A Novel Use of Transfer Function Estimation for Early Assessment of Brain Injury Outcome

Normal blood flow in the brain as a response to pressure fluctuations is commonly referred to as cerebral autoregulation. Linear, nonparametric models of cerebral autoregulation were estimated for 77 human subjects afflicted with brain injury, with mean arterial pressure used as input, and invasively measured regional cerebral blood flow used as output. The data were continuously monitored from the beginning of subject hospital stay. Mean transfer function gain as a function of frequency was calculated for each subject over a limited time window spanning 48 h, starting postsurgery. The mean transfer function gain of the cerebral autoregulation model provided a highly accurate, statistically significant, assessment of patient outcome. Subjects were accurately grouped by outcome, with a high significance (
$p < 0.013$
) across the entire measurement spectrum (0.005–0.25 Hz). In addition, the frequency band from 0.1 to 0.25 Hz contains particularly low variance in transfer function gain. This spectral region provides a highly statistically significant (
$p < 1.7{e}\hbox{-}3$
) assessment of brain injury patient health that is useful for outcome prediction.