Research On The Computer-aided Diagnosis Of Diabetes Based On Clinical Data

Diabetes is a common metabolic disease characterized by hyperglycemia.The cause is impaired insulin secretion or its biological effects.Diabetes mellitus can cause eye,kidney,cardiovascular and nervous physiology damage in the long term.It has many complications,and all of them do great harm to the patients.With the advance of medical information technology,the medical industry is eager to get information from these massive and complex data to help the development of the medical industry,also,the medical industry is eager to alleviate the tense situation of medical personnel while they obtaining a large numbers of the medical data.So the development of medical wisdom is imminent.The use of machine learning and other artificial intelligence methods of medical data analysis and treatment to achieve the computer-aided diagnosis is one of the very practical significant performances of the medical wisdom.Biochemical indicators of diabetes mellitus include glycated hemoglobin(Hb A1c),fasting blood glucose(Glu)and insulin release test(Ins).Glycosylated hemoglobin(Hb A1c)is the indicators of red blood cells in the human blood hemoglobin and blood glucose products,usually can reflect patients with nearly 8 to 12 weeks of blood glucose control;fasting blood glucose(Glu)is the indicators that can represent the basal insulin secretion;and insulin release test(Ins)reflects the reserve function of pancreatic beta cells.The main indicators of the diabetes have a lot of usage such as adjustment of treatment regimen,evaluation of patient status,and measurement of treatment outcome.The trend of indicators’ change is affected by the basic characteristics of patients and also there is a clear correlation between different indicators.Therefore,the index prediction of diabetes can be seen as a part of the computer-aided diagnosis of diabetes.In this paper,we constructed a predictive model of diabetes biochemical index based on neural network.This model is based on BP neural network,and adds a buffer layer of last calculation result in the hidden layer,which can be used to fit the time series of diabetes index data.The main work of this paper is as follows:The data used in the article are real medical data obtained from the hospital,in order to ensure that the next analysis can be built on a good structured data,the article first carried out the data pretreatment and data cleaning.After clarifying the goal of predicting diabetes mellitus index,the article defined the data source of the follow-up analysis according to the data source,the analysis target and the crowd.In order to understand the characteristics of the data and some basic information implied in the data,the article makes a multidimensional analysis,the analysis obtains some data characteristics and based on the result,we build a data model.After analyzing the characteristics of the data model,the BP neural network is selected as the forecasting model,and the data model is improved to meet the irregular timing characteristics of the data model and the complex interrelationships within the data model.,Thus can obtain the better forecast result to assist the doctor to determine the treatment plan,thus achieves the auxiliary diagnosis and treatment goal.Then,in order to verify the model proposed in this paper have a better performance,the study also joined the contrast experiment.The experimental results show that the proposed model is more effective than the BP neural network and XGBoost algorithm in predicting the tendency of biochemical indexes of diabetes.Finally,we designed a system to carry out the diagnosis and treatment of diabetes based on clinical data through the system.