Attention-Enhanced Residual Network for Breast Cancer Detection in Histopathological Images

One of the biggest causes of death worldwide is still cancer, with breast cancer now the most prevalent form, posing significant challenges to women’s health. Although pathological image analysis remains the most reliable diagnostic method, it is often slow and resource-demanding, especially in areas with limited resources. This study introduces a deep learning framework that combines ResNet101 with a Residual Convolutional Block Attention Module (CBAM), utilising channel-wise and spatial attention mechanisms to improve feature extraction. A robust preprocessing pipeline, including geometric transformations, CLAHE, and brightness/contrast adjustments, is employed to improve data diversity and model’s generalization ability. The model was assessed using the BreakHis and IDC datasets, delivering state-of-the-art (SOTA) performance and highlighting its effectiveness in breast cancer detection from histopathology images. The code for our proposed model is available at Github.