The Study Of Classification Algorithm Of The Type Ⅰ And Ⅱ Diabetes Based On CGM

Diabetes a chronic disease is harming human health seriously.According to the etiology and pathogenesis of diabetes,it has divided into four categories,the type of 1,2,specific and gestational diabetes by the Expert Advisory Committee of WHO.At present,classification the type of diabetes mainly depends on the clinic test results and symptom of patient.However,because the identification of diabetes is a process of long-term development,it’s difficult for doctor to diagnose and classify some part of the type 1 and type 2 diabetes usually.Aim to classify the type of 1 and 2 diabetes;a novel indicator of classification the type of diabetes is proposed in this paper based on the glucose features extracted from the time series of glucose data,using the glucose data collected by CGM.What’s more,a novel algorithm of classifying the type of 1 and 2 diabetes is put forward.The main work of this paper includes below:(1)Analyzing and studying the characteristics of the time series of glucose data.An extraction feature model that can extract 17 statistical features including average value of blood glucose,area under the curve of glucose concentration,mean amplitude of plasma glucose excursions and so on from curves of glucose data collected by CGM is designed,finally.(2)A variety of classification algorithms are studied,and an algorithm of classifying the type of 1 and 2 diabetes base on the algorithm of double-Class AdaBoost is proposed.In order to improve the coincidence rate between the classification and clinical diagnosis,firstly,the thresholds of extraction conditions is set at 7mmol/L,8mmol/L,9mmol/L,10mmol/L and 11mmol/L separately in the stage of feature extraction based on classification guidelines of diabetes by WHO.According to the characteristics of collected diabetes data and then the AdaBoost algorithm,an improve AdaBoost algorithm is proposed.(3)1050 curves of diabetics’ glucose data are used to verify the classification indicator in our experiment.The experimental results indicate that the coincidence rate of our novel indicator of classification the type of diabetes and clinical diagnosis is over 90%.The novel indicator expands the standard of diagnosing the types of 1 and 2 diabetes and provides doctors with a new reference to classify diabetes.