Bayesian and lagged general linear modelling strategies in breath-hold induced cerebrovascular reactivity mapping with muti-echo BOLD fMRI

Cerebrovascular reactivity (CVR), the ability of blood vessels to dilate and constrict in response to a vasoactive stimulus, is an important indicator of cerebrovascular health and can be estimated using BOLD functional MRI. In this work, we evaluated a novel variational Bayesian method for mapping CVR dynamics which yielded similar CVR results and reproducibility to a time-shifted, “lagged” general linear model approach, and uncovered the need for more research into negative CVR. This novel approach is more time efficient and has the potential to improve CVR mapping by incorporating non-linear modelling and physiologically meaningful information (in the form of priors).