Research On Medical Decisions Based On High-rank Matrix Completion

Medical decision analysis is a rising research direction based on traditional medical research combined with big data analysis.In the era of big data,the amount of information and data types in health care have increased rapidly.In response to such problems,medical decision analysis has emerged.Etiology analysis is an important composition of medical decision analysis.In this paper,we choose Autism Spectrum Disorders(ASD)as an example to study the impact of autistic factors on children's ASD condition.Autism has become an important and damaging pediatric mental illness,which is detrimental to the growth of children with ASD and causes heavy psychological pressure and financial burden on the families of those children.In order to solve the problem of data missing in the establishment of the relationship model of influence factors of ASD,we introduce matrix completion to fill in the missing data.According to some characteristics of autistic children's information data?many valuable information and high rank,we improve the existing low-rank matrix completion algorithm and propose two algorithms that can be applied to high-rank matrix completion.They are the high-rank matrix completion algorithm(HRMC)based on alternating direction multiplier algorithm(ADMM)and the matrix completion algorithm(AEMC)based on auto-encoder.The former combines the logic of the matrix completion and the fact that the importance of each variable in the data are different,so that the weight factors of the control variable can take different values according to the importance of variables themselves.By doing so,HRMC will devote more resources to the recovery of the important variable data,which improves the efficiency of the algorithm.The latter uses auto-encoder and simultaneously derives the derivative of the loss function to the matrix in the process of derivation of back propagation.After the problem of data missing is solved,this paper establishes an artificial neural network model studying the relationship between autistic influence factors and ASD.Firstly,the index measuring the severity of autism symptoms and the influence factors of ASD are determined.Then,a neural network model is established on the relationship between ABC score and those influence factors.By changing training algorithm,the proportion of training samples,number of nodes in hidden layer and other parameters,the accuracy of the neural network is improved.Finally,we conduct sensitivity analysis to the optimal network,and find several factors that are more susceptible to ASD,namely,the children's genes abnormity and the mothers disease history during pregnancy.In order to verify the validity of the matrix completion algorithms,and make clear whether matrix completion can improve the precision of models,a numerical experiment of matrix completion is carried out.Firstly,the test matrix is generated and used to test the performance of HRMC and AEMC,and then the two algorithms are applied to the filling process of the actual data of children with ASD.The filling results are compared with multiple imputation(MI)and singular value thresholding(SVT),and result shows that the accuracy of HRMC and AEMC are better than the existing algorithms.Finally,the data before and after matrix completion are input into the neural network model obtained before.The results show that appropriate data filling can improve the accuracy of the model effectively.