Tractography in the fetal brain with correction of fetal and maternal motion using model-based slice to volume registration

Introduction Diffusion tensor imaging (DTI) of the fetal brain is attracting increasing interest [1-3] with recent papers starting to present tractography results [4]. To overcome maternal motion, many studies have been performed using breath-hold methods, although this severely reduces the amount of available data. However, even in breath-held acquisitions, the fetus can move substantially, so that individual diffusion weighted (DW) images acquired at the same geometrical slice position can be anatomically inconsistent. This problem was addressed by Jiang et al. [5], who extended the Snapshot to Volume Reconstruction (SVR) technique [6, 7] for application to DTI. The SVR approach works on single shot image slices with image registration to determine their correct location in an anatomical space, and then uses scattered data interpolation to reconstruct self consistent high resolution volume images. For snap-shot images of consistent contrast the registration process can employ cost functions like cross correlation (CC) and this has been shown to position slices in anatomical space with sub-voxel accuracy. In the case of DW images, the contrast changes with sensitization direction, so a more general cost function has to be used. Using this approach, high quality apparent diffusion maps of the fetal brain were made, but fractional anisotrophy (FA) maps proved less robust, and tractography was not demonstrated. In this work, we improved the DTI-SVR algorithm using more robust model-driven registrations, which improved the final reconstructed images and allowed white matter tracts to be obtained using a standard tracking approach.