Automatic Detection Of ECG Signals Based On LIBSVM And GPU

With changes of lifestyle and acceleration of work pace,heart disease has become one of the most pathogenic ones in modern society.Early detection,diagnosis and treatment of heart disease are the keys to reduce heart disease mortality.Electrocardiogram(ECG),which reflects the cardiac activity,is an important basis for diagnosis and prevention of heart disease.However,traditional ECG diagnosis requires to be done by professional doctors.Due to large population and the increasing proportion of elderly population in China,automatic detections of ECG will better help people prevent heart disease and reduce burdens on doctors.How to utilize modern information technologies for automated ECG analysis and detection has become a hot topic in the field of biomedical engineering.This thesis proposes an automatic detection method of ECG based on support vector machine and GPU platform.The method improves ECG diagnostic efficiency in three aspects.Firstly,the wavelet transform is used to remove noises from ECG,including frequency interference,EMG interference and baseline drift.Secondly,LIBSVM toolkits are used for automatic classification of ECG signals.We extract wavelet coefficients as morphological features,and pro-RR interval,pre-RR interval and local RR interval as dynamic features,which are used as training features in LIBSVM.Thirdly,GPU-based parallel computing platform is used to accelerate the training process.At the same time,we optimize LIBSVM serial algorithm.As a result,the classification accuracy is over 98%and the model training process is accelerated by up to 18 times.