Research On Anomaly Detection Algorithm For Multi-lead ECG

Cardiovascular disease(CAD)is the leading killer of human health and has become a major public health problem.It is urgent to detect CAD in time.Electrocardiogram(ECG)detection is the most convenient and effective method among many existing CAD detection methods.However,the lack of medical resources has led to a large number of patients with CAD that cannot be detected in time,so the disease cannot be effectively controlled at the initial stage.The use of computers to quickly detect ECG signals can not only reduce the work pressure of medical staff,but also allow the masses to initially understand their own health conditions.In order to realize the rapid detection of abnormalities in multi-lead ECG signals,this thesis first takes the single-lead ECG signals with relatively simple forms and mature algorithms as the research object,and proposes an ECG signal classification model based on dilation causal convolution.Based on the research of single-lead ECG abnormality detection,considering the complexity of twelve-lead ECG,we first tried to apply the proposed model in the normal-abnormal classification task,and achieved good results through debugging.Then increase the types of abnormal ECG,and proceed to the next experiment.In order to complete the experiment,a twelve-lead ECG signal classification algorithm based on resonance sparse decomposition is proposed;the above algorithms have passed simulation experiments and cross-validation,and the actual performance is good.The main research contents of the thesis are as follows:1.Aiming at the problem that the existing ECG classification algorithm model is relatively fixed,an ECG signal classification model based on dilated causal convolution is proposed.This model uses dilated causal convolution to replace the traditional structure of CNNs+RNNs.Aiming at the problem of poor application of dilated convolution on small objects,further adjustments were made to the network model,changing the value method of dilatation factors,and selecting the most suitable combination of dilatation factors for ECG classification.2.Apply the proposed ECG signal classification model based on dilated causal convolution to a twelve-lead ECG to perform simple ECG classification tasks.In the course of the experiment,the network parameters were constantly adjusted.Aiming at the shortcomings of the existing multi-lead ECG classification algorithm,a new twelve-lead ECG classification algorithm was proposed,and the heart rate variability was added as an additional feature.In the process of classification.3.On the basis of the second step,the detection types are increased from 2 types to 9 types(8 abnormal + normal).In order to further improve the effectiveness of the algorithm,according to the characteristics of abnormal ECG signals,the resonance sparse decomposition is used for signal decomposition,Separate the continuous oscillation component and the instantaneous impact component in the original signal,and perform feature extraction on the continuous oscillation component and the instantaneous impact component to complete anomaly detection.