Early diagnosis of breast cancer is critical for effective treatment. Artificial intelligence (AI) technology has shown promise in assisting physicians with diagnosis. However, the combination of qualitative and quantitative information in surveillance data leads to ambiguity and uncertainty. Belief rule bases (BRB) can address these issues by incorporating confidence distributions. However, existing BRB models rely on offline training and lack adaptability to changes in patient metrics. In addition, the ethical implications of using BRB for breast cancer diagnosis require attention to the interpretability of the model. Therefore, this paper presents an online belief rule base breast cancer diagnosis method with interpretability. The method uses online learning to achieve dynamic growth. It also overcomes the problem of interpretability loss in the optimization process by implementing interpretability constraints. The proposed method achieves competitive accuracy and interpretability in breast cancer diagnosis, as demonstrated by experiments using a large dataset of breast cancer cases.