Detection of shockable ventricular cardiac arrhythmias from ECG signals using FFREWT filter-bank and deep convolutional neural network.

Among various life-threatening cardiac disorders, ventricular tachycardia (VT) and ventricular fibrillation (VF) are shockable ventricular cardiac arrhythmias (SVCA) which require immediate defibrillation therapy for the survival of patients. Timely and accurate detection of rapid VT or VF episodes using ECG signals is extremely important before initiating external defibrillator (AED) and implantable cardioverter-defibrillator (ICD) therapies. In this paper, a novel approach for the detection of SVCA using ECG signals is proposed. The fixed frequency range empirical wavelet transform (EWT) (FFREWT) filter-bank is introduced for the multiscale analysis of ECG signals. The modes evaluated using FFREWT of ECG signals are used as input to a deep convolutional neural network (CNN) for the detection of SVCA. The architecture of the proposed deep CNN comprises of four convolution, two pooling, and four dense layers. The ECG signals from various public databases are used to evaluate the proposed FFREWT domain deep CNN approach. The results show that the proposed approach has obtained an accuracy of 99.036%, 99.800%, and 81.250% for the classification of shockable vs non-shockable, VF vs Non-VF, and VT vs VF, respectively using 8 s ECG frames with 10-fold cross-validation (CV) strategy. Our proposed approach has obtained an average accuracy value of 97.592% using 8 s ECG frames with subject-specific CV. The hardware implementation of the proposed SVCA detection approach can be done using an Internet of things (IoT) driven patient monitoring system.