Cardiac arrhythmia, which is an abnormal heart rhythm, is a common clinical problem in cardiology. Arrhythmias can be divided into many Categories, and accurate detection of arrhythmias can effectively prevent heart disease and reduce mortality. However, existing screening methods require long time monitoring and are low cost and low yield. Our goal is to develop a mixed depth model for processing time series to predict multi-classification electrocardiograph (ECG).
In this study, we developed a new, more robust network model named Hybrid Convolutional Recurrent Neural Network (HCRNet) for the time-series signal of ECG. This model utilized a nine-class ECG dataset containing tens of thousands of data to automatically detect cardiac arrhythmias. At the same time, a large imbalance arose because some of the cases in our selected MIT-BIH atrial fibrillation database had less than 100 records, but some had more than 10,000 records. Therefore, during data preprocessing, we adopted a scientific and efficient method to solve the ECG data imbalance problem. In the experimental studies, 10-fold cross validation technique is employed to evaluate performance of the model.
In order to fully validate our proposed model, we conducted a comprehensive experiment to investigate the performance of the proposed method. Our proposed HCRNet achieved the average accuracy of 99.01% performance and the average sensitivity of 99.58% performance on this dataset. Results suggest that the proposed model outperformed some state-of-the-art studies in ECG classification with a high overall performance value.
The HCRNet model can effectively classify arrhythmia signals in nine categories and obtain high efficiency, accuracy and F1 values. These improvements in efficiency and accuracy explain the rationality and science of setting up the modules in the HCRNet. By using this model, it can help cardiologists to correctly identify heartbeat types and perform arrhythmia diagnosis quickly.

Copyright © 2021. Published by Elsevier B.V.

Author