| Summary: | Breast cancer remains one of the leading causes of death among women worldwide, highlighting the need for early and accurate detection. Recent advancements in AI-driven techniques, particularly Machine Learning (ML) learning and Deep Learning (DL), have significantly improved breast cancer diagnostics in breast cancer recognition. However, the intraclass variability, which is the subtle difference between malignant and benign within the same class, is a major challenge that leads to misclassification, misdiagnosis, reduced model effectiveness, increased health costs, and challenges clinical decision-making. This review provides a comprehensive analysis of ML and DL-based techniques, with a particular focus on addressing intra-class variance in breast cancer imaging. We reviewed articles from the past five years, examining publicly available datasets, the limitations and advantages of various ML and DL techniques, performance metrics, and the clinical applicability of multiple approaches. Furthermore, we also identified dataset challenges, including solutions to imbalanced datasets, particularly focusing on GAN-based data augmentation. We synthesized the current trends and highlighted the future directions. This review aims to support researchers and professionals in developing more robust and interpretable AI-driven breast diagnostic systems.
|
|---|