Dection Of Arrhythmia In Multi-lead ECG Based On BiLSTM

With the aging of the population in China and the increasing pressure of modern young people,the incidence of arrhythmia is also gradually rising.The current method of detecting arrhythmia is to manually interpret the 12-lead electrocardiogram(ecg)and require the ecg doctor or clinical technician to conduct abnormal examination of the ecg.Manual interpretation of ecg is too time-consuming and labor-intensive,and requires doctors to have rich knowledge of ecg pathology and experience in ecg disease diagnosis.Therefore,how to detect arrhythmia in 12-lead ecg automatically is a very important and challenging task.In this paper,nine types of arrhythmias are identified for multi lead ECG signals: Normal,Atrial fibrillation,First-degree atrioventricular block,Left bundle branch block,Right bundle branch block,Premature atrial contraction,Premature ventricular contraction,ST-segment depression,ST-segment elevated.They are considered as the representatives of arrhythmias.The main work of this paper is as follows:1.Aiming at the 12-lead ECG digitization problem,HSV thresholds and mathematical morphology are used to convert the ECG into digital signals.Aiming at the problem of crossover of lead waveforms commonly found in ECGs,a bidirectional curve prediction method is proposed to track the lead waveforms.Ensure the accuracy of ECG digitization and success rate.2.Aiming at the problem of noise interference in ECG signals,a zero-phase low-pass Butterworth filter combined with double-tree complex wavelet transform (DTCWT)was first used to filter out high and low frequency noise in ECG signals.3.Aiming at the problem of feature extraction in ECG signals,wavelet transform and adaptive double-threshold method are used to identify QRS complexes.For the problem that the R wave amplitude of individual samples is not obvious,linear amplification and sliding window integration are used to process the ECG signals to highlight R-wave,flatten other waveforms and use the "refractory period" method on the R-wave detection result to avoid QRS wrong picking.4.For the 12-lead ECG signal arrhythmia detection problem,this paper uses deep learning to detect the ECG signal,combines the model results with HRV statistical characteristics,and uses Catboost algorithm to perform model fusion work to further improve the heart rhythm F1 value for aberration detection.The data set of CPSC2018 Signal Challenge was used for model training and testing,and the test was performed on the ECG sample data set provided by Yifu Hospital of Nanjing Medical University and obtained excellent results.