Annu Int Conf IEEE Eng Med Biol Soc. 2025 Jul;2025:1-7. doi: 10.1109/EMBC58623.2025.11252764.
ABSTRACT
Cardiac arrhythmias constitute a significant subset of cardiovascular diseases (CVDs) and can lead to severe complications such as stroke, heart failure, and sudden cardiac arrest. Accurate and timely diagnosis of arrhythmias is crucial for effective intervention, yet traditional electrocardiogram (ECG) interpretation relies heavily on expert analysis, which can be prone to variability. Recent advancements in machine learning (ML) and deep learning (DL) have shown promise in automating ECG analysis. However, existing one-dimensional (1D) approaches often fail to fully capture the spatial and temporal complexities of ECG signals. To address these limitations, we propose two novel two-dimensional representations for 12-lead ECG signals-Recurrence Polar Plots (RPP) and Cross Recurrence Polar Plots (CRPP)-which effectively reflect signal features while capturing the cyclic and directional patterns inherent to different leads. A multichannel neural network model is trained on these 2D maps to classify four major arrhythmia types. Our model achieves high classification accuracies: 84.0% for Atrial Fibrillation (AF), 94.5% for Sinus Tachycardia (ST), 91.5% for Sinus Bradycardia (SB), and 93.5% for Ventricular Tachycardia (VT), demonstrating the efficacy of leveraging complementary information from RPP and CRPP maps in a unified model for robust arrhythmia detection.
PMID:41337407 | DOI:10.1109/EMBC58623.2025.11252764