43 Using a new technique of ASL | Trigonocephaly showed poor performance, even though a dedicated template for young children was used.13 Our normalization was complicated by the ongoing myelination in these young patients. Here, we were able to combine our previously developed CBF-contrastbased registration with a low-degree-of-freedom nonlinear component to improve the registration for the deformed skulls reaching a better registration.6 In this previous study of Mutsaerts et al., T1w images of elderly patients with frontotemporal dementia were spatially normalized to MNI and ASL was aligned with T1w images using the CBF-GM contrast. The joint transformation of ASL to T1w and T1w to MNI space was then used for spatial normalization of ASL to MNI. In the current study, we extended the previous work by aligning ASL to MNI directly using the CBF vs. GM contrast and investigating, also, the DCT transformation to assess the qualitative and quantitative benefit of this method in a pediatric patient cohort with skull deformations. Both regT1 and regASLrigid performed poorly on the TC and qualitatively. For the regT1 we attribute this to the relatively low WM-GM contrast of these images. The poor performance of regASLrigid is likely caused by the fact that brain size and shape differ with age. Also, it is challenging to register with MNI with only a rigid registration that preserves size. The regASLaffine and regASLdct registrations show a better alignment. RegASLaffine and regASLdct addresses both issues and therefore performed better in TC and on the qualitative rating. The advantage of regASLdct over regASLaffine is that it can adapt to the nonlinear deformations present in the trigonocephaly patients. Therefore, rigid and affine registrations followed by nonlinear deformations using a linear combination of three-dimensional DCT basis functions were considered sufficient for alignment of the individual data to the template.20 The quality of the pseudo-CBF image is essential for the registration. Typical values of CBF of 60 and 20 ml/min/100 g in GM and WM, respectively, were used. Even though these values are only rough approximations, and the population and individual values differ largely, they still capture the general differences in contrast between the tissue types, and this approximation was sufficient for the regASLrigid and regASLaffine registration. This, however, was not sufficient for the regASLdct registration with more degrees of freedom. To obtain good results, the regASLdct registration had to be performed in two iterations, where the regional CBF in GM and WM were first approximated from the ASL image after a rough alignment and used to construct a more realistic pseudo-CBF image in the second iteration. While the spatial normalization reached a good quality in most subjects, further improvement could be achieved in the future in large populations using population-specific templates. To address the relatively high deformation variability, the Cerebromatics approach could be used to create a template that covers a range of deformation types and severity. Subgroups other than trigonocephaly patients could also possibly be considered. To address more severe deformations, further deformation algorithms might need to be tested that provide potentially a better performance than the DCT registration.23 2
RkJQdWJsaXNoZXIy MTk4NDMw