Arrhythmia Detection Based On ECG Signal

Cardiovascular and cerebrovascular diseases,with are frequently led by cardiac arrhythmia,are extremely dangerous to human's health.The famous Framingham study proves that diseases such as stroke are closely related to arrhythmia symptoms such as atrial fibrillation.Automatic analysis technology for ECG signals makes cardiac arrhythmia timely and accurately detected,which has great significance for preventing cardiovascular diseases.Based on the investigation into related research in arrhythmia detection,this thesis focus on a series of methods such as signal preprocessing technology,feature extraction method and classification model,combined with wavelet analysis,support vector machine,deep residual network and stacking model ensemble.New methods for ECG signal denoising and classification model are proposed and validated by different types of ECG dataset.In addition,the pathological correlation between arrhythmia and cardiovascular diseases is explored by clustering method.The main point of this thesis is as follows:1.As the low signal-to-noise ratio and complex noise components of ECG signals,the wavelet transform technique is used to decompose the signals by multi-scale wavelet,which allows effective signal and noise signal separated in the wavelet domain.Experiments show that the proposed method can effectively improve the signal-to-noise ratio of the original signal and improve the performance of the classification model.2.A wavelet transform method combined with dynamic threshold analysis is proposed to detect QRS complexes and extract samples such as conduction block and premature beats.This method has lower false detection rate and missed detection rate than the traditional Pan & Tompkins algorithm.Then,a sliding window sampling method is proposed for samples of atrial fibrillation type.Compared with the traditional segmentation method,the sliding window can keep the features of signals more effectively and significantly increase the number of the extracted samples.3.Two classification model,which are SVM and Res Net,are applied to detection of various types of ECG signal.Four feature extraction methods are designed and particle swarm optimization algorithm is used for SVM model,while a 50-layer convolutional neural network with residual block is built for 1-D ECG signal input,called Res Net50-ECG.Stacking algorithm is used for ensemble modeling with SVM and Res Net,which is proved to significantly improve the precision of arrhythmia detection.4.Based on the detection result of ensemble model for the atrial fibrillation amples,a series of features were extracted in terms of the time and duration of atrial fibrillation,and different cases of atrial fibrillation are distinguished with cluster analysis.The preliminary exploration of the pathological correlation between atrial fibrillation and cardiovascular diseases,such as stroke,has laid a foundation for further research of this thesis.At the end of this thesis,several related researches are selected to compare with the methods and results showed in this thesis,which proved that the methods and models proposed by this thesis can effectively classify the ECG signal and has better ability for arrhythmia detection.