Breast cancer continues to be one of the leading causes of cancer death among women. Mammogram is the standard of care for screening and diagnosis of breast cancer. The American College of Radiology developed the Breast Imaging Reporting and Data System (BI-RADS) lexicon to standardize mammographic reporting to assess cancer risk and facilitate biopsy decision-making. However, because substantial inter-observer variability remains in the application of the BI-RADS lexicon, including inappropriate term usage and missing data, current biopsy decision-making accuracy using the unstructured free text or semi-structured reports varies greatly. Hence, incorporating novel and accurate technique into breast cancer decision-making data is critical. Here, we combined natural language processing and deep learning methods to develop an analytic model that targets well-characterized and defined specific breast suspicious patient subgroups rather than a broad heterogeneous group for diagnostic support of breast cancer management.