The paper introduces the research status of negative and positive classification tasks in clinical findings, proposes a system for clinical findings classification in medical dialogues based on the pre-trained model, elaborates the construction method of the system and analyzes the experimental results. The proposed method achieves 77.87% of Macro-F1 value in the 7th China health information processing conference (CHIP 2021) classifying positive and negative clinical findings in medical dialogues task.