most discriminative features in medical images, improving its ability to detect abnormalities. Additionally, GRAD-CAM visualizations demonstrate that the model effectively highlights pathological regions while minimizing attention on non-relevant areas. The model is evaluated on three benchmark datasets-APTOS-2019, Figshare, and SARS-CoV-2-demonstrating its effectiveness in Diabetic Retinopathy grading, Brain Tumor Classification, and COVID-19 detection. Experimental results show substantial improvements, achieving 84.56% accuracy for retinopathy grading, 99.60% accuracy for tumor classification, and 99.35% accuracy for COVID-19 detection. These results, along with superior performance across other metrics such as ROC-AUC, and F1-scores, demonstrate the effectiveness of the proposed model with Adaptive Fusion Attention over existing approaches.</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

Adaptive Fusion Attention for enhanced classification and interpretability in medical imaging

  • 期刊论文

  • 发表日期：

    2025年3月16日

查看全文
  • Nagur Shareef Shaik,
  • N. Veeranjaneulu,
  • Jyostna Devi Bodapati

摘要

Accurate medical image classification is crucial for effective clinical decision support, improving patient outcomes and reducing healthcare costs. However, developing expert Computer-Aided Diagnosis systems for accurate medical image classification remains a challenging task. Recent advancements in attention mechanisms have revolutionized deep learning-based approaches, leading to improved performance even in applications with limited labeled data. Despite these advances, challenges such as overfitting and poor generalization persist. This work introduces a novel deep learning-based model that incorporates Adaptive Fusion Attention to enhance medical image analysis. The proposed attention module employs a hierarchical fusion of spatial and temporal attention mechanisms, complemented by adaptive refinement. This enables the model to focus on the most discriminative features in medical images, improving its ability to detect abnormalities. Additionally, GRAD-CAM visualizations demonstrate that the model effectively highlights pathological regions while minimizing attention on non-relevant areas. The model is evaluated on three benchmark datasets-APTOS-2019, Figshare, and SARS-CoV-2-demonstrating its effectiveness in Diabetic Retinopathy grading, Brain Tumor Classification, and COVID-19 detection. Experimental results show substantial improvements, achieving 84.56% accuracy for retinopathy grading, 99.60% accuracy for tumor classification, and 99.35% accuracy for COVID-19 detection. These results, along with superior performance across other metrics such as ROC-AUC, and F1-scores, demonstrate the effectiveness of the proposed model with Adaptive Fusion Attention over existing approaches.