| Atrial Fibrillation(AF)is one of the most common persistent arrhythmias.The number of patients with atrial fibrillation exceeds 8 million in China.It has the early and continuous detection means that Electrocardiogram(ECG)is the most direct and effective diagnostic tool for detecting heart disease.With the development of artificial intelligence technology and the popularization of remote cloud medical models,it is of great significance to carry out research on atrial fibrillation detection based on surface electrocardiogram for the construction of an automatic analysis system for cardiovascular diseases.The research direction of atrial fibrillation detection can be roughly divided into two points.The method is a detection method based on feature joints,and the other method is an unsupervised method based on neural networks.However,both feature detection methods and unsupervised methods have shortcomings.The feature joint detection method has extremely high requirements on feature quality,and the unsupervised detection method requires a large number of labeled samples for training.These two types of methods have high requirements on the processing performance of the classification model and medical platform.To solve the problem,the paper proposed an ECG waveform recognition algorithm and the LSTM-based atrial fibrillation detection mechanism.First,characteristics of atrial fibrillation are analyzed from the mechanism of ECG generation and characteristics of atrial fibrillation;then the wavelet transform and differential correction method is used to locate the waveform reference point of ECG signal;finally,the method is used to identify atrial fibrillation that LSTM is modeled as the classification.The main research contents of the paper are as follows:(1)It is analyzed in detail that the mechanism of ECG signal generation and the noise characteristics of body surface acquisition signals.A sparse reconstruction noise reduction algorithm based on the non-convex penalty term is designed for the problem of waveform detail being removed by the noise reduction in the traditional algorithm.Through experimental comparison in the MIT-BIH Arrhythmia Database,it is proved that the non-convex penalty term can effectively reduce the loss of waveform spectral components.The non-convex penalty term can better save the effective information in the ECG signal compared with noise reduction algorithms such as wavelet transform.(2)The paper proposed waveform detection algorithm combining wavelet transform and differential correction after analysis characteristics of the starting point of P waves in ECG signals in detail.A joint detection mechanism is implemented in the detection window to correct the error caused by the reference point offset by setting the P-wave initial beacon node.After verification in the QT database,the detection sensitivity and specificity of the P-wave reference point reached 99.12% and 91.14%,respectively.(3)The paper studied and analyzed the influence of waveform on atrial fibrillation detection in detail,and designs and optimizes the LSTM network as a detection classifier.It is studied which the relation of the close correlation between the RR interval and P wave state for the expression of atrial fibrillation.Bi-directional LSTM with RR interval and P wave as input was constructed.The detection accuracy of the MIT-BIH Atrial Fibrillation database has reached 92.44%,which is superior to other detection algorithms.The paper has carried out related research on waveform recognition and atrial fibrillation detection,and achieved satisfactory results. |
PDF Full Text Request