ABSTRACT Breast cancer remains one of the most common malignancies in women, where early diagnosis is critical for effective treatment. While deep learning shows significant promise for histopathological image analysis, its performance is often hindered by two key challenges: the scarcity and imbalance of medical datasets, and the inherent difficulty of classifying images with large intraclass and small interclass variations. To address these issues, we propose a novel Center‐Border Dual‐Branch Network with Dynamic Weighted Fusion (CBDF‐Net) for breast cancer histopathology image classification. Our framework first counteracts data imbalance through a pathology‐informed augmentation pipeline (HistoAugment), then employs a Center‐Border Partition Module (CBPM) for spatially differentiated feature extraction. These distinct representations are subsequently processed by a dynamically fused dual‐branch network (DFD‐Net), and its outputs are then integrated by a dynamic fusion mechanism with a compensation shortcut. The entire architecture is trained under the supervision of our novel Adaptive Entropy Penalty (AEP) Loss, which mitigates overconfidence and enhances generalization. Extensive experiments on multiple public datasets demonstrate that CBDF‐Net consistently and significantly outperforms state‐of‐the‐art methods.
Zeng et al. (Sun,) studied this question.