Hybrid Neural Architecture for Enhanced ECG Classification: Integrating Parallel Convolutional, State Space, and Sequential Learning

Abstract

This paper presents a novel deep learning architecture that integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and State Space Models (SSM) for electrocardiogram (ECG) classification. The proposed model utilizes a parallel structure with multiple convolutional branches using large kernel sizes and an innovative ParallelConvSSMBlock, consisting of several parallel branches for convolution and state space processing. Positional embeddings and Mamba-inspired blocks are incorporated to capture both local and global temporal dependencies in ECG signals. This hybrid design effectively captures complex ECG signal patterns at multiple scales while ensuring computational efficiency. The proposed approach represents a significant advancement in automated ECG analysis, providing enhanced accuracy and reliability in the diagnosis of heart conditions through advanced deep learning methods. Furthermore, the model achieves an accuracy of 98.70%, demonstrating its effectiveness in ECG signal classification.