Session S72.2

Wavelet Transform Coherence Estimates in Cardiovascular Analysis: Error Analysis and Feasibility Study

K Keissar*, LR Davrath, S Akselrod

Tel Aviv University
Tel Aviv, Israel

Wavelet Transform Coherence (WTC) is a non-parametric method, which can provide insight into the transient linear coupling between two time series in cardiovascular variability analysis, via the computation of time-frequency maps of the time-variant coherence. WTC could allow quantitative analysis of transient events in specific times or frequency bands which is not possible when using classical steady state coherence approach. Computer simulations were performed to asses the bias and standard deviation (SD) of the WTC estimates over variable wavelet coefficient (6 < Wo < 30){Wo=Omega zero}and coherence real value. Next, a method for determining the coherence threshold for specific frequency band was developed and false positive and false negative error rates were used to evaluate this threshold. Furthermore, the WTC ability to detect short epochs of coherence discontinuities was evaluated. Subsequently, WTC was applied on heart rate (HR) and respiration signals from two experimental protocols: Supine rest for 30 min, and Change of Posture (CP) from supine to standing recorded from 8 subjects. The accuracy of the WTC estimates increased for higher levels of coherence and higher values of wavelet coefficient. The relative bias for the WTC estimate was <5% for coherence > 0.5 for Wo > 15. The SD of the coherence estimates increased to up to 15% for Wo > 15 as real coherence decreased. With respect to increasing Wo, the coherence threshold decreased from 0.8 to 0.3. Error rates decreased for larger Wo. The minimum discontinuity epoch time increased from 10 to 50 second with respect to increasing Wo. Coherence between the HR and respiratory signals focused mainly within the HF band (0.18 - 0.4 Hz) throughout the supine rest. Although consistent, the coherence in the HF band was far from stable, with epochs of weak or no coherence at all (below the threshold level). The dynamic trend of the coherence in the case of a time defined physiological stimulus was demonstrated during CP. The WTC responds with a drop of coherence which reaches its minimum value 20 seconds after CP and recovers to a value significantly lower than in supine position (p<0.001). Results from simulated signals have shown that 15 < Wo < 20 was optimal when considering bias, error rate and the ability to detect coherence discontinuities. Analysis of the supine rest data exhibited a dynamic behavior, previously considered as steady state. Analysis of the CP revealed the dynamic response of the coherence strength between HR and respiratory signals.

(Abstract Control Number: 217)