To apply partial volume correction to ASL data from patients with MS to improve the accuracy of grey matter (GM) CBF estimates and facilitate the measurement of perfusion changes as a biomarker of treatment response.ASL is a low resolution imaging modality known to suffer from the partial volume (PV) effect, leading to an underestimation of GM perfusion. The PV effect has two principle causes: blurring from the point spread function (PSF) in the slice direction, due to the long echo train employed, and inadequate sampling resolution due to the need for large voxels to achieve sufficient SNR¹. Both these effects may act as confounders for the measurement of GM CBF abnormalities. A decrease in GM perfusion could reflect neuronal loss or metabolic dysfunction; PV correction techniques allow decoupling of structure and function. It is hypothesized that reduced GM perfusion after PV correction may serve as a biomarker of reversible neuronal dysfunction prior to substantive tissue loss². In this work, we present the first application of a complete PV correction solution for ASL to a cohort of MS patients.Six RRMS patients (5F, 1M aged 24-56Y, mean=43.5Y, EDSS scores ranged from 1.0 to 3.5; mean=2.3) were imaged on a GE MR750 3T scanner with 8 channel head coil. Acquisitions included: T2 FLAIR (TE/TR=162/8000ms, TI=2181ms, matrix=512x512x480, resolution=0.47x0.47x0.6mm³) for lesion characterization and T1-weighted anatomical IR-SPGR for tissue concentration (TE/TR=2.8/7.1ms, TI=450ms, FA=12°, matrix 512x512x248, resolution=0.47x0.47x0.7mm³). The ASL sequence was pCASL labeling with 8-arm 3D stack-of-spirals read-out (inflow time=1525ms, bolus length=1525ms, matrix 128x128x32, resolution=1.88x1.88x5mm³). MS lesions were manually segmented on the FLAIR image using Jim³ software to create lesion masks. FSL's⁴ lesion infilling was used to in-paint the lesions on the anatomical image, prior to segmentation into GM, WM, CSF components using FSL FAST. The segmentations were downsampled to perfusion space using FSL’s flirt and applywarp functions to ensure the PV maps within each ASL voxel represented the average PV estimate across the high-resolution region⁵. The 1D PSF in the slice direction was estimated using the Extended Phase Graph algorithm for the difference and proton density images⁶. These images were both deblurred using the calculated PSF with a Richardson-Lucy deconvolution algorithm, in order to restore signal to the voxel of origin. The images were PV corrected using a 3 x 3 x 3 kernel in a linear regression algorithm before combining to produce separate maps of GM and WM CBF⁷. Regions of interest (ROI) were created from a Freesurfer parcellation⁸ which was transformed to ASL space in the same manner as the PV estimates and the mean GM or WM CBF was calculated for each of these.The 1D PSF in the slice direction was estimated to be 1.6 and 1.3 voxels at full width half maximum for the ASL difference and proton density images respectively. ASL and proton density images for patient 002NN before and after deblurring are shown in fig. 1,2. Middle slice CBF and PV-corrected GM and WM CBF are shown in fig. 3. Fig. 4 shows the mean GM and WM CBF with and without deblurring for all 6 subjects, which shows a broad trend of increased GM and reduced WM CBF. There was a consistent increase in GM CBF with the inclusion of deblurring compared to without, ranging from 5-12%, mean 8%. In 4 out of 6 subjects, the WM CBF decreased by -2 to -11%, mean -5%. For subject 004SS, there was no change, and a 1% increase for 006JF. Fig. 5 shows the mean regional CBF and GM CBF post-correction, as well as the WM CBF.

We observed significant through-plane blurring in 3D ASL images, even in highly-segmented acquisitions. The combination of deblurring and PV-correction produces an increase in mean GM CBF and a decrease in WM CBF over and above that produced by either correction alone. GM CBF values are modified to a greater extent than WM CBF, due to the cortex being significantly thinner than the width of the PSFs and GM voxels therefore suffering more signal attenuation. WM CBF values for these MS patients were higher than usually observed in healthy cohorts. Elevated WM CBF preceding lesion development has previously been observed in MS patients, which suggests that CBF may be sensitive to inflammation and serve as a early stage biomarker of lesion formation^9,10.

More accurate estimations of GM and WM CBF can be obtained by applying PV-correction solutions that correct for both PSF blurring and tissue heterogeneity. This will be particularly important for SPMS patients, who exhibit tissue atrophy over the disease course.