Atrial activity selection for atrial fibrillation ECG recordings

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Atrial fibrillation (AF) detection relies on accurate analysis of electrocardiogram (ECG) signals.
• Independent Component Analysis (ICA) and Second Order Blind Identification (SOBI) are used to isolate atrial activity (AA) sources from ECG recordings.
• Existing methods lack robust criteria for selecting the most representative AA source.

Purpose of the Study:

• To propose and validate novel parameters for selecting the most representative atrial activity (AA) source from ICA-SOBI processed atrial fibrillation (AF) ECG recordings.
• To introduce correlation coefficient with lead V1 (CV1) and peak factor (PF) as new indicators for AA source selection.
• To compare the efficacy of the proposed selection method against established algorithms.

Main Methods:

• Application of Independent Component Analysis (ICA) followed by Second Order Blind Identification (SOBI) to 12-lead ECG recordings.
• Development and application of three parameters: correlation coefficient with lead V1 (CV1), peak factor (PF), and spectral concentration (SC) for source selection.
• Validation using synthesized data and 218 real-world AF ECG recordings.

Main Results:

• The proposed three-parameter method achieved high agreement (93.3%) in selecting the AA source for synthesized data.
• For real ECG data, the peak factor (PF) consistently fell within a defined range (2-4.5) for 89.5% of selected AA sources.
• The novel parameters (CV1, PF) captured features overlooked by spectral concentration (SC), and the overall method outperformed QRST cancellation, PCA, and standard ICA-SOBI.

Conclusions:

• The proposed parameters (CV1, PF, SC) provide a reliable method for identifying the most representative atrial activity (AA) source in AF ECGs.
• The peak factor (PF) offers a robust indicator for AA source selection within a specific value range.
• This novel approach demonstrates superior performance in isolating atrial activity compared to existing signal processing techniques.