Scholarlink

<p>Breast cancer (BC) is a serious global health concern, impacting around 2.1 million people each year. Because no permanent therapy is currently available, early and precise diagnosis is critical to lowering mortality. Advances in computer-aided detection (CAD) systems and deep learning (DL) models have greatly enhanced the ability to detect and classify breast cancer from photographs. This review examines state-of-the-art DL techniques for BC detection across different publically available datasets, including BreakHis, ICIAR 2018, CBIS-DDSM, MIAS, BUSI, and DDSM. Model efficacy is evaluated and compared using a variety of performance indicators, including accuracy, precision, recall, and the F1-score. Based on these metrics, the evaluation determines top-performing models for each dataset: Hybrid CNN-LSTM (99.03% on BreakHis), DOLL with E-SVM (97.70% on ICIAR 2018, IMPA-ResNet50 (98.32% on CBIS-DDSM), VGG16 (98.96% on MIAS), InceptionV3 (99.75% on BUSI), and OMLTS-DLCN (99.75% on DDSM). The findings highlight the importance of deep learning in BC diagnoses and provide researchers with insights into the most successful models and datasets to investigate further.</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

Advances in Deep Learning Techniques for Breast Cancer Classification: A Comprehensive Review

  • 期刊论文

  • 发表日期：

    2025年6月28日

查看全文
  • Kanika Kansal,
  • Sushil Kumar,
  • Kajal Kansal

摘要

Breast cancer (BC) is a serious global health concern, impacting around 2.1 million people each year. Because no permanent therapy is currently available, early and precise diagnosis is critical to lowering mortality. Advances in computer-aided detection (CAD) systems and deep learning (DL) models have greatly enhanced the ability to detect and classify breast cancer from photographs. This review examines state-of-the-art DL techniques for BC detection across different publically available datasets, including BreakHis, ICIAR 2018, CBIS-DDSM, MIAS, BUSI, and DDSM. Model efficacy is evaluated and compared using a variety of performance indicators, including accuracy, precision, recall, and the F1-score. Based on these metrics, the evaluation determines top-performing models for each dataset: Hybrid CNN-LSTM (99.03% on BreakHis), DOLL with E-SVM (97.70% on ICIAR 2018, IMPA-ResNet50 (98.32% on CBIS-DDSM), VGG16 (98.96% on MIAS), InceptionV3 (99.75% on BUSI), and OMLTS-DLCN (99.75% on DDSM). The findings highlight the importance of deep learning in BC diagnoses and provide researchers with insights into the most successful models and datasets to investigate further.