Abstract: The automatic identification and extraction of diagnosis and treatment relationships helps doctors make diagnosis and treatment decisions. The traditional relationship extraction model does not have good interpretability for part of the data. Therefore, this paper used neural network for rule learning and generalization, designed a scoring mechanism, and achieved relationship extraction through rule matching. A targeted deep learning model for incorrectly matched data training was proceeded to complete the final diagnosis and treatment relationship extraction. The relevant texts of the disease-centric diagnosis and treatment process were used for experiments to verify the effect of this method. The results show that the text method not only increases the interpretability of relationship extraction through a few manual rules, but also significantly improves the effect of relationship extraction.