INTRODUCTION

Cerebrovascular reactivity (CVR), an index of the dilatory function of cerebral blood vessels, is typically assessed with a carbon dioxide (CO₂) stimulus combined with BOLD fMRI¹. Recently, resting-state BOLD fMRI has been shown capable of generating CVR maps without the need for a gas (CO₂ inhalation) or drug (acetazolamide administration) challenge^2,3, indicating a potential for broader CVR applications in neuroimaging studies. However, compared with CO₂-CVR, resting-state CVR (RS-CVR) is known to suffer from lower SNR and prior RS-CVR studies have primarily been performed at a spatial resolution of 3-4mm³ voxel sizes. As the fMRI field marches towards higher spatial resolutions, it remains unknown whether RS-CVR can also be obtained at high resolution without a major degradation in image quality. Therefore, the goal of the present study is to investigate the effect of spatial resolution of the resting-state BOLD sequence on CVR mapping. To achieve this goal, resting-state BOLD scans were collected with three resolutions at decreasing voxel size, and the resulting CVR maps were compared against maps obtained with the conventional CO₂-inhalation method.

METHODS

Image Acquisition: Data were collected from eight participants (3F/5M, 26.5±4.9 years) using a 3 Tesla Prisma Siemens scanner with a 32-channel head coil. Four BOLD scans were acquired in each subject: three ten-minute resting-state scans with different resolutions (2mm, 2.4mm, and 3mm isotropic resolution), and one five-minute scan (2mm isotropic resolution) with CO₂ challenge (two 50s periods of 5% CO₂ inhalation interleaved with room air inhalation). Other BOLD imaging parameters were: TR/TE/FA=720ms/37ms/52º, multiband factor=8; for the 2mm, 2.4mm and 3mm resolutions, FOV=208*208*144 mm³, 210*210*155 mm³ and 210*210*168 mm³, respectively. A high-resolution T1-MPRAGE sequence (1x1x1 mm³) was also acquired.
Image Analysis: The resting-state and CO₂-inhalation BOLD data were analyzed following previous literature⁴. For resting-state data, after motion correction and detrending, the BOLD images were temporally filtered to [0, 0.1164 Hz], and a whole-brain-averaged time course was calculated³. Then, using the whole-brain BOLD time course as the independent variable, motion vectors and linear drift as covariates, voxel-wise regression analysis was performed to calculate a CVR index map in %BOLD signal change, which was then normalized to the whole-brain average to yield a relative CVR map. For the CO₂-inhalation data, after motion correction, the BOLD images were smoothed by 0mm, 1.33mm and 2.24mm Gaussian kernels to match the intrinsic resolution of the 2mm, 2.4mm and 3mm resting-state scans, respectively. CVR maps were then generated for the three resolutions following the conventional method⁵. Briefly, the EtCO₂ time course was shifted to account for the lung-to-brain delay, and then this shifted EtCO₂ was used as the independent variable in a similar voxel-wise regression analysis, generating an absolute CVR map in %∆BOLD/mmHg CO₂ which was then normalized to the whole-brain average to yield a relative CVR map. Accuracy of RS-CVR mapping was measured using Pearson’s R², intraclass correlation (ICC) A1 (absolute agreement), and ICC-C1 (consistency) to compare the RS-CVR maps to the standard CO₂-inhalation CVR maps. Sensitivity was explored using Z-score maps from the voxel-wise analysis. Mean Z-scores of each resting-state scan of each subject were calculated and compared across difference resolutions.
Statistical analysis was performed using Kruskal-Wallis Non-parametric ANOVA with Dunn’s test and Holm correction as a post-hoc analysis.

RESULTS

Figure 1 shows the CVR maps obtained from each scan of each subject. Visual inspection suggested that good map quality is seen in all scans, with clear gray/white matter contrast. Lower resolution yielded images that are somewhat blurry but appear more stable. Figure 2 demonstrates the accuracy comparisons among the three resolutions. The 2mm resolution showed slightly lower correlation with the CO₂-inhalation CVR maps, whereas the 2.4mm resolution showed slightly higher correlation. Nonetheless, no significant differences between the resolutions in Pearson’s R² or ICC were found (p>0.089), suggesting comparable accuracy among different resolutions. We conducted the same accuracy analysis using BOLD data pre-smoothed with 4mm and 8mm kernels, and found similar results, with no significant differences between resolutions with different image smoothing.

Histograms of the mean Z-score distributions across all subjects for the three resolutions are shown in Figure 3a. Boxplots of the mean Z-scores for each subject are shown in Figure 3b. A significant difference between the resolutions was found, and post-hoc analysis showed a pattern of increasing mean Z-score with increasing voxel size, with the 2x2x2 mm³ and 2.4x2.4x2.4 mm³ resolutions being significantly lower than the 3x3x3 mm³ resolution (p=0.0007, p=0.03569 respectively). This is expected as larger voxel size should have higher SNR, and thus, yield higher sensitivity⁶.

DISCUSSION

In this study, we have investigated the feasibility of high-resolution CVR mapping based on resting-state fMRI data. Our results show that good CVR quality can be obtained at 2-3 mm³ voxel sizes. Within this resolution range, the higher resolutions gave more resolved information but the lower resolution yielded more robust maps. These findings suggest that the resting-state CVR mapping method can be applied at a similar resolution as functional connectivity mapping, which should broaden the use of CVR mapping in basic science and clinical applications.

Acknowledgements

No acknowledgement found.