Session PB9.5

Ventricular Activity Residual Reduction in Remainder ECGs Based on Short-Term Autoregressive Model Interpolation

P Bonizzi*, M Stridh, L Sörnmo, O Meste

University of Nice
Sophia Antipolis, France

Analysis of the atrial activity (AA) extracted from the surface ECG in subjects with atrial fibrillation (AF) have been demonstrated to provide useful information on the characteristics of this arrhythmia. The accuracy in this extraction is related to the accuracy and the detail degree of subsequent analysis. Up to now, there is no AA extraction method which can produce a remainder ECG which is completely free from ventricular residuals. The present study puts forward a new method for detection of ventricular residuals and their reduction in the remainder ECGs. The importance of QRS residuals is evaluated in each QT segment by a new residual index, based on suitable preprocessing and the QT segment amplitude. Each time the remainder ECG in the QRS interval exceeds a threshold based on the residual index, autoregressive (AR) interpolation is applied to reduce its amplitude. The QRS interval is then treated as an interval with missing samples using a two-stage procedure: in the first stage, the AR model coefficients are estimated from the known part of the signal. In our case, this is represented by the two SQ segments surrounding the current QRS interval. In the second stage, the estimated model coefficients are used to replace the missing samples in the QRS interval through interpolation. Simulated AA remainder ECGs have been exploited for determining a suitable AR model order and the residual index threshold. The performance has been evaluated in terms of amplitude and spectral concentration (SC, being a measure of the spectral compactness of the dominant AF frequency) of the processed signals. The real dataset was composed of 19 patients with AF for which AA extraction was applied in order to get the remainder ECGs. Mean (±SD) amplitude of QRST segments of V1 was 0.23±0.16 mV, compared to 0.05±0.03 mV of the original remainders and to 0.05±0.03 mV of the remainders after applying interpolation. Mean (±SD) SC improved from 56.7±12.8 % of the original remainders to 58.1±13.3 % of the interpolated ones. It is concluded that the proposed algorithm improves the quality of the extracted AA remainders (improvement in the mean SC value) without attenuating their mean amplitudes inside the QRST segments.

(Abstract Control Number: 46)