ECG feature extraction based on the bandwidth properties of variational mode decomposition

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Detecting arrhythmia in long-term ECG recordings is challenging.
• Computer-aided diagnosis (CAD) systems support physicians in improving diagnostic accuracy for heartbeat detection.

Purpose of the Study:

• To explore the bandwidth properties of modes obtained using Variational Mode Decomposition (VMD) for classifying arrhythmia electrocardiogram (ECG) beats.
• To assess the efficacy of VMD-based features in distinguishing between different types of heartbeats.

Main Methods:

• ECG signals from the MIT-BIH arrhythmia database were decomposed using VMD.
• Amplitude modulation bandwidth (B AM), frequency modulation bandwidth (B FM), and total bandwidth (B) of the modes were extracted as feature vectors.
• Bandwidth estimations were based on instantaneous frequency (IF) and amplitude (IA) spectra.

Main Results:

• VMD-based features demonstrated sufficient class discrimination capability for ECG beats.
• The Bagged Decision Tree (BDT) classifier, using VMD-extracted bandwidth features from four modes, achieved high accuracy rates.
• Accuracy rates ranged from 98.71% to 99.51% for different types of ECG beats, with an overall accuracy of 99.06%.

Conclusions:

• The proposed VMD-based feature extraction method is effective for classifying arrhythmia ECG beats.
• The VMD approach, combined with classifiers like BDT, offers a promising tool for computer-aided diagnosis of heart conditions.
• This technique can significantly enhance the accuracy and efficiency of arrhythmia detection in clinical practice.