Quantifying Variability in Dynamic Cerebral Autoregulation in the Human Brain

Abstract

Introduction: The brain is a highly sophisticated system that operates dynamically which is in constant need of a continuous supply of oxygen and glucose via blood flow to sustain functionality and remain healthy. Despite changes in Arterial Blood Pressure (ABP), the blood pressure and flow are maintained at certain levels within the human cranium using a mechanism called dynamic Cerebral Autoregulation (dCA). The major challenge in this research thesis is that the dCA mechanism is a nonstationary process which means that measurements are varying over time. There are determinant physiological factors that make dCA a nonstationary mechanism including Carbon dioxide (CO2), body temperature and Intracranial Pressure (ICP). Aim: This work aims to quantify the variability found in dCA by applying Transfer Function Analysis (TFA) using the MATLAB platform on a univariate scale and multivariate scale. There are several input variables influencing the dCA mechanism and HR input will be included as a measure of sympathetic control in this research. However, CO2 input is widely used in multivariate analysis where HR recordings were used for beat-to-beat averaging or filtering. Also, this thesis aims to examine and quantify the temporal variability of dCA which is found in measurement variability (across multiple recordings) and subject variability (across all recordings) on both univariate and multivariate scales. Methods: The study included 20 subjects recording their vital signs during 5 visits for each subject at 3 conditions (normocapnia, hypercapnia and thigh cuff conditions), forming a dataset of 300 vital signs recordings of ABP, CBv (right and left sides), CO2, and HR as well as mean recordings for ABP and both sides of CBv. Using univariate and multivariate techniques, this study will analyse measurements using the TFA technique. Then, apply reproducibility and covariance analyses where the first will measure ICC levels and the latter will quantify variabilities in measurement and subject variability recordings. Results: For the univariate analysis, the results demonstrate different behaviours in coherence, gain, and phase for normocapnia, hypercapnia, and thigh cuff conditions under different frequency ranges (HF, LF, and VLF). Overall, the thigh cuff condition shows the lowest variation patterns, especially at the LF band. In the HF band, the thigh cuff condition shows low variation in gain and phase but not in coherence. However, the normocapnia condition shows different pattern of variation in the VLF band where normocapnia coherence and gain show narrower measurement variability (the variability between the different visits for each subject for each condition) but larger subject variability (the variability both within each subject 5 recordings, and between the overall subjects recordings for each condition) than hypercapnia and thigh cuff conditions. On the other hand, normocapnia VLF phase variations are significantly narrower in both measurement and subject variabilities compared to the other conditions. Besides, covariance results show that measurement variability are significantly smaller than subject variability in normocapnia, hypercapnia and thigh cuff conditions at all frequency bands. For the multivariate analysis, using 2-inputs and 3-inputs in the TFA significantly increased coherence results compared to the univariate analysis results. Also, ICC results are significantly higher than the ICC results from the univariate analysis where the covariance results show measurement variability significantly smaller than subject variability at all physiological conditions across the frequency spectrum. Conclusion: On a univariate scale, the thigh cuff condition at the LF band exhibits the lowest variation levels among both the right and left sides of the brain compared to the HF and VLF bands. However, the multivariate analysis shows that coherence results appear to improve TFA parameter results as well as ICC values, particularly, when using 3-inputs analysis where covariance results appear to be similar to those found from univariate analysis. Overall, adding HR (Coh ABP+HR = 0.685, ± 0.160, mean, ± Std) to the TFA appears to have more influence than adding CO2 (Coh ABP+CO2 = 0.676, ± 0.137) in increasing coherence results, which has not previously been shown (Coh ABP+CO2+HR = 0.732, ± 0.128).