A Research On The Prediction Model Of Diabetic Complications Based On Bayesian Network

Type 2 diabetes mellitus(T2DM)is one kind of metabolic disorder disease,which leads to the increased costs of noncommunicable diseases with increasing incidence in recent years.Furthermore,T2DM complications seriously affect the quality of life.Due to the limited social resources,and the fact that T2DM complications could not be cured completely,it is important for patients to take actions for prevention and self-management.The related studies explained that the prediction result with warning factors was beneficial for patients.Most of the studies generally used the factors by professional medical technology and instruments,for example,serum endostatin levels by enzyme-linked immunosorbent assay and erythrocyte sedimentation rate(ESR),elongation index at 3 Pa(EI)measured using microfluidic hemorheometer,or clinical factors by machine learning method,such as age,white blood cell,total cholesterol and some like these selected by random forest,support vector machine and feature selection method.However,the indicators were inconvenient and impractical for the patients or the models were not able to deal with the disease prediction based on missing data,it is not conducive to the self-prevention management of the patients.Bayesian network(BN)is a combination of graph theory and probability theory,based on which BN models are generally used to solve problems in complex areas and are capable to deal with incomplete data.Therefore,this thesis used warning factors to predict T2DM complications based on BN from the perspective of self-prevention.Then prediction models were analyzed and compared to validate the model built by this thesis so as to prove the helpfulness of the model for patients to make self-prevention.There are two parts in BN learning that structural learning and parameter learning.Parameter learning is based on the results of structure learning,on which analysis of warning factors is also based,so the structural learning part is particularly important.In order to obtain the more accurate warning factors which are used to predict T2DM complications so as to achieve the purpose of self-prevention in patients with T2DM,this thesis introduced the improved meta-heuristic algorithm called artificial bee colony algorithm(ABC)and Gibbs Sampling algorithm into the structural learning of BN to obtain a more accurate BN which is also able to learn on missing data.Then,T2DM complications were predicted using warning factors based on Markov Blanket(MB).The result demonstrated that the model built in this thesis is effective in predicting the other four complications except diabetic peripheral neuropathy and diabetic retinopathy.The main innovation points are as follow:(?)In view of the T2DM complications prediction application scenario,the issue was translated.According to the idea of classification,the discretized Lévy flight algorithm and Differential Evolution algorithm were integrated into ABC algorithm.Meanwhile,Gibbs Sampling algorithm was used to update the missing data so that structural learning and parameter learning could be done.As a result,a bayesian network structure learning algorithm based on an improved ABC algorithm is proposed.The result showed that the precision and running time were both improved.(?)According to the expert knowledge and common sense of demographic characteristics,whitelist and blacklist were created,which were used to combine with the improved BN learning algorithm to avoid learning invalid network so that the efficiency of learning BN was improved and the T2DM complication prediction model was optimized.(?)Based on MB,warning factors were analyzed on the model.They were instantiated respectively and were used to deduce the probability of complications so as to be helpful with self-management and disease prediction of patients.Then the warning factors were used to predict complications based on the established BN model.The T2DM complication prediction model this thesis built has good effect for patients in T2DM to prevent and manage the occurrence of diabetic nephropathy,diabetic retinopathy,diabetic macrovascular and diabetic ketoacidosis,and can assist patients in the diagnosis of complications so as to go to the hospital in time.Meanwhile,it is beneficial for society to save hospital resources through self-prevention and self-management of patients without T2DM complications.