Abstract：Objective To explore the application value of the BiLSTM-Attention hybrid neural network model in the prediction of
arrhythmia. Methods A total of 27036 electrocardiogram (ECG) data from the Chinese Cardiovascular Disease Database were
selected and divided into training set, validation set and test set according to the ratio of 8∶1∶1. The median filter method and
wavelet transform threshold method were used to denoise and preprocess the original ECG data. The BiLSTM model was used to
learn the features of the data, and the attention mechanism was integrated to allocate the weight. The BiLSTM-Attention model was
constructed to predict the classification of arrhythmia. The BiLSTM-Attention model was compared with long short-term memory
(LSTM), LSTM-Attention and BiLSTM models, and F1 score and area under curve (AUC) were used to evaluate the model.
Results The F1 score of the BiLSTM-Attention model was 0.799. The atrial fibrillation, first degree atrioventricular block, sinus
rhythm abnormalities, and sinus rhythm obtained high F1 scores, which were 0.955, 0.862, 0.954, and 0.917 respectively. The AUC
of nine types of arrhythmia was greater than 0.87. Conclusion The BiLSTM-Attention arrhythmia classification model has strong
classification ability and strong recognition ability for some arrhythmia abnormalities. After training, it can better assist clinical
diagnosis of arrhythmia, and has certain practical value.
杜丛强a，崔昊b. 基于BiLSTM-Attention混合神经
网络的心律失常预测[J]. 中国医疗设备, 2023, 38(11): 67-72.
DU Congqianga, CUI Haob. Arrhythmia Prediction Based on BiLSTM-Attention Hybrid Neural Network. China Medical Devices, 2023, 38(11): 67-72.
