Robust Heartbeat Classification for Wearable Single-Lead ECG via Extreme Gradient Boosting

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Wearable electrocardiogram (ECG) devices facilitate continuous heart monitoring but often yield noisy signals, challenging automated analysis, particularly for single-lead data.
• Existing beat-level single-lead ECG diagnosis methods require improvement in accuracy and efficiency due to signal variability and noise interference.

Purpose of the Study:

• To develop and evaluate a novel beat-level ECG analysis method using new morphological features and an extreme gradient boosting (XGBoost) model.
• To enhance the accuracy and efficiency of classifying five-class heartbeats according to the Association for the Advancement of Medical Instrumentation (AAMI) standard.

Main Methods:

• Extraction of new morphological features from single-lead ECG heartbeats.
• Implementation of an extreme gradient boosting (XGBoost) algorithm for beat-level classification.
• Validation using the MIT-BIH Arrhythmia Database (MITDB) for static ECG data and a self-collected wearable ECG dataset for real-world conditions.

Main Results:

• The proposed method achieved 99.14% accuracy on the MIT-BIH Arrhythmia Database (MITDB).
• The method demonstrated robustness in wearable ECG monitoring, achieving 98.68% accuracy.
• Performance surpassed other state-of-the-art models in both static and wearable ECG analysis.

Conclusions:

• The novel morphological features and XGBoost-based approach significantly improve beat-level single-lead ECG diagnosis accuracy and efficiency.
• The method is robust and effective for analyzing ECG data acquired from wearable devices, addressing challenges posed by noise and signal variability.
• This work offers a promising solution for computer-aided automated ECG analysis in both clinical and remote monitoring settings.