Surface-enhanced tractography (SET)
Introduction
Diffusion MRI (dMRI) is a non-invasive technique that allows the reconstruction of the white matter (WM) structure. Axonal pathways can be reconstructed in-vivo by following the local orientation of the water diffusion with a process called tractography. This in-vivo reconstruction of the white matter can be used for structural connectivity studies (Wakana et al., 2007, Hagmann et al., 2007, Fornito et al., 2013). Structural connectivity mapping (connectomics) can be estimated through streamline endpoints produced by tractography (Yo et al., 2009, Jbabdi et al., 2015).
However, limitations in dMRI and tractography can lead to biased measurements and conclusions (Jones, 2008, Descoteaux et al., 2009, Yo et al., 2009, Jones and Cercignani, 2010, Jbabdi and Johansen-Berg, 2011, Tournier et al., 2011, Jones et al., 2013). Compared to standard anatomical MRI image, dMRI has an intrinsically low signal-to-noise ratio and low spatial resolution (Tournier et al., 2011). Partial volume effect (PVE), caused by the poor spatial resolution (typically 2 mm isotropic), and angular discretization reduces the precision of streamlines tractography. Another problem is that most streamlines stop in the middle of the white matter without reaching the gray matter (GM) regions (seeding and masking problems) (Jbabdi and Johansen-Berg, 2011, Smith et al., 2012b, Girard et al., 2014). Moreover, the low resolution of dMRI can also lead to the gyral bias of tractography (Van Essen et al., 2013a, Reveley et al., 2015), as shown in Fig. 1, where current iterative tractography algorithms are unable to capture the fanning structure in the gyral blade.
The goal of this work is to confront dMRI limitations in the superficial WM and improve tractography methods to overcome the poor resolution, the partial volume effect and the gyral bias. We therefore propose a surface-enhanced tractography technique, a surface-based differential geometry approach, to model the main orientation of the WM structure under the cortex, without the need of any diffusion information. We also present a surface seeding and stopping strategy, utilizing the cortex normal and curvature to improve and enhance current tractography algorithms. Together, these methods lead to more reproducible and less variable tractograms and connectivity matrices.
Section snippets
Background
We now present some of the existing literature on cortical folding models along with the underlying axonal structure. The proposed surface-enhanced method is derived from these gyrification and fiber models.
Methods
In this work, we propose to apply a geometric flow to the WM-GM boundary surface to model the predominant fiber pathway and improve the tractography near the cortex. As specified in the next section, each point of the mesh will flow inwardly, orthogonal to the cortical surface, creating a less convoluted surface. Consequently, this surface flow will be used to initialize and terminate a traditional streamlines tractography algorithm using the dMRI information from the fiber orientation
Human Connectome Project (HCP) data
Surface Flow From Fig. 2, the surface flow evolution and the points moving along the inward normal direction and mean-curvature can be observed. Surface flow can be seen as a smoothing weighted by positive curvature (stiffness) and area (mass) (Fig. 2-a). From each vertex composing the mesh we can reconstruct a line going towards the inside of the gyri, in the surface normal direction, bending with the curvature and area compression (Fig. 2-b). For SET, the inner surface generated from the
Discussion
We have proposed a novel surface-enhanced tractography strategy to overcome the gyral bias of classical tractography algorithms. We showed that tractography can be improved using a surface flow via the WM-GM surface mesh obtained from a high-resolution T1, seeding from the endpoints of this flow and using the flow to back-project streamlines to the WM-GM, subcortical and brainstem surface meshes. Hence, dMRI tractography is improved by our surface-based approach because:
- •
it improves the
Conclusion
Surface meshes from a classical T1-weighted image can boost the resolution and performance of dMRI tractography. Since an anatomical T1 image is generally acquired with dMRI and it is often 1 mm isotropic, this is a straightforward and elegant way to improve existing tractography processing pipelines and structural connectivity studies.
A cortical surface flow model of the superficial white matter streamlines can enhance and improve tractography. Surface-enhanced tractography is able to reduce
Conflicts of interest
We have no conflict of interest to declare.
Acknowledgements
A Special thanks to Eleftherios Garyfallidis, François Rheault, Giorgio Innocenti, Jasmeen Sidhu and Tim Dyrby for their help and insights. Acknowledgements to Maxime Chamberland, Kevin Whittingstall and the Sherbrooke Molecular Imaging Center for the data acquisition of the test-retest dataset. For research founding, thanks to FRQNT, NSERC Discovery Grant (RGPIN-2015-05 29 7), CREATE-MIA program (150682827) and Université de Sherbrooke Institutional Research Chair in NeuroInformatics.
References (66)
- et al.
Mechanical forces in cerebral cortical folding: a review of measurements and models
J. Mech. Behav. Biomed. Mater.
(2014) - et al.
A generalised framework for super-resolution track-weighted imaging
Neuroimage
(2012) - et al.
Cortical surface-based analysis: I. segmentation and surface reconstruction
Neuroimage
(1999) - et al.
A sulcal depth-based anatomical parcellation of the cerebral cortex
NeuroImage
(2009) - et al.
Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain
Neuron
(2002) - et al.
Graph analysis of the human connectome: promise, progress, and pitfalls
Neuroimage
(2013) - et al.
Towards quantitative connectivity analysis: reducing tractography biases
Neuroimage
(2014) Studying connections in the living human brain with diffusion mri
Cortex
(2008)- et al.
White matter integrity, fiber count, and other fallacies: the do's and don’ts of diffusion mri
Neuroimage
(2013) - et al.
Automated 3-d extraction and evaluation of the inner and outer cortical surfaces using a laplacian map and partial volume effect classification
Neuroimage
(2005)