INTRODUCTION
Spatial normalization is an important step for image processing and quantification of regional brain perfusion values using arterial spin labeling (ASL) MRI and is typically performed via high-resolution structural scans. Structural segmentation and/or registration is complicated when gray-white matter T1w contrast is low and changing in early phases of myelination in newborns. This is complicated even further in children with craniosynostosis, whose skulls are deformed in various nonlinear ways. ^1-3 Both situations are present in scans of children with craniosynostosis. Craniosynostosis is a condition where skull sutures close prematurely and the decision for surgical treatment in the first years of life is supported by brain imaging.^4-7 In this study, we investigate if the ASL CBF image contrast can be directly used for spatial normalization, in both healthy controls and trigonocephaly patients - a non-syndromic type of craniosynostosis.

METHODS
Between 2018 and 2020, 36 trigonocephaly patients (median age 6 months, IQR 4, 11 females) and 16 healthy controls (median age 10 months, IQR 66, 10 females) underwent brain MRI with ASL to measure cerebral perfusion. Image processing and evaluation was performed with ExploreASL.^8-10 After aligning, motion correcting, and quantifying the ASL images [ref xASL], we compared four registration approaches: 1) Registration via segmentation and spatial normalization of the T1w images (regT1w); and a direct registration of ASL to MNI using 2) rigid-body (regASLrigid), 3) affine (regASLaffine), or 4) non-linear Direct Cosine Transform (DCT) transformation (regASLdct). For regT1, the T1w image was segmented and registered using SPM12 with a 1-year-old infants template (NITRC, version V2.3).¹¹ The M0 and T1w images were rigidly aligned and a joint transformation from ASL native space to MNI common space was applied. For the regASLrigid, regASLaffine, regASLdct, a pseudo-CBF image was constructed in the MNI space by assigning GM and WM CBF voxels of 60 and 20 mL/min/100 g, respectively. The native ASL CBF image was then registered with the pseudo-CBF image using rigid-body (2), affine (3), and DCT transformations (4), respectively. Registration performance was expressed by the Tanimoto coefficient (TC) and compared quantitatively by a linear mixed model.

RESULTS
For the total cohort, regASLdct showed the highest TC score (Figure 1). Registration performance did not differ between registration types in patients and controls (Figure 2). For both patients and controls, regASLdct showed the highest registration performance. RegASLaffine outperformed regT1 (TC improvement of 0.03, p-value < 0.05) and regASLdct outperformed regT1 (TC improvement of 0.04 p-value < 0.05). The interaction plot (Figure 3) shows that regASLdct and regASLaffine have a higher TC in comparison to regASLrigid of regT1in both patients and controls. TC will decrease using the registration method regASLrigid and regT1 in patients.

DISCUSSION
We have shown that direct registration to MNI space using ASL CBF as image contrast outperforms spatial normalization based on T1w segmentation in both patients and controls. The non-linear registration outperformed both rigid and affine registration. While better results were shown for the control group, the difference between regT1 and regASL was even higher in patients showing the increasing importance of this method in a cohort with poor GM-WM T1w contrast and large skull deformations. One main limitation is that we tested whole-brain alignment only. A more thorough validation is needed to demonstrate if this registration method allows regional CBF evaluation in MNI space using gray and white matter masks. Many ASL studies in neonates, those two included, rely on manual ROI delineation.^{12, 13} Using the proposed method to automate ASL analysis gives tools to delineate the neuroanatomical information leading to a higher reproducibility of results. For trigonocephaly patients it is therefore beneficial to use regASLdct.

CONCLUSION
In conclusion, the ASL CBF contrast is a viable spatial normalization alternative if structural images are not available or have poor contrast. The results of this study may be an important step for the feasibility of future large pediatric ASL studies.

Acknowledgements

No acknowledgement found.