Löwner-Based Tensor Decomposition for Blind Source Separation in Atrial Fibrillation ECGs

Abstract : The estimation of the atrial activity (AA) signal in electrocardiogram (ECG) recordings is an important step in the noninvasive analysis of atrial fibrillation (AF), the most common sustained cardiac arrhythmia in clinical practice. Recently, this blind source separation (BSS) problem has been formulated as a tensor factorization, based on the block term decomposition (BTD) of a data tensor built from Hankel matrices of the observed ECG. However, this tensor factorization technique was precisely assessed only in segments with long R-R intervals and with the AA well defined in the TQ segment, where ventricular activity (VA) is absent. Due to the chaotic nature of AA in AF, segments with disorganized or weak AA and with short R-R intervals are quite more common in persistent AF, posing some difficulties to the BSS methods to extract the AA signal, regarding performance and computational cost. In this paper, the BTD built from Löwner matrices is proposed as a method to separate VA from AA in these challenging scenarios. Experimental results obtained in a population of 10 patients show that the Löwner-based BTD outperforms the Hankel-based BTD and two well-known matrix-based methods in terms of atrial signal estimation quality and computational cost.
Complete list of metadatas

Cited literature [17 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02271029
Contributor : Pedro Marinho Ramos de Oliveira <>
Submitted on : Monday, August 26, 2019 - 2:53:21 PM
Last modification on : Wednesday, August 28, 2019 - 1:17:30 AM

File

EUSIPCO2019_VSubmitted.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02271029, version 1

Collections

Citation

Pedro Marinho R. de Oliveira, Vicente Zarzoso. Löwner-Based Tensor Decomposition for Blind Source Separation in Atrial Fibrillation ECGs. 27th European Signal Processing Conference, Sep 2019, A Coruña, Spain. ⟨hal-02271029⟩

Share

Metrics

Record views

18

Files downloads

22