Benchmarking Deep Learning Algorithms for Breast Cancer Detection: A Comprehensive Review and Evaluation Across Public Imaging Datasets

Breast cancer remains a leading cause of cancer-related mortality among women globally, emphasizing the need for early and accurate detection. This study combines a systematic review of artificial intelligence (AI) applications in breast cancer imaging with empirical benchmarking of deep learning models across public datasets. The review analyzed ۲۱ studies, highlighting convolutional neural networks (CNNs) as the dominant AI approach, with mammography being the most studied modality. Benchmarking involved evaluating a baseline CNN, ResNet-۵۰, and U-Net on datasets including DDSM (mammography), INbreast (mammography), and BUSI (ultrasound). Results demonstrated that ResNet-۵۰ significantly outperformed the baseline CNN in classification tasks, with a mean AUC improvement of ۰.۰۷۳ (p = ۰.۰۲۳). U-Net achieved robust segmentation performance on BUSI, particularly for malignant lesions (Dice coefficient = ۰.۸۷). The study underscores the superiority of transfer learning and deeper architectures in breast cancer imaging while identifying gaps in multimodal integration and explainability. Future directions include expanding to multimodal datasets, incorporating interpretability tools, and validating models in real-world settings. These findings contribute to the growing body of evidence supporting AI's role in enhancing breast cancer diagnostics and pave the way for clinically actionable solutions.

Keywords:

Breast Cancer Detection , Artificial intelligence , Deep Learning , Medical Imaging , Benchmarking