Detection of heart arrhythmia based on UCMFB and deep learning technique

Severe cardiovascular diseases (CVD) are the leading cause of death worldwide. In the emergency scenario, reliable electrocardiography (ECG) is critical for the rapid diagnosis and management of acute CVD. Deep learning (DL) is the most important leading technology for automatic computer-aided ECG detection of cardiovascular disorders. This work proposed a ResNet-50 model that classifies healthy people and patients with four types of CVD based on ECG abnormalities. The ECG signals were decomposed using a uniform cosine modulated filter bank (UCMFB) that helps in the easy identification of irregular and regular heartbeats. The study was performed on four different types of ECG databases specifically for short segmented (i.e., 2sec and 5sec) and long segmented (5min and 8min) time frames, and these sub-signals are converted into 2-D images using wavelet transform packet (WTP). The extensive tests result in the identification of AF, CHF, HT, and NSR classes with an accuracy, recall, precision, and F1-score of 99.93%, 99.96%, 99.89%, and 99.95%, respectively for multi-class classification. The proposed approach undergoes different fold cross-validation techniques and has achieved high classification accuracy when compared with different state-of-the-art models, demonstrating the superiority of our system over previous systems. It is discovered that the proposed technique achieves decayed computational complexity; thus, it is recommended for categorization challenges. The suggested approach has the potential to gain essential clinical acceptability and be used for ECG prioritisation of CVD detection in clinics and out-of-hospital situations.