人工智能驱动下的中医智能诊疗研究进展与挑战OA

Objective Focusing on the"six-step procedure for diagnosis and treatment in traditional Chinese medicine(TCM)",to systematically review the key artificial intelligence(AI)paradigms(including supervised learning,unsupervised learning,reinforcement learning,and deep learning)in intelligent TCM diagnosis and treatment,summarize representative progress and application boundaries,and outline practical technological routes for deployment.Methods A search and synthesis of recent high-quality literature was conducted.Based on the full TCM diagnostic and treatment process of"inspection,listening/smelling,inquiry,pulse-taking,syndrome differentiation,prescription,and outcome prediction",a comparative analysis was performed on AI algorithm categories,data resources,and evaluation strategies.Additionally,a"problem-method"mapping was constructed from the perspectives of data standardization,interpretability,and privacy/security.Results ①Research on the objectification of TCM's four diagnostic methods evolved from"feature engineering+classical machine learning"to a generator-discriminator synergy with"Transformer algorithms/diffusion models+multimodal fusion".②Generative adversarial networks(GANs)and large language models(LLMs)significantly improved the ability to identify TCM syndromes and recommend individualized prescriptions,with verifiable gains in related clinical diagnosis and treatment scenarios.③Reinforcement learning demonstrated potential in the clinical"dynamic prescription adjustment-efficacy feedback"loop in TCM yet it was constrained by high-dimensional heterogeneous states,sparse/delayed rewards,offline bias,and safety exploration.④Actionable routes for AI research in TCM were proposed,including cross-modal alignment and shared representation,knowledge graph-enhanced explainable modeling,federated learning with differential privacy,and digital-twin-driven safe-reinforcement learning for virtual-to-real training and validation.Conclusions AI is reshaping the TCM intelligent diagnosis and treatment workflow and the logic of syndrome differentiation and treatment,but scalable deployment hinges on data standards and trustworthy,safe mechanisms.Grounded in the"six-step procedure for diagnosis and treatment in TCM"framework,multimodal integration and LLM alignment can drive a transition from experience-driven to data-and model-driven TCM,enabling precise syndrome differentiation,personalized prescriptions,and traceable decision-making.