Partial volume (PV) effects are a well-recognized challenge in arterial spin labeling (ASL), and can have profound effects on the accuracy of the ASL-derived cerebral blood flow (CBF) maps^1,2,3. Several algorithms exist to correct for these errors^4,5,6,7. A limitation of the current strategies is the use of PV-maps from segmented T1-weighted images (pGM), which differ geometrically from the ASL acquisition⁸. This study aims to quantify the impact of common errors in PV maps caused by mismatches in geometric distortion and spatial resolution between the low-resolution ASL and high-resolution T1-weighted acquisitions. Furthermore, we show how PV-correction relate to the commonly used calculation of the mean gray matter (GM) CBF within a GM mask, which is essentially a form of PV-correction. Lastly, the CBF maps were smoothed to simulate a larger acquisition point-spread-function (PSF) and sensitivity to motion.Three methods to estimate the mean GM CBF (within a pGM-mask obtained by thresholding the pGM map) were compared: 1) GM-Mask: mean CBF within a pGM mask; 2) GM-Weighted: within the pGM mask, the mean CBF is divided by the mean pGM; 3) PVEC: within the pGM mask, the mean of the PV-corrected GM-CBF values (the Asllani’s method with a 3x3x1 kernel³). Both 2D-EPI pseudo-continuous ASL⁹ and T1-weighted (T1w) images were acquired for a young healthy volunteer (Table 1). The ASL and M0 images were motion corrected and the perfusion-weighted difference (PWI) was calculated. The PWI and M0 were upsampled to the T1w space and the T1w volume was segmented into GM/WM using SPM12. To estimate the deformations between ASL and T1w volumes they were co-registered using four different methods: A) T1w-Rigid: rigid transformation (between T1w and M0); B) GM-Rigid: rigid transformation (between PWI-GM); C) Nonlin: SPM’s affine transformation followed by nonlinear transformation¹⁰ (between PWI-GM); D) DARTEL: creating a DARTEL¹¹ template from the PWI and GM image (Figure 1). Each of these transformations was separately applied to the original GM/WM maps to create four deformed volumes, which were subsequently used to simulate the deformation between the ASL-image and T1-based PV-maps. The results were downsampled using an anisotropic Gaussian kernel identical to the acquisition resolution. The downsampled GM/WM maps from method D (DARTEL) were used to generate a simulated CBF (sCBF) image by assuming a uniform GM-CBF of 80 ml/min/100g, GM/WM ratio of 3, and SD-4 Gaussian noise. The mean CBF in sCBF was evaluated using the four different downsampled PV-maps for the three PV-correction methods. A second set of sCBF maps was created from method D by increasing FWHM from 3x3x7mm³ to 7x7x7mm³. The mean CBF was evaluated using PV-maps with FWHM 3x3x7mm³ to simulate the effect of having sharp PV-maps and blurred CBF (Figure 2). The globus pallidus and thalamus were excluded from the analysis due to T1-segmentation issues.To assess the differences across the 4 methods, for each PV-correction algorithm, we computed: the mean relative voxel-wise difference between all pGMs and the ground-truth pGM (D), the relative error in the mean GM-CBF, and the mean relative error in the GM-CBF calculated on a local neighborhood of 15x15x7 mm³. Deformation methods A-D (Figure 3), and FWHMs from 3x3x7mm³ to 7x7x7mm³ for the method D (Figure 4) were compared. For a typical threshold (pGM>0.7), the mean pixel-wise error in pGM-maps was between 10-15%. Without PV-correction, global and local CBF errors were 16-21% for all methods, including D. With both PV-corrections, the local error was under 8%. A local error of 4% with perfect PV-maps was achieved for GM-Weighted PV-correction. The global CBF error was <5%, however, this is interpreted carefully as a uniform GM-CBF value was used. The CBF-map blurring caused 3-13% error in pGM. This added 2-10% to the error in the mean global/local CBF calculation without PV-correction. This effect persisted with PV-correction regardless of deformation for the global error. However, the blurring did not affect the local CBF error for the deformations methods A-C with PV-correction.Without any PV-correction, the ASL-T1w deformations had only a small effect on the calculated mean CBF values, however, blurring further underestimated the CBF. Blurring of data caused by acquisition, motion or post-processing thus presents an important factor for the mean CBF evaluation. This needs to be taken into account in multi-center ASL studies, especially for 3D sequences with large PSF and in patients that are prone to move. Even on a local scale, both PV-correction methods decreased the errors in mean CBF calculation. The use of PV-correction did always improve the results even in the presence of deformations and data blurring, despite a decrease of the accuracy of the PV-maps.