Blind Source Separation in Persistent Atrial Fibrillation Electrocardiograms Using Block-Term Tensor Decomposition...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computational Cardiology

Background:

• Atrial fibrillation (AF) analysis relies on estimating atrial activity (AA) from ECGs.
• Existing tensor factorization methods like Hankel-based block term decomposition (BTD) struggle with weak AA in persistent AF.
• Persistent AF presents challenges due to short R-R intervals and disorganized AA signals.

Purpose of the Study:

• To develop a novel tensor-based approach for improved atrial activity (AA) signal estimation in electrocardiogram (ECG) recordings.
• To address the limitations of existing methods in analyzing persistent atrial fibrillation (AF) with weak AA signals.
• To enhance the noninvasive analysis of AF by accurately separating AA from ECG data.

Main Methods:

• Proposed a tensor approach to estimate and subtract QRS complexes from ECG signals, isolating the AA component.
• Formulated the problem as blind separation of rational functions, modeling QRS complexes explicitly.
• Developed a variant of the constrained alternating group lasso (CAGL) algorithm, incorporating Löwner structure via orthogonal projection (LCAGL).

Main Results:

• The proposed Löwner-constrained AGL (LCAGL) algorithm demonstrated consistency in extracting desired sources from synthetic data.
• Experiments on 20 persistent AF patients showed LCAGL outperforms other tensor-based methods.
• The method achieved superior atrial signal estimation quality, even from single-heartbeat ECG records.

Conclusions:

• The novel tensor approach effectively estimates atrial activity (AA) by first separating QRS complexes.
• This method overcomes limitations of prior techniques for analyzing persistent atrial fibrillation (AF).
• The proposed LCAGL algorithm offers a significant advancement in ECG-based AF analysis.