Vascular autorescaling of fMRI (VasA fMRI) improves sensitivity of population studies: A pilot study

The blood oxygenation level-dependent (BOLD) signal is widely used for functional magnetic resonance imaging (fMRI) of brain function in health and disease. The statistical power of fMRI group studies is significantly hampered by high inter-subject variance due to differences in baseline vascular physiology. Several methods have been proposed to account for physiological vascularization differences between subjects and hence improve the sensitivity in group studies. However, these methods require the acquisition of additional reference scans (such as a full resting-state fMRI session or ASL-based calibrated BOLD). We present a vascular autorescaling (VasA) method, which does not require any additional reference scans. VasA is based on the observation that slow oscillations (< 0.1 Hz) in arterial blood CO2 levels occur naturally due to changes in respiration patterns. These oscillations yield fMRI signal changes whose amplitudes reflect the blood oxygenation levels and underlying local vascularization and vascular responsivity. VasA estimates proxies of the amplitude of these CO2-driven oscillations directly from the residuals of task-related fMRI data without the need for reference scans. The estimates are used to scale the amplitude of task-related fMRI responses, to account for vascular differences. The VasA maps compared well to cerebrovascular reactivity (CVR) maps and cerebral blood volume maps based on vascular space occupancy (VASO) measurements in four volunteers, speaking to the physiological vascular basis of VasA. VasA was validated in a wide variety of tasks in 138 volunteers. VasA increased t-scores by up to 30% in specific brain areas such as the visual cortex. The number of activated voxels was increased by up to 200% in brain areas such as the orbital frontal cortex while still controlling the nominal false-positive rate. VasA fMRI outperformed previously proposed rescaling approaches based on resting-state fMRI data and can be readily applied to any task-related fMRI data set, even retrospectively.