example visualizations created by ExploreASL, for template of vascular territories141 in the odd rows and average ATT template142 in the even rows.
These are examples for a mean T1 (left), between-subject SD T1 (middle) and between-subject SNR T1 (right image == left image divided by the middle image). On the mean T1, it is clear that this is an elderly population with some atrophy and enlarged ventricles. The (between-subject) SD and SNR give an idea of the between-subject registration (residual variance between subjects). Hence why in the large WM regions the between-subject variance is low because misalignment between subjects will matter little due to the large size of the WM region. On the GM-WM border, any misalignment results in high variance because of the thin cortex.
These are the same examples for the pGM, showing again that misalignment has smallest penalty in larger regions. In case of GM, the largest size are the subcortical regions and the insula. Note that this analysis was performed with the older priors (poorer subcortical segmentation).
Same for the pWM, the smallest misalignment penalty is in the middle of the WM, and largest at the edges with GM. Note also the higher between-subject variation in the periventricular WMH areas.
Same for FLAIR images. These appear a bit more fuzzy than the T1 images, because their between-subject registration is also dependent on their registration with the T1. Note the WM lesions.
Same for WMH_SEGM images, but the mean WMH_SEGM projected on the mean FLAIR image.
Same for CBF images. Note the bit poorer readout resolution in this cohort. Nevertheless, still WM CBF and CSF (0) can be distinguished, which is a sign of good ASL quality. First, we want the GM/WM CBF contrast to be good, but then we can also look at the WM/CSF contrast, which will only be good in high quality ASL scans. With poor effective resolution, this will be a smear/blur, and with poor SNR, this will also look similar to the WM CBF, since the SNR was too poor to obtain good WM CBF. Note the lower SNR in the back, which could be ATT-related. Likewise, the variance in the subcortical structures is highest for the thalamus, which is mostly fed by the posterior circulation.
Same for the mean_control image. This provides an idea of the effective resolution, which should appear similar to the effective resolution of the CBF image (otherwise there may be specific perfusion-related artifacts). Note that the SNR of the mean_control (2D EPI image before subtraction) is good everywhere, it is not better in the region with high signal (CSF).
Same for the temporal SD image. Note that this shows an angiogram, showing that the largest perfusion fluctuations happen in the large arteries, similar to the variation of the perfusion fluctuation between-subjects (2nd image) or how "consistent" perfusion fluctuations are between subjects (3rd image) -> they are most consistent in the circle of Willis. The between-subject SD and SNR images are difficult to interpret though, since the quality of these images is low.
Same for the temporal SNR images, as discussed in the population module, for the use of the creation of population-based quality maps. Note the holes where the large cerebral vessels run, most notably the MCA. Note the contrast between WM and CSF, showing that we can in fact detect WM CBF. Note also how fuzzy the GM CBF is, as compared to the thin GM in the average pGM image above. This clearly shows the partial volume problem for ASL. Moreover, if images are compared between-subjects (as with statistical analyses), the images become even more fuzzy (as illustrated by the "SNR tSNR map" on the right, showing the "stability between-subjects" of the tSNR map ), probably due to residual registration errors.
Overlay of the population average CBF image (yellow) over the population average pGM image (red) illustrates the extent of the average geometric distortion.