Scholarlink

Breast cancer poses a significant challenge globally, underscoring the critical need for accurate early detection to enhance treatment outcomes. Detecting breast calcifications, important signs of cancer, is tough due to tumor diversity and complexity. This paper introduces a novel method for real-time classification of mammograms, aiming to distinguish between benign and malignant cases uniquely. The approach begins with meticulous image preprocessing to reduce noise and enhance quality. It combines two innovative parts: a vision transformer (ViT) for understanding the overall context of a mammogram image and a CNN using transfer learning to extract detailed visual features. Using the real-time dataset, proposed model achieves significant improvements in accuracy, reaching 98.26% with an F1-score of 92.58%, precision of 93.25%, recall of 95.45%, and an ROC AUC score of 0.96, demonstrating its robust ability to distinguish between benign and malignant cases. This hybrid approach shows promise in enhancing breast cancer diagnosis, leveraging advanced deep learning techniques for accurate and reliable mammogram analysis.

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

Hybrid Fusion Approach Using CNN and Vision Transformer for Breast Cancer Classification from Mammograms

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tunisha Varshney,
  • Karan Verma,
  • Arshpreet Kaur,
  • Sunil Kumar Puri

摘要

Breast cancer poses a significant challenge globally, underscoring the critical need for accurate early detection to enhance treatment outcomes. Detecting breast calcifications, important signs of cancer, is tough due to tumor diversity and complexity. This paper introduces a novel method for real-time classification of mammograms, aiming to distinguish between benign and malignant cases uniquely. The approach begins with meticulous image preprocessing to reduce noise and enhance quality. It combines two innovative parts: a vision transformer (ViT) for understanding the overall context of a mammogram image and a CNN using transfer learning to extract detailed visual features. Using the real-time dataset, proposed model achieves significant improvements in accuracy, reaching 98.26% with an F1-score of 92.58%, precision of 93.25%, recall of 95.45%, and an ROC AUC score of 0.96, demonstrating its robust ability to distinguish between benign and malignant cases. This hybrid approach shows promise in enhancing breast cancer diagnosis, leveraging advanced deep learning techniques for accurate and reliable mammogram analysis.