Three‐dimensional motion corrected sensitivity encoding reconstruction for multi‐shot multi‐slice MRI: Application to neonatal brain imaging
As for brain imaging, although non‐rigid motion 1 can occur due to pulsatile motion 2 and localized movements can arise such as from eyeball motion 3, the current clinical paradigm relies on the subjects holding their head still enough for the duration of the acquisition to avoid gross motion artefacts. However, in many brain studies, such as for those subjects who have difficulty remaining still enough 4 and in ultra‐high resolution imaging applications 5, large or small motion inconsistencies become a limiting factor, so that appropriate acquisition or reconstruction strategies have to be put in place.
Usual sequences in structural brain studies are based on multi‐shot methods, where a fraction of the k‐space is acquired after a single radio frequency excitation or shot. This type of sampling, particularly when combined with slice selective excitation, provides flexibility to achieve the desired contrast while simultaneously balancing the signal to noise ratio, image resolution, and scanning time requirements. A common operating regime is to use an interleaved scheme in which several slices are sequentially excited within a given repeat time (TR), that way permitting efficient sampling for large TR's 6. In this setting, changes in the head position among different shots and slices, which may be acquired very distant in time, will provoke degradation of individual slices and inconsistencies in volumetric information.
Head motion estimation and/or retrospective motion‐compensated reconstruction in multi‐shot sequences has been extensively studied in the past 7. However, to the best of our knowledge, none of the proposed reconstruction methods can be used to correct for through‐plane motion in multi‐slice acquisitions 1. On the other hand, image‐based alignment methods have been introduced to assemble the information coming from snapshot acquisitions, where individual slices are acquired fast enough to approximately freeze movement, into self‐consistent three‐dimensional (3D) representations of the imaged structures. This has been performed either by matching the structures along slice intersections 16 or by slice to volume registration 18.
In Ref. 15, we proposed a framework for rigid body motion estimation and motion‐compensated reconstruction in volumetric multi‐shot sequences using parallel imaging that is grounded on the sensitivity encoding (SENSE) reconstruction paradigm 21. The method is fully data driven, using a common functional to estimate the motion and the image. Moreover, it is suitable for a wide range of settings as it does not require external sensors, navigators, or particular samplings. In that work, we provided an empirical characterization of the conditions for which, in the absence of noise, fully rigid motion corrected reconstructions are still possible in terms of the amount of motion, number of shots, encoding trajectories, and availability of prior information. Here, the framework is extended to multi‐slice sequences, thus fusing the multi‐shot and aligned snapshot families of motion compensation methods. Moreover, differently from most image‐based alignment methods, our procedure aims to correct for within‐plane and moderate through‐plane motion requiring only a single slice orientation. To support this aim, on the acquisition side, slices are sampled in an overlapped manner (where the slice separation is maintained while the slice thickness is increased for same total acquisition time).