Session S34.3

Analysis of Surface Atrial Signals Using a Spectral Method for Time Series with Missing Data

R Sassi*, VDA Corino, LT Mainardi

Università di Milano
Milano, Italy

Analysis of atrial signals (AS, often called “fibrillatory waves”) extracted from the surface ECG on subjects undergoing atrial fibrillation (AF) have been documented to provide significant information on the properties of AF events and on the responsiveness to anti-arrhythmic drug or cardioversion. In particular, the main fibrillatory frequency (FF) has been related to atrial refractoriness and has been widely used to monitor electrophysiological properties of the fibrillating atria for analysing anti-arrhythmic drug actions or spontaneous diurnal variability. Quantification of FF is obtained through the spectral analysis of AS, which is usually extracted from surface ECG by removing waves induced by the ventricular activities. The extraction of AS requires advanced signal processing techniques, since atrial and ventricular activities, during AF, overlap in time and frequency. These techniques include averaged beat subtraction (ABS), spatio-temporal QRS-T cancellation or methods based on principal component analysis (PCA). However, if detecting the main fibrillatory frequency is the only goal, subtraction of the ventricular activity may be avoided by performing spectral analysis on those ECG intervals where ventricular activity is absent, (i.e. the T–Q intervals). In this case, two different strategies are at hand. The first, already applied in early works, is based on averaging periodograms computed on several short TQ intervals. The second, which we introduce in this paper, blinds the portion of ECG related to QRS and T waves leading to a single, discontinuous in time, atrial signal. The so formulated problem might be recast to that of missing data in a long time series and proper spectral analysis methods might be applied. With this objective in mind, we evaluated two methods: the Lomb Periodogram (LP) and the Incomplete Time Series Periodogram (ITSP). In particular we explored the capability of these methods in detecting the main FF when applied on an ECG recordings with QRS and T wave removed. We analyzed the data coming from the Physionet AF Challenge Database containing 80 AF records, each 1 minute long, sampled at 128Hz. For this dataset, we took as a reference the FF obtained by spectral analysis of AS using a modified ABS technique in which two separate templates were used for QRS and T waves. We observed that the mean absolute error was 0.76 +/- 1.21 Hz, (mean+/-SD), for LP and 0.41 +/- 0.46 Hz for ITSP. We concluded that estimation of FF is feasible without applying QRS-T subtraction.

(Abstract Control Number: 57)