Continuous wavelet based transfer function analysis of cerebral autoregulation dynamics for neuromonitoring using near-infrared spectroscopy

Abstract

Introduction: Near-infrared spectroscopy is now a popular method in neuromonitoring. Derived parameters like cerebral oxygenation index or Fast Fourier Transform based coherence estimates between blood pressure and cerebral blood flow have their limitations of use for stationary data and low time resolution. Wavelet transfer function analysis can be employed to estimate coherence, gain and phase relationship between two signals without these restrictions. Methods: We aimed to extend the previously described Grinsted wavelet package with rectified bias of power and transfer function gain estimation of cerebral autoregulation assessment. The algorithm was validated in simulated signals and data of five healthy male subjects undergoing a protocol to produce large changes in hemodynamics by using lower body positive pressure (LBNP) with mild hypoxia and lower body negative pressure (LBNP). We intended to compare wavelet-based observations with FFT-based estimates. Results: We found good agreement between wavelet and FFT-based coherence and gain of cerebral tissue oxygenation index, in Bland Altman Plot and linear correlations for repeated measurement especially in the low frequency range (0.04 -0.15 Hz, coherence: r = 0.69, p < 0.001, gain: r = 0.74, p = 0.001), but was less in the very low frequency range (≤0.04 Hz, coherence: r = 0.65, p < 0.001, gain: r = 0.66, p < 0.001) and high frequency range (0.15-0.4 Hz, coherence: r = 0.39, p < 0.001, gain: r = 0.71, p = 0.001). FFT-based coherence was smaller than wavelet estimates for values <0.5. Discussion: We demonstrated good agreement in power, coherence, transfer function gain estimates between FFT-based method and our modified wavelet method. This was confirmed in simulated data and in healthy subjects undergoing LBNP and LBPP with hypoxia. Near-infrared spectroscopy-derived wavelet transform could be useful for exploring cerebral autoregulation dynamics, especially in non-stationary data.