Introduction

The signal measured in a blood oxygenation level-dependent (BOLD) contrast fMRI experiment reflects changing neural activity through its coupling to local cerebral blood flow (CBF). However, the representation of neural activity is modulated by local vascular properties. These include the BOLD-sensitive cerebral blood volume (CBV) and cerebral vascular reactivity (CVR), which is the relative increase in CBF in response to a vasodilatory stimulus. CVR can be probed using carbon dioxide (CO₂) manipulations through, for example, breath-holding. In a breath-hold (BH) BOLD fMRI experiment, parametric maps of CVR are calculated based on the measured trace of exhaled CO₂.¹ However, this approach quantifies CVR in terms of voxel-wise BOLD signal rather than relative change in CBF, with its own dependence on CBV, and also relies on accurate quantification of end-tidal CO₂, which can be experimentally challenging. Here, we investigate an alternative approach to account for variability in the local BOLD response using a BOLD-sensitive relative CBV (rCBV) marker. The rCBV marker represents the ratio of the voxel-wise BH-induced BOLD signal change to that induced in the superior sagittal sinus (SSS), where CBV approaches 1. Here, we compare the use of a BH BOLD-CVR covariate with our proposed rCBV-weighted covariate in accounting for inter-subject variability in the BOLD fMRI response to a visuomotor task.

Methods

Twenty-seven healthy subjects (38.1±11.0 years old, fifteen females) underwent MRI on a GE Signa HDx 3 T system (GE Medical Systems, WI, US) using an 8-channel head coil. The MRI protocol included structural T₁-weighted, stimulus-evoked and BH BOLD fMRI, and B₀ inhomogeneity field map acquisitions with the parameters shown in Table 1.

During BOLD fMRI, subjects performed a serial reaction time (SRT) task in which the position of a visual stimulus was localised by pressing the corresponding key on a four-key keypad.² The task comprised Sequence (16 s), Random (16 s) and Rest blocks (15 s). Sequence blocks included four repeats of eight stimuli. One Random block was presented every three Sequence blocks. Rest blocks presented a static image matching the task.

During BH-BOLD fMRI (total duration=250 s), subjects were instructed to perform a BH consisting of four 15-s BH repeats interleaved with 35-s recovery periods. Continuous CO₂ partial pressure signals were measured in the expired air via a nasal cannula from which end-tidal CO₂ partial pressure (PetCO₂) was estimated.

BH-BOLD fMRI images were motion-corrected using MCFLIRT,³ expressed as fractional BOLD changes relative to their temporal mean, high-pass (60 s) gaussian filtered and aligned with the MNI template using FNIRT.³ CVR maps expressed as BOLD signal percent change per unit of PetCO₂ were then calculated using the General Linear Model framework using the subject-wise PetCO₂ traces as a regressor.

To provide a marker of local rCBV, a region of interest (ROI) was manually delineated in the posterior SSS based on the T₁-weighted image and aligned with the BH fMRI image using FLIRT.³ Grey matter segmentations were obtained from the T₁-weighted images using FAST⁴. Parametric maps of rCBV were calculated as follows. To mitigate partial volume effects in the SSS, the voxel-wise BH-BOLD signal within the SSS was regressed against the mean grey matter BH-BOLD signal. Based on this analysis, the average BH-BOLD signal (SSS_BH-BOLD) was calculated from the high covarying voxels (≥90^th percentile). Relative CBV maps were calculated by regressing the whole-brain voxel-wise BH-BOLD signal change against SSS_BOLD-BH.

Task BOLD fMRI images were analysed using FEAT.^5,6 Subject-level analyses were performed by modelling the BOLD signal with a block task regressor plus its temporal derivative.⁷ Group-level analyses were performed using three models: the first modelled the task response; the second (SRT+covrCBV) and third (SRT+covCVR) modelled the task response and added the individual rCBV or CVR maps as voxel-dependent covariates. To assess the extent of brain activation in the motor cortex (precentral gyrus), for each model, group Z-score statistical maps were compared at increasing thresholds (Z=3, 3.5, 4, 4.5 and 5).

Results

Figure 1 shows the mean and standard deviation (SD) of the CO₂ traces (A) and SSS_BH-BOLD signals (B) across subjects during the repeated breath-holding. Both group-level corrections for vascular covariates (SRT+covrCBV and SRT+covCVR) increased the significance of activation to the task in the motor cortex at superior levels, where the largest increase was observed in SRT+covrCBV (Figure 2A). Moreover, the rCBV covariate showed a spatially larger and stronger correlation with the task response (Figure 2B). Finally, both SRT+covrCBV and SRT+covCVR resulted in larger areas of activation compared to SRT (Figure 3A), where SRT+covrCBV had the largest increase reflecting the larger area of association with the task response (Figure 3B).

Discussion and Conclusion

The improved sensitivity of group-level fMRI when using SRT+covrCBV and SRT+covCVR agreed with previous results demonstrating that correction for BH responsiveness or vascular reactivity increases the spatial extent and statistical significance of detected activation.¹ The larger improvement in sensitivity observed when using SRT+covrCBV compared to SRT+covCVR suggests that an SSS-derived CBV marker, which does not require recording end-tidal CO₂, may offer a more robust way to account for local vascular properties than BOLD-CVR measurements relying on CO₂ recordings. This finding may reflect the effect of inter-subject variation in end-tidal CO₂ recordings, which can be heavily influenced by erroneous readings in practice.

Acknowledgements

The study was supported by the MS Society UK. This work was partially conducted under the framework of the Departments of Excellence 2018–2022 initiative of the Italian Ministry of Education, University and Research for the Department of Neuroscience, Imaging and Clinical Sciences (DNISC) of the University of Chieti-Pescara, Italy. This project was partially supported by the UK Engineering and Physical Sciences Research Council (EP/S025901/1).