Nonlinear modeling of dynamic cerebral autoregulation in humans

Experimental data acquired from healthy subjects and under normal conditions, using recent high temporal resolution measurement techniques, namely transcranial Doppler ultrasonography and laser Doppler flowmetry, were analyzed following a novel nonlinear modeling approach in order to obtain an accurate representation of the system dynamics. The results demonstrated that cerebral autoregulation is a nonlinear and frequency-dependent phenomenon, exhibiting two distinct components with fast and slow dynamics respectively. Comparisons between linear and nonlinear models demonstrated that the nonlinear component of the system is more prominent in low frequencies, where autoregulation is more effective. The nonlinear component was also proven to be significant, since linear models achieved predictions with normalized mean square errors (NMSE) around 45%, whereas inclusion of nonlinear terms in the models improved the prediction NMSE to around 25%. Moreover, the relative contribution of the nonlinear terms displayed considerable variability.