Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2014 Oct 15;100:414-26.
doi: 10.1016/j.neuroimage.2014.05.069. Epub 2014 Jun 2.

MSM: A New Flexible Framework for Multimodal Surface Matching

Affiliations
Free PMC article

MSM: A New Flexible Framework for Multimodal Surface Matching

Emma C Robinson et al. Neuroimage. .
Free PMC article

Abstract

Surface-based cortical registration methods that are driven by geometrical features, such as folding, provide sub-optimal alignment of many functional areas due to variable correlation between cortical folding patterns and function. This has led to the proposal of new registration methods using features derived from functional and diffusion imaging. However, as yet there is no consensus over the best set of features for optimal alignment of brain function. In this paper we demonstrate the utility of a new Multimodal Surface Matching (MSM) algorithm capable of driving alignment using a wide variety of descriptors of brain architecture, function and connectivity. The versatility of the framework originates from adapting the discrete Markov Random Field (MRF) registration method to surface alignment. This has the benefit of being very flexible in the choice of a similarity measure and relatively insensitive to local minima. The method offers significant flexibility in the choice of feature set, and we demonstrate the advantages of this by performing registrations using univariate descriptors of surface curvature and myelination, multivariate feature sets derived from resting fMRI, and multimodal descriptors of surface curvature and myelination. We compare the results with two state of the art surface registration methods that use geometric features: FreeSurfer and Spherical Demons. In the future, the MSM technique will allow explorations into the best combinations of features and alignment strategies for inter-subject alignment of cortical functional areas for a wide range of neuroimaging data sets.

Keywords: Discrete optimisation; Functional alignment; Multimodal; Surface-based cortical registration.

Figures

Figure 1
Figure 1
Discrete optimisation for volumetric registration: A) a lower resolution control point grid (red) is shown over the moving image. B) a set of label points (blue/purple) defines the restricted deformations possible for each control point. The spacing of the candidate label points (blue/purple) is intermediate between the lower-resolution control point grid (red) and the voxel grid of the fixed image, F (grey). C) The Euclidean deformation of a single control point to the location of its optimal label point (only one control point is moved for illustrative purposes).
Figure 2
Figure 2
Discrete optimisation for spherical registration: A) The control point grid (red) is formed from a regular subdivision of an icosahedron. B) Regularly spaced label points are placed around the control point using a higher resolution icosahedron to define a sampling grid for the candidate label points (purple/blue points); this grid has lower resolution than the fixed (and moving) mesh (grey); C) Deformation of a control point to its optimal label point (only one control point is moved for illustrative purposes), using rotations about the centre of the sphere (not Euclidean vectors).
Figure 3
Figure 3
Results of univariate feature alignment using 25 subjects. The algorithms used are: FreeSurfer, Spherical Demons, MSM with low regularization (MSMlo) and MSM with high regularization (MSMhi). Panel A) show results of sulc-driven alignments: cross subject averages (first row) and variances (second row) across subjects. Panel B) shows curvature alignments. Panel C) shows myelin alignments. See text (section 5.2) for details of how the methods are run and initialised. Note that the cross-subject myelin average shows more structure at the positions of the frontal eye fields (yellow arrow) and intra-parietal sulcus (white arrow) for the MSM results.
Figure 4
Figure 4
Areal distortions from univariate alignment, across the four methods shown in figure 3: FreeSurfer, Spherical Demons, MSMlo, and MSMhi. Rows A to C correspond to panels A to C in the previous figure: registrations driven by Sulc, Curvature or Myelin. Areal distortion maps are averages across subjects, after being transformed to the population average target, and the absolute values of log2 of the area ratio is used (see text). FreeSurfer generates maps with localised distortion, particularly along gyral crowns, whereas MSMhi and Spherical Demons generate a much smoother pattern of mean distortions for Sulc- and Myelin-driven alignments. Note that MSMlo and MSMhi have substantially different distortions when driven by Sulc data, but very similar distortions for Curvature-driven or Myelin-driven cases, indicating that the differences in initialisation have little effect in these cases. Histograms (rightmost column) show all methods in each row, colour-coded as: black (FreeSurfer), blue (Spherical Demons), red (MSMlo), and green (MSMhi).
Figure 5
Figure 5
Task fMRI alignment driven by RSN features. The maps show group activation results from 196 subjects, thresholded at |z| > 10 for the Gambling task: reward contrast. Results are shown using a range of registration methods: FS (FreeSurfer), SulcLo (MSM-sulc with low regularization), SulcHi (MSM-sulc with high regularization), Myelin (MSM-myelin) and RSN (MSM-RSN). In the bottom left the boxplots show percentage improvement in cluster mass, across 86 task contrasts, using a threshold of |z| > 10. Median values are: 0.8%, 4.0%, 6.9% and 21.5% for SulcLo, SulcHi, Myelin and RSN respectively (and for comparison, the values for the gambling reward contrast are −0.3%, 3.0%, 5.7% and 20.5% respectively). The equivalent unthresholded maps can be found in the supplementary material.
Figure 6
Figure 6
Multimodal alignment of sulc, curv and myelin features for a variety of costfunction weightings. Results are compared through contrasting the sharpness of the intersubject averages of the curvature and myelin features, specifically referencing the alignment of curvature in the frontal and temporal lobes (light blue arrow and white loop/arrow respectively), and myelin in the IPS (pink circle) and frontal eye fields (yellow arrow). A) presents the results obtained following univariate alignment of the curvature and myelin features individually, and is shown as a reference. Four different combinations of costfunction weightings were examined: B) no weighting; C) upweighting of the geometric features (i.e., sulc and curv) relative to myelin throughout the whole brain; D) upweighting of myelin features relative to the geometric features throughout the brain; E) upweighting of myelin features regionally according to a binary mask generated by thresholding the group myelin map (top row).

Similar articles

See all similar articles

Cited by 104 articles

See all "Cited by" articles

Publication types

MeSH terms

Feedback