Automatic classification of heartbeats using ECG morphology and heartbeat interval features

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Automated analysis of electrocardiogram (ECG) data is crucial for diagnosing cardiac arrhythmias.
• Accurate heartbeat classification is essential for effective patient monitoring and treatment.
• Existing automated systems often struggle with precise identification of various ectopic beat types.

Purpose of the Study:

• To develop and validate an automated method for classifying heartbeats into five standard categories (normal, VEB, SVEB, fusion, unknown).
• To evaluate the performance of a supervised learning-based statistical classifier using ECG morphology and interval features.
• To compare different classifier configurations for optimal automated heartbeat detection.

Main Methods:

• Utilized the MIT-BIH arrhythmia database, splitting 44 non-pacemaker recordings into two datasets (~50,000 beats each).
• Compared twelve classifier configurations based on two-lead ECG features (morphology, heartbeat intervals, RR-intervals) using supervised learning.
• Selected the best configuration on the first dataset and validated performance on the second independent dataset.

Main Results:

• Achieved a sensitivity of 77.7% and positive predictivity of 81.9% for ventricular ectopic beats (VEB).
• Reported a sensitivity of 75.9% and positive predictivity of 38.5% for supraventricular ectopic beats (SVEB).
• Demonstrated a false positive rate of 1.2% for VEB and 4.7% for SVEB, outperforming previous automated systems.

Conclusions:

• The developed automated ECG processing method effectively classifies heartbeats, particularly VEBs and SVEBs.
• The system shows improved performance over existing automated heartbeat classification techniques.
• This method holds promise for enhancing automated cardiac arrhythmia detection and analysis.