Session S32.4

Significance of Mixing Matrix Structure on Principal Component Based Analysis of Atrial Fibrillation Body Surface Potential Maps

P Bonizzi*, M Guillem, F Castells, AM Climent,
V Zarzoso, O Meste

University of Nice
Sophia Antipolis, France

Clinical investigation on atrial fibrillation (AF) is a crucial task for deciding about the clinical intervention. New technologies like Body Surface Potential Maps (BSPM) increase spatial sampling compared to conventional ECG and allow more accurate analysis of AF. Guillem et al. (2009) showed that using BSPM recordings it is possible to get information about the degree and wavefront direction of AF. This work aims to test if the degree of the AF organization is reflected on the stationarity of the mixing matrix given by a principal component analysis (PCA) derived for consecutive segments along the BSPM recording. Mixing matrix pseudo-stationarity is evaluated on atrial activity (AA) signals obtained concatenating only the TQ segments present in a BSPM recording, and analysing them 10s by 10s. PCA was carried out on each 10s segment and the ability of the mixing matrix of one segment to retrieve the AA components of subsequent segments was analysed. Two real datasets composed of 6 BSPM recordings each one, the first (G1) previously classified as AF type I (organized AA) and the second (G2) as AF type III (more disorganized AA), have been employed. Stationarity along the recording was analysed as follows: the mixing matrix from the PCA of the first 10s segment was exploited to recover the AA component of the subsequent 5 segments. The recovered AA components were then compared with the original ones obtained from the PCA model derived for each segment. Recovered AA and original AA components were compared in terms of Pearson’s coefficient (Pc) and spectral coherence (SPC), and the number of significant singular values was retrieved from the PCA analysis of each segment. Mean (±SD) Pc of G1 was 0.43±0.09, compared to 0.15±0.06 of G2 (p <0.001). Mean (±SD) SPC of G1 was 0.57±0.06, compared to 0.47±0.04 of G2 (p <0.05). Number of significant components giving a percentage of variance greater than 95% was 3.23±0.71 for G1, compared to 7.64±3.30 of G2 (p <0.05). The results show that a greater AF organization is reflected in a greater pseudo-stationarity of the mixing matrix along the BSPM recordings and in a lower number of components needed to represent the AA, which can be interpreted as a lower complexity in the underlying AA in patients with AF type I.

(Abstract Control Number: 48)