Atrial Fibrillation Detection Method Based on Inception-CNN-BiLSTM Model

Electrocardiogram (ECG) automatic classification is an important auxiliary diagnostic method for diagnosis of arrhythmia. In order to improve the accuracy of beat extraction of dynamic ECG anomalies, a method for atrial fibrillation de-tection and beat classification based on Inception-CNN - BiLSTM is proposed. First, wavelet denoising and heartbeat splitting are carried out for ECG signals in the original dataset to remove interference in the original ECG signal and unify data dimensions for model training. Then, the data is trained and tested by using the Inception-CNN-BiLSTM model. The model outputs are divided into three categories: atrial fibrillation (A), normal (N) and other arrhythmia (O). Finally, several evaluation metrics are used to evaluate the performance of the model. It is found that the performance of the proposed model is improved and the number of parameters could be reduced compared to the CNN-BiLSTM model, and the accuracy rate is increased by 0.75% and 4.87% compared to the CNN and BiLSTM models, respectively.