Scholarlink

<p>Breast cancer (BC) remains a ubiquitous malignancy affecting millions of women worldwide, with rising cases in the number of diagnosed cases. Advanced computer-aided diagnosis technology and deep learning (DL) methods have significantly improved early diagnosis, leading to higher survival rates and better treatment outcomes. This study proposes an efficient hybrid DL-based approach that combines a convolutional neural network (CNN), attention mechanism (AM), and residual connections (RC), termed the CAR model. The model uses CNN for custom-designed feature extraction to learn image-specific representations. The AM block focuses on specific image features, and the RC block addresses the vanishing gradient problem mapping the original extracted features and the attended weight features. The CAR model is evaluated for both binary and multi-class breast cancer classification across two benchmark datasets—BUSI and MIAS—using various color schemes (RGB, RGBA, and gray scale) and stratified cross-validation (3-CV, 5-CV, 7-CV). The results demonstrate that gray scale images consistently yield better classification performance compared to RGB and RGBA, achieving accuracies of 98.33% on BUSI and 98.90% on MIAS datasets. These findings underscore the significance of color information in medical imaging and highlight the model's ability to generalize across different cross-validation folds, with 7-CV yielding the best results. The CAR model also outperforms existing state-of-the-art models while maintaining computational efficiency, making it an efficient approach for breast cancer diagnosis.</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

Efficient Breast Cancer Classification Using a Hybrid CNN-Attention-Residual Connection (CAR) Model

  • 期刊论文

  • 发表日期：

    2025年4月17日

查看全文
  • Sasanka Sekhar Dalai,
  • Bharat Jyoti Ranjan Sahu,
  • Ibanga Kpereobong Friday

摘要

Breast cancer (BC) remains a ubiquitous malignancy affecting millions of women worldwide, with rising cases in the number of diagnosed cases. Advanced computer-aided diagnosis technology and deep learning (DL) methods have significantly improved early diagnosis, leading to higher survival rates and better treatment outcomes. This study proposes an efficient hybrid DL-based approach that combines a convolutional neural network (CNN), attention mechanism (AM), and residual connections (RC), termed the CAR model. The model uses CNN for custom-designed feature extraction to learn image-specific representations. The AM block focuses on specific image features, and the RC block addresses the vanishing gradient problem mapping the original extracted features and the attended weight features. The CAR model is evaluated for both binary and multi-class breast cancer classification across two benchmark datasets—BUSI and MIAS—using various color schemes (RGB, RGBA, and gray scale) and stratified cross-validation (3-CV, 5-CV, 7-CV). The results demonstrate that gray scale images consistently yield better classification performance compared to RGB and RGBA, achieving accuracies of 98.33% on BUSI and 98.90% on MIAS datasets. These findings underscore the significance of color information in medical imaging and highlight the model's ability to generalize across different cross-validation folds, with 7-CV yielding the best results. The CAR model also outperforms existing state-of-the-art models while maintaining computational efficiency, making it an efficient approach for breast cancer diagnosis.