Abstract

Deep Learning (DL) has been suggested for the automated classification of heart abnormalities using ECG signals, although its practical applications in the medical field are limited. An extensive evaluation is presented focusing on the DL methodologies and the potential for DL-based ECG arrhythmia classification. The exploration of implementing DLdriven ECG classification involves the development of innovative DL models and further investigation into inter-patient models. Reviews rely on the MIT-BIH Arrhythmia Database for constructing DL models. Convolutional neural networks have been the predominant architecture utilized in recent research, incorporating various forms of DL. The assessment frameworks, specifically intra and inter-patient paradigms, are clearly addressed in the studies, revealing a significant gap in the inter-patient paradigm. Notably, the average F1 score in the studies reviewed is considerably lower. To enable the practical application of DL-based ECG classification in clinical settings, the exploration of diverse ECG databases, integration of novel DL frameworks, and deeper analysis of inter-patient models are potential avenues for future research.