IMPROVE THE CARDIAC ARRHYTHMIA DETECTION WITH HYBRID APPROACH USING CONVOLUTIONAL NEURAL NETWORKS AND RECURRENT NEURAL NETWORKS

Abstract

Cardiac arrhythmias pose a serious problem in the field of cardiovascular diagnostics, where fast and precise detection is necessary. This study suggests a new hybrid deep learning framework for better arrhythmia detection based on electrocardiogram signals. The method proposed combines Convolutional Neural Networks and Recurrent Neural Networks in such a way as to use their respective strengths and offer the best results. CNNs are used for powerful spatial and morphological feature extraction from ECG signals of single heartbeats, and subtle patterns of each ECG beat are closely seen during this process. The extracted features are then passed on to RNNs, in particular Long Short-Term Memory networks, which are performing the task of characterizing the temporal dependencies and the contextual rhythm across the flowing beats. The two-stage analysis is what allows the system to detect both the isolated distorted waveforms and the inconsistencies in the heart's rhythmic sequence. Our hybrid model, when tested with the MIT-BIH Arrhythmia Database, shows impressive performance by being able to secure a greater accuracy, sensitivity, and specificity in comparison to the models based on only CNN or only RNN. Therefore, it can definitely be considered a powerful tool for automated cardiac diagnosis.