Session S84.2

Multilead T-Wave Alternans Quantification Based on Spatial Filtering and the Laplacian Likelihood Ratio Method

V Monasterio*, JP Martínez

Universidad de Zaragoza
Zaragoza, Spain

This work addresses the 2008 Physionet/Computers in Cardiology Challenge. The aim of this challenge is to detect and quantify the magnitude of T-wave alternans (TWA) in every record of a given database. The database contains 100 multichannel ECG records sampled at 500 Hz with an approximate duration of two minutes. The subjects include patients with different conditions, as well as healthy controls and synthetic cases with calibrated amounts of TWA.
To automatically detect and estimate TWA, a multilead scheme is applied. This scheme was presented in last year's conference [CinC 2007;34:5-8], and combines principal component analysis (PCA) of the multilead ECG signal with a previously developed TWA analysis method [JP Martínez et al. IEEE TBME, 2006; 53:701-711].
In a preprocessing stage, baseline wander is removed by a time-variant filter with a cut-off frequency coupled to the prevailing heart rate. The signal is decimated to obtain a sampling frequency of 125 Hz, and low-pass filtered with a cut-off frequency of 15 Hz. On each beat, an interval of 300 ms is selected for TWA analysis (ST-T complexes). The starting point of each ST-T complex is located within a certain time from the QRS fiducial point, depending on the heart rate.
PCA is then applied to the multilead signal to obtain the principal components where alternans is better projected. The Laplacian Generalized Likelihood Ratio Test is used to decide whether TWA is present or not. A reconstructed signal is computed with truncated inverse PCA, where only the transformed leads with detected TWA are considered. Finally, Maximum Likelihood Estimation for Laplacian noise is applied to the reconstructed signal to estimate the alternans waveform.
The proposed scheme was tested on the challenge database. TWA was detected in 41 records, and a preliminary score of 0.795 was obtained (the best possible score is 1). These results suggest that accurate measurements of TWA can be obtained with the proposed scheme in multilead ECG records.

(Abstract Control Number: 363)