Application Of Bayesian Networks In Diabetes Mellitus Aided Diagnosis System

Digital medical technology is the development trend of medical information technology. Using of disease diagnosis system, the basic health units can share large hospital resources and experience, improving the accuracy of diagnosis of diseases, and can provide better medical service for the people. Diabetes mellitus (DM) is the world’s one of the highest incidence of lifestyle diseases, the incidence of diabetes and its complications is increasing year by year in China. If a diabetes diagnosis system can be developed to assist doctors in all levels of hospitals with diagnosis and prediction of diabetic complications by using the digital medical technology, the diagnostic accuracy of early complications symptoms will be improved, and the risk of disease will be reduced.The application of Bayesian network in diabetes aided diagnosis system is studied in the paper, and a diabetes-aided diagnosis system is designed and implemented. Firstly, the hospital original data were collected, and a diabetic patient sample database was established by a series of data preprocessing operations (data integration, feature selection, discretization, feature reduction and missing data processing); Then the NB, TAN and ATAN models were applied to the diabetes-aided diagnosis system, and Bayesian networks were established for predicting diabetic complications (macrovascular disease, microvascular disease, neuropathy and diabetic foot); Finally, the Bayesian network inference algorithms of NB, TAN and ATAN models were implemented. Doctors can obtain diagnosis results by using the algorithms to analyzing patient’s information. Based on the TAN model, a new improved TAN Bayesian network model named ATAN (attributes augmented tree augmented Bayesian network model) was presented in the paper. The ATAN able to make better use of expert knowledge, more truly reflect the dependencies between the characteristics, and thus having a higher rate of diagnostic accuracy. The test results (obtained from an open data mining platform, Weka) show that in most cases, ATAN has higher classification accuracy than TAN. The test results of our diabetes-aided diagnosis system show that the system can achieve an average diagnostic accuracy rate of more than60%, and can be applied in practice.