A hierarchical method based on weighted extreme gradient boosting in ECG heartbeat classification.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiology

Background:

• Electrocardiogram (ECG) is crucial for diagnosing heart disease.
• Automated ECG analysis enables remote heart monitoring, particularly with wearable devices.
• Accurate classification of heartbeats is essential for automated arrhythmia recognition.

Purpose of the Study:

• To develop an accurate classification method for electrocardiogram (ECG) heartbeats.
• To investigate the effectiveness of extreme gradient boosting (XGBoost) for single heartbeat classification.
• To improve automated arrhythmia detection for clinical applications.

Main Methods:

• A hierarchical classification approach based on weighted extreme gradient boosting (XGBoost) was proposed.
• Extensive features from six categories were extracted from preprocessed heartbeats.
• Recursive feature elimination was employed for feature selection, followed by a hierarchical classifier composed of threshold and weighted XGBoost classifiers.

Main Results:

• The method achieved high sensitivities for normal (92.1%), supraventricular (91.7%), and ventricular (95.1%) beats.
• Positive predictive values were 99.5% for normal, 46.2% for supraventricular, and 88.1% for ventricular beats.
• The approach demonstrated effectiveness in an inter-patient experiment conforming to AAMI standards.

Conclusions:

• Weighted extreme gradient boosting (XGBoost) was effectively applied to single heartbeat classification for the first time.
• The novel method demonstrated superior performance compared to existing approaches.
• The method is suitable for clinical use, offering high positive predictive value for normal heartbeats and high sensitivity for abnormal ones.