Scholarlink

In the realm of medical image analysis, enhancing the interpretability of convolutional neural networks (CNNs) is critical for their adoption in clinical settings. This study presents a novel approach to improve the explainability of a CNN-based medical image classification by integrating Gradient-weighted Class Activation Mapping (GradCAM) with attention mechanisms. This integration not only highlights crucial image regions that influence classification decisions but also ensures the model’s focus on clinically relevant features and enhances the model performance which is crucial in medical domain. Our experiments on a digital pathology images demonstrate that this method enhances both model performance and interpretability, thereby making AI-driven decisions more transparent and trustworthy for healthcare professionals. Attention in CNN is improving the model performance by 3.17% in accuracy as compared to without attention mechanism and Grad-CAM is on attention-based CNN is improving the interpretability of the model thereby making it more trustworthy.

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

Improving Model Performance and Explainability of Attention-Based CNN Models on Health Image Datasets Using Grad-CAM

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shruti Jaiswal,
  • Gopalan Lakshmi Narasimha,
  • Dharani Prabhu Natarajan,
  • S. Supritha,
  • Krity Kansara,
  • Sushovan Chakraborty,
  • S. Sree Niranjanaa Bose,
  • M. S. H. Shankar Datta

摘要

In the realm of medical image analysis, enhancing the interpretability of convolutional neural networks (CNNs) is critical for their adoption in clinical settings. This study presents a novel approach to improve the explainability of a CNN-based medical image classification by integrating Gradient-weighted Class Activation Mapping (GradCAM) with attention mechanisms. This integration not only highlights crucial image regions that influence classification decisions but also ensures the model’s focus on clinically relevant features and enhances the model performance which is crucial in medical domain. Our experiments on a digital pathology images demonstrate that this method enhances both model performance and interpretability, thereby making AI-driven decisions more transparent and trustworthy for healthcare professionals. Attention in CNN is improving the model performance by 3.17% in accuracy as compared to without attention mechanism and Grad-CAM is on attention-based CNN is improving the interpretability of the model thereby making it more trustworthy.