Influence of realignment-induced interpolation errors on the estimation of cerebral blood flow using arterial spin labeling fMRI

Purpose/Introduction: The realignment transformation needed for motion correction in fMRI has been shown to have the adverse effect of smoothing the realigned images1. This effect is independent of the accuracy of the estimated motion parameters (it can occur even for accurately estimated motion parameters) and can, in ASL, cause gray-matter (GM) cerebral-blood-flow (CBF) underestimation compared to an acquisition without motion. Here, we investigated the smoothing of the realignment transformation in ASL by creating simulated CBF maps based on T1-weighted (T1w) images and motion parameters obtained from ASL images acquired on patients2.
Methods: ASL data from 66 children (age: 8-18y, median: 12.8y, 34 males) were obtained from the NOVICE study2. This dataset is characterized by a relatively high head motion often associated with scanning children. Images were acquired on Philips 3T Ingenia (pCASL: voxel-size 3x3x6.6mm3, 30 controls/label pairs; T1w: voxel-size 1x1x1mm3).
To simulate ASL time-series that reflect motion from real acquisition, the following processing steps were implemented: (1) T1w-image was co-registered to the first ASL volume and segmented into gray and white-matter posterior probability maps (pGM/pWM) using SPM12. (2) ASL images were realigned with respect to the first volume3 (average motion >1mm or >1° was considered 'high'). (3) The motion-parameters from step 2 were applied to pGM/pWM maps, and the maps were down-sampled to the ASL resolution. This resulted in 60 pGM/pWM volumes mimicking a real ASL acquisition (Fig. 1b). (4) The pGM/pWM maps were realigned to their first respective volume (Fig. 1d) and averaged (Fig. 2d).
Two sets of simulated CBF maps were created: (1) from the pGM/pWM corresponding to the first ASL volume (CBF-static), (2) from the realigned pGMs/pWMs described above (CBF-motion), assuming GM-CBF of 80 mL/min/100g and GM/WM CBF ratio of 3. Local and global GM-CBF values of CBF-motion were compared against the idealized CBF-static case.
Results: Figures 1d and 2d show ‘blurring’ of the realigned pGM in high movement cases. The local GM-CBF in CBF-motion was up to 10% lower than in CBF-static (Fig. 2e); negligible differences (<1%) were seen in relatively low movement cases (Fig. 2f). In all 'high-motion' participants, 4-7% lower global GM-CBF was observed (Fig. 3).
Discussion/Conclusion: Relatively high motion during ASL acquisitions appears to result in a GM-CBF underestimation of 4-7% globally and up to 10% locally. This finding can have important implications in studying patients who tend to move more than their healthy counterparts, which may lead to a GM-CBF underestimation in patients relative to controls.