Research And Development Of The Diabetes Risk Identification System Based On Information Fusion

With the development of Internet technology and medical information technology,the mass data collected by medical institutions has great significance to the diagnosis of diseases.However,because of the broad-scale,source diversification of the data,it is difficult to analyze the data deeply to obtain potential knowledge.Focusing on the problems existed in the medical aided diagnosis systems,this study designed a diabetes risk identification system to improve the diagnostic efficiency and reduce the misdiagnosis by information fusion methods,and effectively assist the medical staff to make decisions.The main research of this study were as follows:(1)Based on the research of domestic and foreign relevant literature,this paper analyzed and summarized the problems existed in the medical aided diagnosis systems,and determined the research direction of the paper.(2)Aiming at the redundant of diabetes features,an improved firefly algorithm(RS-IFA)was proposed to reduce the diabetes features.Firstly,the initial population of firefly was optimized by rough set theory.Secondly,the location of firefly was associated with the global optimum firefly.The RS-IFA algorithm was compared with the basic firefly algorithm and the particle swarm optimization algorithm in experimental simulation test,and the performance of the Naive Bayes classifier was used to evaluate the feature subset.The result showed that the feature subset obtained by the RS-IFA algorithm has a significant improvement in the accuracy rate,and it solved the problem of slow convergence speed and improved the efficiency of the feature selection process.(3)Aiming at the diagnosis of diabetes,a Two-stage diabetes prediction model was established by theories and methods of information fusion.In the first stage,primary diabetes diagnosis was processed using BP neural network,and in the second stage,the outputs from feature level are fused by D-S evidence theory methods to get the final diagnosis result.The combination of BP neural network and D-S evidence theory can not only overcome the uncertainty of diagnosis and improve the accuracy of diabetes diagnosis,but also give full play to the self-learning ability of BP neural network.(4)Aiming at designing the diabetes risk identification system,the detailed design ofthe system was carried out.Experiments showed that the system achieved the expected goal,which realized the dynamic analysis and prediction of diabetes medical data.