Imaging Poster Presentation

P0589 - Implications of registration of white matter structures for MS connectome analysis (ID 1945)

Speakers
  • J. Garber
Authors
  • J. Garber
  • M. Barnett
  • F. Calamante
  • C. Wang
Presentation Number
P0589
Presentation Topic
Imaging

Abstract

Background

To assess the connectome disruption in patients with multiple sclerosis (MS), methods have been developed that utilise connectomes derived from healthy controls (HC) which require the accurate registration of white matter structures from patients with MS onto HCs whole brain streamline tractograms.

MRI T1 based image registration may be inaccurately mapped to HC derived tractograms due to the lack of information within the relatively homogenous T1 signal of white matter.

Fibre orientation distributions (FODs), created by the constrained spherical deconvolution (CSD) model of white matter diffusion data, have the potential to improve white matter registration as it holds information of all fibre directions for each voxel.

Objectives

Here we compare the accuracy and utility of FOD-based, T1-based and FA-based registration of white matter tracts in MS.

Methods

10 MS patients and 10 age matched healthy controls underwent 3T MRI scanning including 1mm 3D T1 and single shell diffusion weighted imaging. FOD maps were created using CSD algorithm designed for single shell data. Tensor and FA maps were created.

Non-linear registration was undertaken directly from each patient to each control (ie. 100 warps) with ANTs using T1 and FA images, and through an MNI template intermediary.

Registration was performed directly from each patient to controls using FOD maps using MRtrix3 FOD-based registration, as well as through a custom population FOD template.

Each registration method was assessed in transforming three white matter tracts, corticospinal tracts (CST), anterior thalamic radiations (ATR) and optic radiations (OR), from patients to controls. The resultant transformed segmentations from each registration method was compared to the control segmentation by calculating its Dice coefficient.

Further to this each method was assessed using a tract segmentation that had the cortical ribbon and juxtacortical tract removed.

Statistically significant differences were assessed by non-parametric Kruskal-Wallis test with Dunn’s post hoc testing.

Results

For combination of all tracts, the highest Dice coefficients were with direct FA (median = 0.727, IQR 0.06215) and direct T1 (median = 0.72185, IQR 0.056525) with no significant difference found.

For combination of all cropped tracts, the highest Dice coefficients were with FOD population template (median = 0.7673, IQR 0.0468), direct FOD (median = 0.76565, IQR 0.050175) and direct FA registration (median = 0.7626, IQR 0.060025) with no significant difference found.

When utilising an intermediary template, both T1 and FA based methods performed worse, whereas the FOD population template performed similarly.

Conclusions

FA and T1 based registration outperformed FOD based, despite more white matter information. This was driven by poorer juxtacortical registration in the FOD based method. This is important in the analysis of MS due to the high prevalence of juxtacortical lesional pathology.

Collapse