The invention relates to a physical engine and data dual-driven trusted medical AI decision-making method, which comprises the following steps: S1, acquiring multi-
source data containing doctor questions and patient
medical information, and carrying out integrated analysis on the multi-
source data based on an external authoritative
knowledge base to obtain a preliminary treatment
hypothesis; s2, calling a physical engine, and constructing a digital twinborn model based on individualized data in the patient
medical information; based on the preliminary treatment
hypothesis, simulating the digital twinborn model according to a
physical law, and generating a feedback result; and S3, if the feedback result does not reach the ideal effect, updating the preliminary treatment
hypothesis, and returning to S2 until the feedback result reaches the ideal effect, and taking the corresponding preliminary treatment hypothesis as a treatment scheme. According to the method, through a brand new data-
physics driven normal form, a systematic and extensible solution is provided for safety landing of a
large model in the medical high-risk field, and the method has important practical value in a high-risk clinical AI application scene.