Research On Clinical Decision Support Methods Based On Collaborative Filtering And Deep Learning

The Outline of the "Healthy China 2030" clearly stated that health is an inevitable requirement for the promotion of man's all-round development.Promoting health science and technology innovation and improving the level of medical and health services have become national development strategies.Clinical decision support systems(CDSS)can effectively reduce the misdiagnosis and missed diagnosis rate of doctors.It has been a research hotspot in the field.In recent years,deep learning and collaborative filtering technologies have developed rapidly and are increasingly applied to different fields.Based on collaborative filtering and deep learning techniques,this paper had conducted in-depth studies on clinical diseaseassisted diagnosis and clinical high-risk disease prediction methods in CDSS.For the diagnosis of clinical disease,this article deeply analyzed the frequent occurrence of clinical disease information and its challenges to clinical decisionmaking.Based on this,two unsupervised learning methods based on user-based collaborative filtering,collaborative filtering based on Restricted Boltzmann machine(RBM),and identification-restricted Boltzmann machine were proposed.Discriminative Restricted Boltzmann machine(DRBM)supervised learning methods to support disease-assisted diagnosis.Using the UCI(University of California Irvine)database of dermatological datasets and chronic nephritis datasets,three missing datasets of varying degrees of deletion were generated by random methods for experimental training and validation.The experimental results showed that the overall performance of the DRBM-based method is the best.When the missing degree of the two data sets was 30%,the classification accuracy was still over 90%.For the prediction of high-risk clinical conditions,this article studied in depth the problem of epileptic seizure prediction based on the e Eelectroncephalogram(EEG)dataset,and conducted an experimental analysis on the probability of pre-seizure seizures.This paper proposed a method for transforming time domain EEG data into frequency domain information through discrete Fourier transform and extracting frequency domain and time series data features based on two-layer convolutional neural network(CNN)and AUC was 0.79.In addition,comparative experiments and systematic analysis of Linear Discriminant Analysis(LDA),Logistic Regression(LR)and Basic Recurrent Neural Network(RNN)were also carried out.The experimental results showed that the prediction performance based on CNN model was better than other models and the prediction probability value was more accurate.