Purpose/Significance To extract the characteristic information of breast ultrasound and mammography reports, and to construct a knowledge graph for breast cancer, so as to promote the intelligent management of breast cancer diagnosis and treatment data, and lay a foundation for the development of clinical decision support system (CDSS) for breast cancer. Method/Process By combining breast cancer diagnosis and treatment guidelines with clinical expert experience, the breast cancer knowledge graph ontology and conceptual layer are constructed. Then, the entity and relationship joint extraction is performed resorting to a RoBERTa model and Global Pointer so as to form the data layer. Finally, Neo4j is adopted to store the constructed knowledge graph. A variety of traditional serial entity relation extraction models are selected for comparative experiments to evaluate the performance of the models. Result/Conclusion The accuracy rate, the recall rate, and the F1 value of the model based on the breast cancer knowledge graph exceed 95%, outperforming the comparison models.