Introduction

Cerebral small vessel disease (SVD) encompasses all pathological processes affecting the small vessels of the brain, being a major vascular contributor to dementia and age-related cognitive decline. White matter hyperintensities (WMH) detected on structural MRI are a hallmark of SVD, but they reflect macroscopic lesions secondary to the disease. This has motivated the search for biomarkers that more directly probe vascular integrity, including perfusion, which may be more sensitive to disease progression. To date only a few studies have investigated perfusion measures as potential SVD biomarkers using arterial spin labeling (ASL) MRI, with somehow discordant results¹; and the predictive power of cognitive impairment remains to be explored. Here, we investigate the potential of ASL perfusion imaging to provide sensitive biomarkers of SVD, by evaluating their predictive power of cognitive impairment in processing speed tasks².

Methods

17 patients with SVD patients (50 ± 9 yrs) were studied on a 3T Siemens Verio system. Structural images included T₁-weigthed MPRAGE (1 mm isotropic) and T₂-weighted FLAIR (0.7 × 0.7 × 3.3 mm³). Image segmentation was performed and regions of interest (ROIs) were extracted for gray matter (GM), normal appearing white matter (NAWM), WMH and cerebrospinal fluid (CSF). The normalized brain volume (nBV) and normalized WMH lesion load (nWMHLL) were estimated. ASL data were acquired using a PASL PICORE-Q2TIPS sequence, with 2D multi-slice GE-EPI readout (TR/TE = 2500/11 ms, 28 contiguous slices, 3.5 × 3.5 × 5.0 mm³ resolution). Eleven inversion times (TI) were sampled, between 400 and 2400 ms, in steps of 200 ms, and 8 label/control repetitions were performed for each TI. Q2TIPS saturation limited the labeling bolus width to τ = 750 ms.
All data were analyzed using FSL³ and MATLAB. ASL data pre-processing steps included: motion correction; control magnetization averaging at each TI (control time series); control-label magnetization subtraction and averaging at each TI (difference time series); and correction for off-resonance effects caused by imperfect inversion slice profile in 2D multi-slice imaging. Maps of equilibrium magnetization of tissue, M_0t, were obtained by fitting a saturation-recovery curve to the control time series. An extended kinetic model with intravascular arterial compartment was fitted to the difference time series using BASIL⁴, with T_1a = 1.65 s, T_1t = 1.3 s, τ = 750 ms, in order to estimate cerebral blood flow (CBF) and bolus arrival time (BAT). Calibration was then performed voxelwise, using the equilibrium magnetization of arterial blood (M_0a) map obtained by smoothing M_0t (FWHM = 10.5 mm) and dividing by the brain average water partition coefficient between blood and tissue, λ=0.9. Both CBF and BAT were averaged across GM and NAWM.
The patients’ cognitive function was evaluated using a battery of neuropsychological tests, including assessment of processing speed using the Trail Making Test - Part A⁵. The Pearson correlation was computed between the processing speed scores and each of the four perfusion metrics considered (CBF and BAT in GM and NAWM), as well as the following covariates: age, nBV and nWMHLL. Multiple linear regression was performed by best subset regression, using R (https://www.r-project.org/), including the perfusion metrics and covariates.

Results

An illustrative example of CBF and BAT maps obtained for one patient is displayed in Fig. 1. The average CBF and BAT in GM/NAWM were 40.4±8.2/25.9±5.5 and 0.74±0.14/0.87±0.10, respectively. Pearson correlation analysis between the processing speed scores and each of the four metrics considered (CBF and BAT in GM and NAWM), as well as the covariates, is shown in Fig. 2. Only CBF in GM was significantly correlated with processing speed scores (p=0.008). Fig. 3 displays the subsets of regressors that best explain the processing speed scores. The best subset model includes as predictors: age (p = 0.030), CBF in GM (p = 0.001), and BAT in GM (p = 0.090). Although only CBF in GM exhibited a significant correlation with processing speed, when considering multiple regression, age and BAT in GM further contribute to explaining the cognitive scores, yielding a total explained variance of ~52%. Nevertheless, CBF in GM is the only regressor that is consistently selected by all subsets.

Conclusions

Our results provide the first evidence that CBF measurements obtained by non-invasive ASL perfusion imaging have the potential to predict cognitive decline in SVD, and therefore further support the hypothesis that it may provide sensitive SVD biomarkers. In particular, CBF in GM is significantly correlated with processing speed, the cognitive domain that has more specifically been associated with SVD². Previous work has reported relationships of other advanced MRI parameters with processing speed in SVD patients, including microstructural changes measured by diffusion-weighted imaging (DWI)⁶ and arterial pulsatility measured by the amplitude of low frequency fluctuations using resting-state BOLD-fMRI⁷. Future studies should investigate the relationship between different MRI parameters, namely perfusion, arterial pulsatility and microstructure, among others. Our preliminary results should be expanded by testing larger patient cohorts, and in longitudinal studies, in order to further investigate the potential of ASL to provide SVD biomarkers.

Acknowledgements

This work was funded by FCT grants PD/BD/135114/2017, PTDC/BBB-IMG/2137/2012, and UID/EEA/50009/2019.