Research On Medical Self-diagnosis Based On Knowledge Graph And Deep Reinforcement Learning

With the available of electronic medical records(EMR),data-driven medical self-diagnosis methods have become main stream.In the current work,researchers train electronic medical records based on various machine learning models,and obtain better disease diagnosis results by adjusting the model structure.However,these methods only focus on whether the model performs well,ignoring the fact that patients cannot accurately provide the information required by the self-diagnostic model in one application.In real medical consultations,patients often complete a comprehensive description of their condition in several questions and answers with the doctor.To simulate this process,this paper builds an interactive intelligent diagnosis model.At the same time,the “black box” of machine learning limits its application in the medical field,and it is necessary to study the interpretable methods of medical diagnostic models.Therefore,this paper proposes an interpretable medical self-diagnosis system DKDR(Disease Knowledge System by dint of Knowledge graph and Deep Reinforcement learning)based on medical knowledge graph and deep reinforcement learning.The main contents of the system are:(1)Realize intelligent diagnosis of diseases: use machine learning classification algorithms to diagnose diseases.This paper compares a variety of classification algorithms,conducts experiments on the real medical datasets MIMIC-III and electronic medical record generator Synthea,choosing better performing multilayer perceptron(MLP)and convolutional neural network(CNN)as intelligent disease diagnosis algorithms.(2)Realize interactive intelligent diagnosis of diseases: construct medical knowledge graph to interact with users,and introduce Q-learning reinforcement learning algorithm in the intelligent diagnosis algorithm of diseases MLP/CNN to improve the accuracy of disease diagnosis.First,this paper builds a medical knowledge graph based on Scrapy crawler framework and Neo4 j graph database.The knowledge graph helps users to improve the description of the disease and achieve interaction with the user.Secondly,combining Q-learning algorithm and MLP/CNN algorithm to diagnose diseases,improve the accuracy of disease diagnosis,the best accuracy of disease diagnosis reaches 91%.(3)Realize the interpretability of disease intelligent diagnosis: build an explanatory model based on visual sensitivity analysis.Use scatter plots,heat maps,and radar charts to present the characteristics of electronic medical records,and observe the impact of different features on the disease.Through the analysis of this effect,re-diagnose the module experiment and observe the model performance.After analysis and improvement,the diagnostic accuracy of the model has increased by 3 percentage points on pneumonia datasets.Visual sensitivity analysis helps researchers and users to better understand the model and enhance people’s trust in the medical self-diagnosis system.