Journal of Research and Health

en A Hierarchical ECG-Based Framework for Inter-Patient Detection of Multiple Cardiac Arrhythmias ● Artificial Intelligence مقاله اصيل پژوهشي Orginal Article Background: Cardiac arrhythmias exhibit diverse electrophysiological characteristics, making their simultaneous detection from electrocardiogram (ECG) signals challenging. Moreover, the generalizability of automated detection methods to previously unseen patients is an important consideration for their practical application. This study proposes a hierarchical ECG-based framework for the inter-patient detection of five clinically relevant arrhythmias: ventricular tachycardia/ventricular fibrillation (VT/VF), atrial fibrillation (AF), premature ventricular contractions (PVC), premature atrial contractions (PAC), and Wolff-Parkinson-White (WPW) syndrome. Methods: Rather than treating all arrhythmias as a single multiclass classification problem, the proposed framework sequentially analyzes the ECG using arrhythmia-specific characteristics. VT/VF detection is based on signal periodicity and complexity measures, AF is identified using heart rate variability and Poincaré-derived features, PVC and PAC are detected using RR-interval characteristics followed by wavelet-based discrimination, and WPW is identified from PR-interval and QRS-duration characteristics. The framework was evaluated using ECG recordings from four databases under a leave-one-subject-out inter-patient protocol. Results: The VT/VF, AF, PVC/PAC detection, PVC-PAC discrimination, and WPW stages achieved accuracies of 98.90%, 97.06%, 98.64%, 97.73%, and 95.15%, respectively. Conclusion: The proposed hierarchical framework achieved high inter-patient performance across all detection stages. These findings indicate that combining arrhythmia-specific ECG characteristics within a sequential framework can provide effective detection of different rhythm- and beat-level abnormalities in previously unseen subjects. Electrocardiogram (ECG), Cardiac arrhythmia, Inter-patient classification, Feature extraction, Hierarchical classification 5 5 http://jrh.gmu.ac.ir/browse.php?a_code=A-10-3113-2&slc_lang=en&sid=1 Mohammad Ahmadabadi mo_ahmadabadi@elec.iust.ac.ir 0021296707 No Faculty of Electrical Engineering, K. N. Toosi University of Technology, Tehran, Iran Hossein Hosseini-Nejad hosseini_nezhad@kntu.ac.ir 0859811141 Yes Faculty of Electrical Engineering, K. N. Toosi University of Technology, Tehran, Iran Arash Rajabipoor Arash Rajabipoor@yahoo.com 0021486875 No University of Medical Sciences, Gilan, Iran