After severe acute brain trauma, cerebrovascular autoregulation (AR) can be impaired, but the performance of this homeostatic mechanism cannot be interrogated directly due to the complexity of the vascular system and existing challenges in assessing cerebrovascular phenomena. When indicated by the severity of brain trauma, clinicians continuously monitor intracranial pressure (ICP) to assess cerebral perfusion as a proxy measure for neural tissue oxygenation. The Monroe-Kellie doctrine states that the sum of the brain tissue, blood in the cerebrovascular bed, and cerebrospinal fluid in the ventricles is held constant within the cranial cavity; the resultant pressure of these volumes within the cranial cavity is ICP which fluctuates during a single cardiac cycle. Where ABP presents two peaks corresponding to systole and diastole, ICP presents three distinct peaks that correspond to cardiac systole (peak 1), cerebrovascular compliance (peak 2), and cardiac diastole (peak 3). Recent research on the morphology of individual intracranial pressure beats indicates the potential to use transient morphological changes in ICP between cardiac cycles in the time domain to gain greater insight into the physiological performance of AR and near-future ICP.

In order to highlight fluctuations in ICP behavior between successive cardiac cycles, this dissertation presents a novel method to analyze ICP morphology by transforming this cerebral pressure data from the time-domain to the phase-domain. Since existing mathematical models of the cerebrovascular performance focus on longer time-scale ICP behavior and clinically measured ICP morphology during cardiac cycles is often erratic, this dissertation demonstrates a novel morphological simulation of ICP to test a phase domain metric, the phase area ratio (PAR) in application to ICP monitoring.

An additive Gaussian simulation of ICP was developed to specifically examine the behavior of ICP Peak 2 that represents cerebral compliance, which is the component of AR that cannot be assessed directly using other existing physiological measures. This dissertation tests the hypothesis that phase domain analysis of ICP is useful as a forecasting tool for intracranial hypertension (IH) after severe acute brain trauma and post-surgical intervention. To test this hypothesis, 300 simulated ICP cycles and over 1 million clinical ICP cycles from 7 patients were analyzed. The simulated data were analyzed in a linear model that showed an R-squared value of no more than 0.76 for PAR and peak 2 amplitude, and the model showed a 0.93 R-squared value or higher between mICP and peak 2 behavior. The Spearman's correlation presented weak positive correlations between PAR and ICP ranging from 0.4 for the 1-hr time span to -0.1 for the 0.1-hr time span in time segments preceding intracranial hypertension (preIH). Overall in the clinical data examination, PAR was successfully able to differentiate between time periods of intracranial normotension and preIH time periods ranging from 1 to 0.1 hours using a Kolmogorov-Smirnov test for 67.9% of time periods tested in the seven patients. PAR performed with a lower area under the curve (0.53) than the time domain metric, Sample Entropy (SE) (0.71), when tested as a threshold classifier using receiver operator characteristic analysis for all time points in the exemplar patient. When analyzing all patient data, the area under the curve for PAR came out to 0.43 for a 1-hr window. A confusion matrix analysis of all patient data that yielded similar results as the receiver operator curve analysis. Using a logistic regression approach for prediction measurement, the results showed that PAR adds value to the performance of the model, where a longer amount of prior information yields better predictions for shorter times into the future. When PAR was used in conjunction with other metrics in a classifier, PAR-based metrics were more valuable that PAR itself. Overall PAR is a parameter that (1) requires less data for calculation than existing metrics, (2) has a bounded range between 0 to 1, and (3) does not have discontinuities like comparable complexity metrics. Ultimately, this work shows that PAR contributes unique information to existing multi-parameter prediction algorithms to forecast IH.