A Robust CNN Approach for BI-RADS Based Breast Cancer Detection

Breast cancer remains one of the most prevalent cancers among women and a major cause of cancer-related mortality worldwide. Early and accurate detection is essential for reducing mortality rates, and mammography remains the most effective screening tool. This study proposes a convolutional neural network (CNN) framework for BI-RADS-based breast cancer classification using three publicly available datasets: CBIS-DDSM, INbreast, and KAU-BCMD. A comprehensive preprocessing pipeline, including noise reduction, contrast enhancement, and region-of-interest extraction, was applied, followed by data augmentation to improve generalization. The model was trained and optimized through grid search across multiple hyperparameter settings. The best configuration, with a learning rate of 0.001 and batch size of 32, achieved 92.28% test accuracy, with precision of 99.1% for BI-RADS4-5 cases and recall of 99.5% for BI-RADS 1 cases. These results demonstrate the potential of a custom CNN with robust preprocessing for BI-RADS based detection of breast cancer and highlight its clinical applicability for improving breast cancer detection.