A Hybrid SwinTConvNeXt With Learnable Dynamic Gating Fusion for Breast Cancer Histopathology Image Classification

ABSTRACT Breast cancer remains a leading cause of cancer‐related deaths among women worldwide, with early and accurate diagnosis being critical to improving survival rates. While deep learning has revolutionized medical image classification, current models often face significant challenges in balancing intricate local features with global features. This study presents a hybrid multi‐class classification framework using the Swin Transformer and ConvNeXt model. The proposed SwinTConvNeXt‐LDGF model dynamically fuses local‐global features within a learnable dynamic gating network and classifies pathological images into different categories. The model was evaluated on an eight‐class histopathology BreakHis dataset, achieving a test accuracy of 95.53%, a precision of 94.74%, a recall of 96.32%, and an F1 score of 95.47%. These results demonstrate the effectiveness of combining the Swin Transformer and ConvNeXt backbones within a unified, learnable dynamic training framework. The proposed approach emphasizes the strong potential of SwinTConvNeXt‐LDGF to support pathologists in the real‐world classification of breast cancer subtypes.