Abstract
Breast cancer (BC), the most prevalent malignancy in the female population, often presents significant difficulties in early diagnosis and identification of molecular subtypes. In addition, due to the lack of obvious clinical symptoms in the early stage and the lack of effective early detection means or specific biomarkers, about 30% of the cases are already in the advanced stage at the time of diagnosis, which directly leads to the patients missing the best treatment period. Unfortunately, BC is also highly heterogeneous, and its different molecular typing directly affects the outcome of treatment regimens such as chemotherapy, immunotherapy, etc., and significantly correlates with patients' 5-year survival rates. Artificial intelligence (AI) has rapidly advanced from proof of concept to prospective and real-world deployments, delivering radiologist level accuracy, improved specificity, and substantial workload reduction (≈44%-68%) without compromising cancer detection. Some studies even report more cancers detected when AI supports readers. These gains translate into earlier diagnosis, fewer unnecessary recalls, and more efficient screening workflows. Concurrently, multi-modal AI (integrating mammography, ultrasound/DBT, MRI, digital pathology, and multi omics) enables robust subtype identification, immune tumor microenvironment quantification, and prediction of immunotherapy response and drug resistance, thereby supporting individualized treatment design and drug discovery. The aim of this review is to systematically illustrate the efficient application of AI technology in BC diagnosis, such as constructing early diagnostic models based on multi-omics data, identifying molecular subtypes of BC, quantifying the tumor immune microenvironment and predicting the immunotherapeutic response, as well as investigating drug resistance of BC and developing new therapeutic agents. In the future, AI technology will be able to provide more accurate individualized diagnosis and treatment for BC patients.