Research On Recommendation Method Of Treatment Regime Based On Graph Embedding And Reinforcement Learning

The application of artificial intelligence technology in the medical and health field is essential for improving the overall level of medical diagnosis and treatment and achieving intelligent and precise diagnosis and treatment.Prescription intelligent recommendation is a key element of TCM-assisted diagnosis and treatment.The early clinical system is mainly based on the experience and knowledge of experts,but different doctors’ diagnosis and treatment concepts and prescriptions are not the same,which leads to uneven clinical diagnosis and treatment levels.In recent years,the use of reinforcement learning technology to predict medical treatment regime is considered to be a feasible way of thinking.However,due to the complexity of clinical samples and high noise,designing a high-performance TCM prescription recommendation method based on reinforcement learning is still a challenge.In response to the above problems,we proposed a method for predicting diagnosis and treatment regime based on reinforcement learning,and achieved good predictive performance.The main work of the article can be summarized as follows:(1)Aiming at the complexity of the composition of TCM prescriptions,we propose a diagnosis and treatment regime prediction model that combines the core principles of TCM treatment and treatment based on a multi-classification method.In view of the large number of patients with symptoms,the severity and importance of the symptoms are different,and the expression of the patient’s state is subjective,this article builds a patient state network based on electronic medical records and uses the network embedding method to learn the patient’s state.The experimental results show that compared with the baseline method,the diagnosis and treatment prediction model we proposed can better retain the characteristics of the original symptom observation vector.(2)Aiming at the problem of lack of real environment in medical reinforcement learning research,we propose a simulation-based TCM diagnosis and treatment environment model,which is based on patient symptoms and diagnosis and treatment plans and other information to infer the possible symptoms and probabilities of the patient.Furthermore,a diagnosis and treatment plan prediction method based on state transition probability and dynamic programming is proposed.Experiments show that the result of this method on the weighted symptom score index is significantly better than doctors.(3)Aiming at the problem that it is difficult to learn state features in diagnosis and treatment plan prediction,this paper proposes a diagnosis and treatment plan prediction framework GS-DRQN that combines network embedding and deep reinforcement learning.The framework uses deep reinforcement learning to obtain the optimal path strategy to predict the diagnosis and treatment plan.The simulation experiment results show that,compared with the doctor’s diagnosis and treatment plan,this prediction framework can significantly improve the patient’s disease reduction,and thus obtain better prescription recommendation results.