Research On Automatic Detection Algorithm Of ECG Signal Feature Points

Cardiovascular disease is a serious threat to human life and health.In clinical practice,electrocardiography is often used as a non-invasive tool for the diagnosis of cardiovascular diseases.Electrocardiography can record the electrical activity process of myocardial cells from the body surface.Doctors can diagnose various types of heart diseases by analyzing the electrocardiogram of patients.However,manual analysis of electrocardiograms is not only time-consuming and labor-intensive,but also requires strong professional knowledge,especially the large number of electrocardiograms generated by the promotion of wearable devices has exacerbated this dilemma.At present,the automatic analysis of ECG signal has become an important means of cardiovascular disease detection.As the premise of automatic analysis of ECG signal,the detection of characteristic points of ECG signal has always been a research hotspot in the field of biomedical signal processing.In this thesis,several types of traditional automatic detection algorithms for feature points were compared and studied,and automatic detection methods of ECG signal feature points based on deep learning algorithm were proposed.The main research contents of this thesis were as follows:(1)Aiming at the problem that traditional feature point detection algorithms cannot systematically compare or describe their advantages in the detection of various ECG waveforms,in order to explore the detection capabilities of traditional detection algorithms for various waveforms,Several typical traditional detection algorithms were used to automatically detect ECG signal feature points in the LU database,and the detection performance of different traditional algorithms was compared in the detection of each type of waveform feature points.The P-T algorithm had the best performance in the detection of R wave peaks,with an average F1 of 97.97%;the geometric relationship algorithm had the best performance in the detection of the onset and offset of the QRS complex,with an average F1 of 99.81%;the differential threshold algorithm had the best performance in the detection of the onset and offset of the P wave,and the average F1 was 83.46%;the detection accuracy of the cumulative integral area algorithm in the identification of the onset and offset of the T wave was high,and the average F1 was 91.78%.(2)Aiming at the problem that the traditional feature point detection algorithm needs to pre-specify the threshold or other assumptions,which leads to the poor generalization ability of the algorithm,an automatic detection algorithm of ECG signal feature points based on the SegNet model was proposed.The model consisted of two parts:the encoder and decoder.The encoder extracted rich ECG signal features through traditional convolution,and used pooling operation to reduce the redundancy of features,so as to obtain high-order semantic information.The decoder compensated for the loss of details caused by the partial pooling process of the encoder through deconvolution operations,while ensuring that the size of the input data and output data remain the same.On the LU database,the detection average F1 of the onset and offset of the P wave,the peak of the R wave,the onset and offset of the QRS complex and T wave were 93.17%,98.39%,98.64%and 97.15%,respectively.The experimental results showed that,compared with the traditional detection algorithm,the SegNet model achieved better detection results and had a certain generalization ability.(3)For traditional convolutional models,the context information is extracted by increasing the network depth or width,which not only significantly increases the model complexity,but also fails to capture the dependencies between time series.An automatic detection method of ECG signal feature points based on the ECG_Segnet model was proposed.The method first introduced a standard dilated convolution module in the encoder path to extract more useful ECG information features without increasing the complexity of the model;then the bidirectional long short-term memory network was added to the encoder structure to capture a large number of time series features;finally,the feature sets of ECG signals at all levels extracted in the different paths of the encoder were connected to the decoder for multi-scale decoding,so as to reduce the information loss caused by the pooling operation in the encoding structure.Compared with the SegNet model,the traditional detection algorithm and the advanced methods in current research,the ECG_Segnet model achieved more accurate detection results.On the LU database,the average F1 at the onset and offset of the P wave detection was 94.74%,the average F1 at the peak of the R wave detection was 98.58%,the average F1 at the onset and offset of the QRS complex detection was 98.88%,and the average F1 at the onset and offset of the T wave detection was 97.53%.