Comparative study of morphological and time-frequency ECG descriptors for heartbeat classification

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Accurate electrocardiogram (ECG) analysis is crucial for computer-assisted cardiac condition detection.
• Reliable heartbeat classification is essential for automated long-term monitoring systems.

Purpose of the Study:

• To compare the heartbeat classification performance of two ECG feature extraction techniques: QRS pattern recognition and Matching Pursuits.
• To evaluate the impact of general versus patient-specific learning datasets on classification accuracy.

Main Methods:

• Feature extraction using QRS morphological descriptors and Matching Pursuits.
• Classification using the Kth nearest neighbour rule on the MIT-BIH arrhythmia database.
• Comparison of performance using general learning sets (GLS) and patient-specific local learning sets.

Main Results:

• Both QRS pattern recognition and Matching Pursuits achieved high classification accuracies.
• Patient-specific local learning sets (optimal size ~3 min) improved accuracy over the general learning set.
• Matching Pursuits excelled at classifying repeating waveforms, while QRS descriptors were better for varied premature ventricular contractions.

Conclusions:

• Both feature extraction methods are effective for ECG analysis.
• Incorporating patient-specific data enhances the accuracy of automated heartbeat classification.
• The choice of method may depend on the specific types of arrhythmias being analyzed.