A Symmetric Prior for the Regularisation of Elastic Deformations: Improved Anatomical Plausibility in Nonlinear Image Registration

Abstract

Nonlinear registration is critical to many aspects of Neuroimaging research. It facilitates averaging and comparisons across multiple subjects, as well as reporting of data in a common anatomical frame of reference. It is, however, a fundamentally ill-posed problem, with many possible solutions which minimise a given dissimilarity metric equally well. We present a novel regularisation method that aims to selectively drive solutions towards those which would be considered anatomically plausible by penalising unlikely lineal, areal and volumetric deformations. In addition, our penalty is symmetric in the sense that geometric expansions and contractions are penalised equally, which encourages inverse-consistency. We demonstrate that our method is able to significantly reduce volume and shape distortions compared to state-of-theart elastic (FNIRT) and plastic (ANTs) registration frameworks. Crucially, this is achieved whilst matching or exceeding the registration quality of these methods, as measured by overlap scores of labelled cortical regions. Furthermore, extensive use of GPU parallelisation has allowed us to implement what is a highly computationally intensive optimisation strategy while maintaining * Corresponding Author reasonable run times of under half an hour.

Anatomical variability, multi-modal coordinate systems, and precision targeting in the marmoset brain

“…However, further validation of brain parcellations are needed in future studies by combining histology data (e.g., Majka et al, 2021), as well as functional connectivity data as has been done in humans (Glasser et al, 2016). This may require refinement of technologies in terms of spatial mapping of 2D-histology data into 3D neuroimaging data (Wang et al, 2020; Majka et al, 2021; Hayashi et al, 2021) and intersubject registration based on multi-modal data for cortical surfaces (Robinson et al, 2018) and brain volume (Lange et al, 2020).…”

Adapting the UK Biobank brain imaging protocol and analysis pipeline for the C-MORE multi-organ study of COVID-19 survivors

“…Data were corrected for gradient and EPI distortions and aligned with each other using linear alignment (Jenkinson and Smith, 2001) (between modalities, within-subject). Subjects were then aligned into standard template space (MNI152) using FNIRT nonlinear alignment driven by T 1 -weighted images (Andersson et al, 2019) and aligned to a study-specific template using MMORF nonlinear alignment driven by both the T 1 and T 2 -FLAIR images (Lange et al, 2020).…”

“…Rigid and affine registrations used to align T 1 and T 2 -FLAIR images within subjects, and T 1 images between subjects, respectively, were performed with FLIRT (Jenkinson and Smith, 2001). Nonlinear registrations between each subject and the template were performed with MultiModal Registration Framework (MMORF) (Lange et al, 2020), which allowed the simultaneous registration of both modalities, and brain and non-brain tissue. In the second step, the average templates served as a reference space for normalisation and the enhanced SNR enabled clear visualisation of the left and right olfactory bulbs (OB), which were manually segmented.…”

Reliability of multi-modal MRI-derived brain phenotypes for multi-site assessment of neuropsychiatric complications of SARS-CoV-2 infection

“…Sub-cortical volumes were estimated utilizing population priors on shape and intensity variation across subjects (Patenaude et al, 2011). Using an additional non-linear registration procedure, regional volumes of the olfactory bulbs were estimated using T1-weighted, T2-FLAIR and dMRI data, and a template derived from over 700 UKB individuals (Arthofer et al, 2021; Griffanti et al, 2021; Lange et al, 2020).…”