Multivariate system identification of cerebral autoregulation
Journal
Journal of Cerebral Blood Flow and Metabolism
Journal Volume
27
Journal Issue
SUPPL. 1
Pages
BP01-07M
Date Issued
2007
Author(s)
Abstract
Background: Blood flow in the brain is controlled by numerous processes in both normal and pathological conditions. Techniques to assess the control of cerebral blood flow by systemic variables (such as arterial blood pressure, ABP) frequently attempt to use a linear time invariant system identification approach to assess the extent to which low frequency content in variables related to cerebral oxygenation (cerebral blood flow velocity, CBFV) can be explained by a linear dynamical system acting upon low frequency content in systemic variability (for example oscillations in mean arterial blood pressure). In the case of a univariate input-output relationship, the coherence function is typically found to be very low at low frequencies, which means that the validity of the procedure in identifying the system dynamics at these frequencies has been questioned. Many studies argue that the presence of low coherence indicates strong non-linearity of cerebral autoregulation. However, CBFV variation is not entirely determined by pressure. Other mechanisms responsible for controlling blood flow include the reactivity of cerebral vessels to arterial CO2 and O2 levels. The observed low value of univariate coherence between ABP and CBFV in the low frequency region might be due to the additional input of CO2 and O2, rather than the strong non-linearity of the transfer path from ABP to CBFV. To test this hypothesis, the method of multiple coherence analysis has been used to investigate the contribution of CO2 and O2 to CBFV variability. Multiple coherence quantifies the extent to which the proposed output of the system can be explained by a linear combination of the input variables. Methods: Data from 13 healthy subjects, with measurements of beat-to-beat spontaneous fluctuations in mean arterial blood pressure (ABP) and cerebral blood flow velocity (CBFV), breath-to-breath arterial CO2 (PaCO2) and O2 (PaO2) pressure fluctuations was used. The multiple coherence function of the relationship between an assumed CBFV output driven by input ABP, PaCO2 and PaO2 was computed using non-parametric system identification (division of cross and power spectra). Results and conclusion: The additional inputs of important physiological variables, PaCO2 and PaO2, resulted in significantly higher values of multiple coherence for frequencies <0.05Hz than the corresponding value obtained for univariate coherence with only a single assumed input of ABP. This illustrates that when additional variables that could be thought to act in the control of cerebral blood flow are accounted for using multiple coherence, the reliability of linear system identification at low frequencies is improved. Moreover, it is also found that the transfer function between ABP and CBFV time series at low frequencies can be modified by CO2 and O2 reactivity and no longer represents pressure autoregulation only. The effect of this variability can be illustrated by demonstrating how autoregulation index as measured by Tiecks using univariate system identification techniques is increased when multiple variability is accounted for.
Type
conference paper