Session SA4.4

Continuous Quantification of Spectral Coherence Using Quadratic Time-Frequency Distributions: Error Analysis and Application

M Orini*, R Bailón, L Mainardi, A Mincholé, P Laguna

Universidad de Zaragoza
Zaragoza, Spain

Quadratic time-frequency distributions (QTFD) have an excellent joint TF resolution, but their applicability is limited by the presence of interferences. Interferences make the measure of TF coherence (TFC) based on QTFD inconsistent, unless a specific framework is set up to reduce them. A methodology for robustly estimating TFC, based on signal-dependent smoothing of the Wigner-Ville distribution, is shown to provide a reliable continuous quantification of cardiovascular interactions during non stationary conditions. In the ambiguity function (AF) domain, the smoothing function iso-contours are ellipsis, whose eccentricity automatically adjusts to the TF structure of the signals being analyzed. The degree of smoothing depends on their major axes. In order to estimate the optimal length of the major axis which controls the time smoothing, the method requires a previous estimation of the central frequencies of LF and HF components: the former is estimated from the AF of the signals by means of the Hough Transform, while the latter can be estimated using an (in)direct measure of the respiratory rate. The length of the major axis which controls the frequency smoothing is estimated by an iterative process which aims at achieving full consistency of the estimator while preserving good TF resolution. Bias, standard deviation and tracking capability of the estimator are evaluated in different physiological situations through simulations involving simulated AM-FM HRV signals with known and controlled theoretical coupling. For theoretical coherence levels >0.5 bias is <0.05. Standard deviation is <0.025. For time-varying theoretical coherence, the estimator perfectly follow gradual slow changes and it is able to correctly localize decorrelating events. In order to assess the autonomic response to orthostatic stress, TFC between heart rate variability and systolic pressure variability has been continuously estimated before, during and after upright tilt testing in 15 healthy subjects. In all subjects but one tilting provokes a quick coherence increase in both LF and HF band. Statistical analysis shows that during tilting the coupling in both LF and HF ranges is significantly higher (p<0.02) than during previous and later supine position.

(Abstract Control Number: 170)