Scholarlink

Breast cancer is the second leading cause of death among women, with new cases rising globally each year. Early detection can significantly reduce mortality risk. Mammography, the primary screening method, uses X-ray images to identify breast abnormalities. Recent advancements in deep learning have enhanced medical image processing. This article describes a hybrid convolutional neural network (CNN) method for mammography scan-based breast cancer diagnosis that uses minimum-redundancy maximum-relevance (mRMR). The study combined top-performing pre-trained CNN architectures—AlexNet, EfficientNet-B0, and Inception V3—from eight deep-learning models. Features derived from Gradient-weighted Class Activation Mapping (Grad-CAM) were merged with trained features. The mRMR technique optimized these features, which were then classified using SVM and KNN algorithms. The hybrid model achieved a 99.35% accuracy rate in detecting breast cancer with the SVM classifier on the Breast Ultrasound Images Dataset (BUSI). These results demonstrate the effectiveness of combining feature selection methods with CNN models for breast cancer classification.

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 Breast Cancer Detection in BUS Images Using Multimodal Deep Learning and Grad-CAM Fusion

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zeeshan Mubeen,
  • Zulfiqar Ali,
  • Rahmat Ullah,
  • Vishal Krisha Singh,
  • Muhammad Haroon Ahmad

摘要

Breast cancer is the second leading cause of death among women, with new cases rising globally each year. Early detection can significantly reduce mortality risk. Mammography, the primary screening method, uses X-ray images to identify breast abnormalities. Recent advancements in deep learning have enhanced medical image processing. This article describes a hybrid convolutional neural network (CNN) method for mammography scan-based breast cancer diagnosis that uses minimum-redundancy maximum-relevance (mRMR). The study combined top-performing pre-trained CNN architectures—AlexNet, EfficientNet-B0, and Inception V3—from eight deep-learning models. Features derived from Gradient-weighted Class Activation Mapping (Grad-CAM) were merged with trained features. The mRMR technique optimized these features, which were then classified using SVM and KNN algorithms. The hybrid model achieved a 99.35% accuracy rate in detecting breast cancer with the SVM classifier on the Breast Ultrasound Images Dataset (BUSI). These results demonstrate the effectiveness of combining feature selection methods with CNN models for breast cancer classification.