Research And Implementation Of Risk Prediction Algorithm For Diabetic Patients Based On Convolutional Neural Network

Diabetes is a common chronic disease,which will not only lead to the instability of blood glucose,but also leads to a series of complications.Diabetics spend most of their time on blood glucose management at home,but sometimes they need hospitalization to adjust their blood glucose levels.Diabetic inpatients may face unforeseen risks,such as high and low blood glucose risk and rehospitalization risk,but it is hard for the existing medical measures to predict these risks in advance.Therefore,this thesis analyzes and studies the hidden risks of diabetes inpatients.Based on the improved Convolutional Neural Network(CNN),an effective risk prediction model is put forward to assist doctors in diagnosing and evaluating patients,reducing the risk of high and low blood glucose and re-hospitalization risk.This thesis has done the following work:(1)A blood glucose prediction model based on Temporal Convolutional Neural Network(TCN)is proposed.The existing blood glucose risk prediction methods are difficult to mine the deep time characteristics of time series.In order to fully mine the time relationship in time series data and make efficient prediction,this thesis combines TCN and extreme gradient boosting(XGBoost)algorithm,and proposes a new time series prediction algorithm model.In this model,TCN combines the characteristics of convolution neural network and causal convolution,and can fully mine the time information in blood glucose historical data.At the same time,XGBoost can carry out fast and efficient regression prediction on this basis.The experimental results show that the model can effectively predict the blood glucose status of diabetic patients in the future,so as to assist doctors in making diagnosis and treatment plans for patients and reduce the risk of hyperglycemia or hypoglycemia.(2)A diabetic re-hospitalization prediction model based on Cost Sensitive Convolutional Neural Network(CSCNN)is proposed.For re-hospitalization,the medical data is often faced with problems of imbalance.Imbalanced data often leads to the classifier being biased toward the majority class,and thus,neglecting the minority class samples.The model proposed in this thesis improves the loss function of convolutional neural network using cost sensitive learning to solve the imbalance problem of medical data.The experimental results show that the model in this thesis can effectively classify and predict unbalanced readmission data,and the prediction results can assist physicians to evaluate the discharge time of patients,thus reducing the risk of re-hospitalization.(3)A risk prediction system for diabetes patients based on convolutional neural network(CNN)is designed and implemented.In order to combine data analysis technology with diabetes treatment,this thesis implements a risk prediction system based on the above glucose risk prediction model and re-hospitalization risk prediction model.The system collects hospitalization and physical examination information of patients through the information collection module,and trains the prediction model in the background with the data.The risk prediction system can not only allow patients to check their physical status,but also use the prediction model to predict the risk of blood glucose and rehospitalization.The prediction results can assist physicians to comprehensively understand the physical conditions of patients and make diagnostic evaluation to reduce the risk of blood glucose and re-hospitalization.