Research and Application of Dialogue Diagnosis and Treatment System Based on Large Language Model

The design, research and practical application of conversational diagnosis and treatment system based on Large Language Model (LLM) has potential value in improving outpatient efficiency and reducing medical error rate compared with traditional diagnosis and treatment scenarios. First, the design of the system is based on BERT's (Bidirectional Encoder Representations from Transformers) architectural idea. GPT-4 (Pre-Trained) model was introduced to generate and expand contexts, and a conversational diagnosis and treatment system based on a large language model was built. The system is pre-trained on a large-scale medical knowledge base, mimics the natural language communication between doctors and patients to carry out in-depth dialogue and interaction, accurately capture the symptoms and health information described by patients, and provide more effective diagnostic suggestions combined with the knowledge of diagnosis and treatment field. Secondly, the doctor-patient dialogue intention recognition algorithm is used to assist the model to capture and analyze the patient's emotional expression, medication history and pathological feedback to ensure the accurate transmission and processing of patient information. Finally, when tested in a Tier 3 hospital, the system showed significant advantages over the traditional single-mode dialogue system in three aspects: automatic generation of medical record reports, real-time response to patient inquiries, and simulated doctor-patient communication. After more than 1000 doctor-patient conversation tests, the system accurately identified the patient's intention ratio reached 95%, and the accuracy of diagnostic recommendations obtained a high evaluation of 93%. It is expected to become an important auxiliary tool for the medical industry in the future.