Recent Advances in An Optimized Dynamic Deep Unfold Network Model for Predicting Cardiac Arrhythmias Based On 12 Lead ECG Signals: A Systematic Review

Abstract

Cardiac arrhythmias represent a significant global health concern, necessitating early and accurate detection to reduce morbidity and mortality. The integration of artificial intelligence with electrocardiogram analysis has revolutionized arrhythmia prediction, particularly through deep learning-based approaches. Recently, optimized dynamic deep unfolding network models have emerged as a promising paradigm by combining the interpretability of model-based methods with the adaptability of data-driven learning. This systematic review explores recent advancements in such hybrid architectures for arrhythmia prediction using 12-lead ECG signals. The review critically examines signal preprocessing techniques, feature extraction strategies, and the evolution of deep unfolding frameworks that incorporate optimization principles into neural networks. Emphasis is placed on dynamic adaptability, computational efficiency, and clinical applicability. Furthermore, the study evaluates benchmark datasets, performance metrics, and real-world deployment challenges. The findings indicate that dynamic deep unfolding networks achieve superior classification accuracy and robustness compared to conventional convolutional and recurrent models, particularly in handling noisy and heterogeneous ECG data. Despite these advancements, challenges such as data imbalance, model generalization, and interpretability persist. This paper provides a comprehensive synthesis of current methodologies and outlines future research directions aimed at enhancing reliability and scalability in clinical environments.