A Symmetric Prior for the Regularisation of Elastic Deformations: Improved anatomical plausibility in nonlinear image registration

Frederik J Lange,John Ashburner,Stephen M Smith,Jesper L.R Andersson

doi:10.1016/j.neuroimage.2020.116962

Frederik J Lange, John Ashburner + Show 2 more

Open Access

https://doi.org/10.1016/j.neuroimage.2020.116962

Copy DOI

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 regularisation method capable of selectively driving solutions towards those which would be considered anatomically plausible by penalising unlikely lineal, areal and volumetric deformations. This penalty is symmetric in the sense that geometric expansions and contractions are penalised equally, which encourages inverse-consistency. We demonstrate that this method is able to significantly reduce local volume changes and shape distortions compared to state-of-the-art elastic (FNIRT) and plastic (ANTs) registration frameworks. Crucially, this is achieved whilst simultaneously matching or exceeding the registration quality of these methods, as measured by overlap scores of labelled cortical regions. Extensive leveraging of GPU parallelisation has allowed us to solve this highly computationally intensive optimisation problem while maintaining reasonable run times of under half an hour.

Highlights

Nonlinear registration is commonly used in neuroimaging to deform images of individual brains into some common space
One which is separable over warp parameters where calculations for one warp parameter can be performed and stored independently by one GPU thread. Following this line of reasoning, we may divide the Hessian calculation of Equation (20) into portions which are constant, separable over samples, and separable over warp parameters. As it is this process which is central to how using the SPRED penalty is made tractable, we provide an intuitive understanding of what implementing the parallelisation of equation (19) looks like in practice by considering a simple 2D example
We have demonstrated that our SPRED penalty is capable of matching the registration accuracy of the most well established largedeformation (ANTs-CC) framework available today

Summary

Introduction

Nonlinear registration is commonly used in neuroimaging to deform images of individual brains into some common space (normalising both size and shape). Often the registration is based on structural, e.g., T1weighted, images. The resulting warp is subsequently applied to both structural images as well as images depicting some function of interest such as BOLD or diffusion-derived connectivity. The rationale behind this is typically to facilitate statistical analysis of those functional data across subjects and populations. An important distinction in this context is between volume- and surface-based methods. The former attempt to find the inter-subject mappings in the original (3D) space. In this paper we present a novel volume based method

Objectives

Methods

Results

Discussion

Conclusion