An Explainable Hybrid CNN–Transformer Framework for Breast Cancer Detection from Histopathology Images

Cancer is one of the most serious health challenges in the world and hence pathologists need accurate and clinically reliable diagnostic support. Even though deep learning algorithms have proven to be very effective in analyzing histopathology images, most of the models currently are not interpretable and prone to false negatives. In this paper, a hybrid CNN-Transformer model detect breast cancer automatically and it combines the local and global contextual modelling of features to capture the complex tissue patterns. The standard preprocessing and augmentation approaches are used to make experiments on large-scale histopathology image dataset robust. Accuracy, precision, recall, F1-score and AUC-ROC are used to evaluate the proposed model, but the main focus of the model is on recall and reduction of false negatives. Grad-CAM and transformer attention maps as explainable AI methods are used to visualize regions of diagnostic interest to enhance model transparency and clinical trust. The experimental outcomes prove that the hybrid framework is more efficient in comparison to independent CNN and transformer models which provide a sensitive, reliable and interpretable solution to AI-assisted histopathology based breast cancer detection.