System Identification of Human Cerebral Blood Flow Regulatory Mechanisms

Abstract

Regulation of cerebral blood flow (CBF) is important to protect the brain against ischemia or excessive capillary pressure and hyperperfusion. In addition to the myogenic and metabolic mechanisms of pressure autoregulation, which tend to maintain CBF relatively constant with changes in arterial blood pressure (ABP), several other variables, like CO2, intracranial pressure, metabolic demand, sympathetic activity, cerebrovascular resistance, and endothelial metabolites are also involved in the dynamic control of CBF. Increasingly, system identification techniques are being used to shed light on the physiology of CBF regulation and to provide clinical tools for diagnosis, monitoring, and prognosis of patients with cerebrovascular conditions. In the frequency domain, the transfer function between ABP and CBF for normal subjects, is characterized by a reduced coherence, below 0.1 Hz, a positive phase response, below 0.2 Hz, and an amplitude frequency response that tends to rise continuously with frequency. One or more of these functions have been shown to be altered in patients with prematurity, carotid artery disease, severe head injury, and subarachnoid hemorrhage. Time-domain approaches are becoming more popular, involving autoregressive moving average structures (ARMA), cross-correlation analysis, or linear system models, such as the second-order differential equation used to calculate the dynamic autoregulation index (ARI), which has been adopted in many clinical studies. Nonlinear methods, based on the Wiener–Volterra approach have also been proposed, including new implementations based on dedicated neural networks. One trend in this area is the use of multivariate time-domain models to include other variables, such as CO2, in addition to ABP. However, the inclusion of other key variables, such as intracranial pressure, metabolic demand, and sympathetic activity, will depend on refinements of noninvasive measurement techniques to quantify their input load. Open access databases, containing representative recording from patients with different conditions, would be important to standardize the validation of system identification techniques by different research groups. The use of hybrid models, incorporating elements of mathematical modeling of well-known physiological phenomena, might also be a worthwhile approach in the future.