Hybrid Deep Learning Framework for Enhanced Cardiovascular Disease Detection Using ECG Signal

Abstract

Cardiovascular Disease (CVD) is a serious medical issue in today's society. The electrocardiogram (ECG) is considered the most appropriate non-invasive diagnostic technique for detecting cardiac conditions. However, interpreting an ECG requires specialist experience and is time-consuming. This underscores the need for automated CVD diagnosis using advanced techniques. Many researchers have proposed various techniques to identify CVD. However, current approaches have been inefficient in identifying small differences due to the irregular and complex nature of the ECG rhythms. This research proposes a novel hybrid deep learning (DL) model called CNN (Convolutional Neural Network)-GRU (Gated Recurrent Unit)-Transformer. The spatial features of ECG are retrieved by the CNN, and temporal features are retrieved by the RNN model. Both features are fused and classified for CVD detection using the Transformer network. The fusion of features helps detect the minor changes in ECG and identify CVD with high reliability. The experimental outcome of the proposed model on the PTB-XL database for ECG classification of CVD shows the highest accuracy of 98.8% and the lowest false negative rate (FNR) and false positive rate (FPR) of 1.2% and 0.3%, respectively. The importance of the proposed network architecture is analyzed through an ablation study. Two ablation studies are conducted: first, the CNN is removed, and the GRU features are given to the Transformer for classification; in the second study, the GRU is removed. The ablation study shows accuracies of 95.8% and 97%, which are significantly lower than the proposed model’s accuracy. Additionally, the proposed network is compared with existing research. The outcome shows that the proposed network outperforms state-of-the-art techniques in detecting changes in ECG for CVD classification. The analysis of the proposed network suggests that it is a promising tool for detecting CVD at earlier stages with high accuracy rates.