Arrhythmia classification from single-lead ECG signals using the inter-patient paradigm

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Arrhythmia, a heart rhythm disorder, necessitates continuous electrocardiogram (ECG) monitoring via Holter devices.
• Large volumes of ECG data require automated systems for efficient arrhythmia detection.

Purpose of the Study:

• To propose an automated system for classifying arrhythmias using single-lead ECG signals.
• To evaluate the system's performance and robustness against segmentation errors.

Main Methods:

• Utilized a combination of RR intervals, signal morphology, and higher-order statistics for feature extraction.
• Validated the system using the MIT-BIH database with an inter-patient paradigm.
• Assessed robustness by introducing jitter to R-wave positions and analyzing feature group performance.

Main Results:

• Achieved high sensitivities for Normal (N), Supraventricular (S), and Ventricular (V) arrhythmias (93.7%, 89.7%, 87.9% respectively) even with added jitter.
• Reported positive predictive values of 99.2% (N), 36.8% (S), and 93.9% (V).

Conclusions:

• The proposed automated arrhythmia classification system outperforms state-of-the-art methods.
• Demonstrated robustness against segmentation errors, making it suitable for real-world clinical scenarios.