Application Of Bayesian Network In The Related Factors And Classification Of Hepatic Encephalopathy Complication Of Hepatic Cirrhosis

Objective:By establishing a Bayesian Network model for the factors that affect Hepatic Cirrhosis' complication of Hepatic Encephalopathy(HE),this paper tries to describe the relationships among Hepatic Encephalopathy and those possible factors.Along with the inference of Bayesian Network,this paper talks about single factor's or multiple factors' role in contributions to the HE.Finally,this article tries to construct HE classification model to explore the classification efficiency of Bayesian network for hepatic encephalopathy identification.Methods:By investigating the medical record data of cirrhosis patients during January,2006 to December,2015,this paper gets intact check data of 950 hepatocirrhosis patients.It constructs a Bayesian Network topology through single factor and multiple factors logistic regression analysis to filter Hepatic Encephalopathy's associating factors.Based on the result that comes from the single factor logistic regression analysis,this paper uses Tabu Search(TS)algorithm to construct General Bayesian Network(GBN),and explores GBN to identify the classification efficiency in the study of hepatic encephalopathy.At the same time,a comparison is conducted among the Naive Bayesian(NB)classifier,Tree Augment Naive Bayesian(TAN)classifier and Logistic probability prediction model in their recognizing ability.Results:1.After the single factor's and multiple factors' logistic regression analysis,it comes out that only seven factors as Electrolyte imbalance,infections,hepatorenal syndrome,depression,diabetes,liver origin prothrombin time,total bilirubin are closely included in the model of regression.The former three are the main risks of HE with coefficients as 6.861,3.467 and 3.021.The last four factors' risk coefficients range from 2.1 to 2.7.2.The network contains 8 nodes and 10 arrows among the factors of hepatic encephalopathy.The area under the ROC curve is 0.843,meaning that the evaluation of network effect is better.Nodes representing hepatorenal syndrome,electrolyte imbalance,infections,depression,total bilirubin,time of prothrombin and diabetes with liver origin contact with hepatic encephalopathy through complex relationship.The diabetes get indirectly association with hepatic encephalopathy through total bilirubin.Hepatorenal syndrome can be in contact with the electrolyte disorder or infection,indirectly affect HE.Bayesian network inferences that,infection,electrolyte disorder and hepatorenal syndrome have a more closer relationship with hepatic encephalopathy.3.This paper study the 950 cases about classification and recognition.From the the aspect of comprehensive evaluation index such as F-means and G-means,the General Bayesian Networks(GBN)constructed through tabu search algorithm can correctly reflect the performance of the minority class and it is better than other models(F-measure of 0.410);For minority or large amount samples,the network's classification performance is better than other models(G-means to 0.739).At this point,the probability of the model in Youden index's biggest cut-off point is 0.10.After 1:2 sampling,GBN's classification of the minority samples performance is still better than other models,compared with 950 cases of data sets.At the same time,the F-measure value doubled(F-measure of 0.820);For minority or large amount samples,it's classification performance is not as good as TAN Bayesian classifier(G-means to 0.754).Compared with 950 cases of data sets,GBN also increased by 2.0%,implying that the data sets is imbalance.As for For minority or large amount samples,the classification is improved slightly.Under the condition of the maximum index,the probability cutoff point is increased from 0.10 to 0.374,which is more reasonable when recognizing HE.Conclusion:The Bayesian Network constructed based on Tabu Search algorithm has a high specific sensitivity degree.It can reflect the relationship between all nodes,and can also reveal the relationship between related factors as roles of hepatic encephalopathy.Bayesian Network can reasoning the patients' situations along with HE according to the doctors' acknowledge order of patients' information,conforming to the sequential process of clinical diagnosis and treatment.Due to the limited data,This paper failed to include all factors affecting HE.That is,this study is a preliminary evaluation criterion of the model efficiency.Extrapolation and application of the model have certain limitation,and nedd further validation and improvement.The constructed HE classification model for screening HE may have some guiding significance.However,when it is used in clinical cirrhosis with HE classified forecast,further research and study are necessary.