Optimization And Application Of Atrial Fibrillation Detection Model For Wearable Ecg

With the accelerating aging of the population,the incidence of cardiovascular disease in China has increased year by year,and the mortality rate ranks first among various diseases.As a clinically common persistent arrhythmia disease,atrial fibrillation(AF)can increase the risk of stroke,heart failure,coronary artery and so on,with high morbidity and mortality.Therefore,it is very important to improve the clinical prevention rate,diagnosis rate and treatment rate of AF.In this paper,clinical data were collected by wearable electrocardiogram(ECG)devices,and AF intelligent detection model was optimized with the help of feature selection and machine learning technology,in order to realize early warning of AF,and reduce the medical burden and economic pressure.Moreover,it can save valuable medical resources.In this paper,the AF intelligent detection algorithm is researched and analyzed from the three dimensions of data,features,and algorithms.First,unconstrained feature selection algorithm is used to optimize the feature set,and combines three machine learning algorithms to build classification models,and then the classification results of all models on multiple databases are compared and analyzed when different feature sets are selected.In addition,three AF detection models are established with selected features for premature beat and AF considering the impact of premature beat in clinical signals on the accuracy of AF detection algorithm,with the expectation of developing a clinically effective AF diagnosis algorithm.The main research contents are as follows:(1)Clinical collection and labeling of the wearable ECG data.The three-lead wearable ECG recorder jointly developed by Southeast University and Lenovo,is applied to collect longterm ECG data from the elderly over 65 years old in Jiangdu People’s Hospital.Then,a longterm ECG AF labeling system is designed with Matlab GUI,to label and screen AF from all the collected data.Finally,the ECG report is generated for each patient with the analysis of the AF detection algorithm.(2)Calculation,unconstrained optimization and constrained selection of AF feature set.102 features commonly used in short-term AF detection are introduced,and they are divided into time-domain,frequency-domain,nonlinear and morphological.Then,unconstrained feature optimization and constrained feature selection are completed respectively for AF and Non-AF,AF and premature beat data.Finally,three different feature sets are obtained to build subsequent classification models.In addition,two features used for screening AF rapidly in long-term ECG signal are introduced.(3)Research on AF detection models under unconstrained and constrained optimization for premature beat.Each feature set is combined with support vector machine,logistic regression and random forest,to build three classification models,and the performance of these models is tested on three public databases and the wearable ECG data Ⅰ.Finally,the accuracy of the three unconstrained AF detection models reaches more than 90% on the four test sets,and random forest algorithm has stronger stability and generalization ability than the other two algorithms.In addition,compared with the unconstrained support vector machine model,the constrained support vector machine model has better performance and can effectively reduce the probability of false detection of AF,so it is more suitable for clinical applications.(4)Clinical validity verification of this study.AF screening rapidly on the wearable ECG data Ⅱ is completed in combination with the labeling system and the features used for screening AF in long-term ECG.Then the constrained support vector machine model is applied to detect AF precisely on the screened ECG signal.Afterwards,the duration of AF onset is counted and compared with the results of manual labeling.Finally,the AF detection rate reaches 96.96%,which proves that this study has certain clinical application and reference value.In this paper,the AF detection model under constraint for premature beat is constructed by designing the optimization and screening algorithm of feature set based on the clinical needs,and the model is applied to the screening of paroxysmal AF and the statistics of AF burden in the elderly over 65 years old,so as to provide some reference for the management and treatment clinically of AF.