Elsevier

NeuroImage

Volume 62, Issue 3, September 2012, Pages 1924-1938
NeuroImage

Anatomically-constrained tractography: Improved diffusion MRI streamlines tractography through effective use of anatomical information

https://doi.org/10.1016/j.neuroimage.2012.06.005Get rights and content

Abstract

Diffusion MRI streamlines tractography suffers from a number of inherent limitations, one of which is the accurate determination of when streamlines should be terminated. Use of an accurate streamlines propagation mask from segmentation of an anatomical image confines the streamlines to the volume of the brain white matter, but does not take full advantage of all of the information available from such an image. We present a modular addition to streamlines tractography, which makes more effective use of the information available from anatomical image segmentation, and the known properties of the neuronal axons being reconstructed, to apply biologically realistic priors to the streamlines generated; we refer to this as “Anatomically-Constrained Tractography”. Results indicate that some of the known false positives associated with tractography algorithms are prevented, such that the biological accuracy of the reconstructions should be improved, provided that state-of-the-art streamlines tractography methods are used.

Highlights

► Modular improvement to diffusion MRI streamlines tractography. ► Effective use of anatomical information and biological priors. ► Prevents spurious streamline terminations for improved connectome reconstruction.

Introduction

Diffusion-weighted (DW) MRI is capable of characterising tissue microstructure in vivo non-invasively. Within the brain white matter, characterisation of the anisotropic nature of the signal (achieved through acquisition of multiple images with varying diffusion sensitisation directions) allows for the estimation of the orientations of the underlying coherent neuronal axon populations (Tournier et al., 2011). This result is naturally amenable to so-called ‘fibre-tracking’ applications, where these orientations are used to define tangents upon continuous trajectories (‘streamlines’) to reconstruct plausible long-range anatomical connections (Mori and van Zijl, 2002).

The diffusion tensor was the first proposed model for characterising the underlying tissue microstructure in terms of the measured diffusion signal (Basser et al., 1994). The diffusion tensor is however only capable of representing a single fibre orientation per voxel; in circumstances of complex fibre architecture (such as bending, crossing and kissing fibres, or partial volume contamination between adjacent fibre populations), the diffusion tensor is a poor representation of the underlying structure, and therefore leads to errors in the estimation of fibre orientations (with concomitant errors in the attempted reconstruction of the underlying fibre pathways). Previous research suggested that at typical DW-imaging resolutions, around one-third of white matter voxels in the brain contain such complex architecture (Behrens et al., 2007); however, more recent evidence indicates that this figure may be as high as 90% (Jeurissen et al., 2012).

Over the last decade, a range of so-called ‘higher-order’ methods have been developed with the capability of estimating the orientations and relative contributions of the fibre populations within each image voxel, without any explicit assumption regarding the number of underlying fibre populations (Tournier et al., 2011). These methods often represent these orientations as a continuous function on the sphere for each voxel, known as the Fibre Orientation Distribution (FOD) (or, alternatively, the fibre Orientation Distribution Function; fODF). By probing the shape of these FODs and applying some simple priors, streamlines tractography can be performed even in white matter regions containing complex fibre architecture.

Tractography algorithms also vary in the way they sample from these orientations as they determine the streamline trajectories. Independent of the particular diffusion model used, these generally fall into two categories. In deterministic streamlines tractography (e.g. Mori et al., 1999), the ‘best fit’ fibre orientation is selected at each point and traced exactly. These algorithms do not take the uncertainty in these orientations into account, and can therefore fail to reconstruct connections between distant regions due to small amounts of noise in the data. Conversely, probabilistic streamlines tractography (e.g. Parker et al., 2003) considers the uncertainty in the calculation of the fibre orientations given the image data available, and statistically samples from these distributions as the streamlines are constructed.

Regardless of the diffusion model or algorithm used, streamlines tractography inherently suffers from a number of limitations that can reduce the accuracy of the reconstructed connections with respect to the underlying biology. The difficulty that is the focus of this paper is the determination of appropriate termination points for the streamlines; although DW-images provide useful fibre orientation information from the underlying axonal populations, they do not provide an appropriate contrast for accurately identifying where these axon bundles end.

Streamlines algorithms typically use a number of criteria for determining when a streamline should cease. Firstly, a threshold may be applied to some metric from the diffusion model, indicating that the anisotropy of the signal is poor; this is typically the Fractional Anisotropy (FA) when tracking using the diffusion tensor, or the amplitude of the FOD if using a higher-order model. This is (perhaps optimistically) considered to be an indicator of the presence of grey matter (where the cytoarchitecture is typically more complex than that of white matter, leading to less coherent organisation of orientations) or CSF.

Secondly, the streamlines may be terminated if their local curvature exceeds a given threshold; if high curvature paths were permitted, streamlines may perform a tight ‘U-turn’ and travel along the same pathway in the opposite direction, a situation that is unlikely to occur in biology. This restriction is typically enforced as a threshold on the angle of curvature between successive steps.

Finally, it is common to employ a binary tracking mask, which defines the volume within which streamlines are permitted to exist. Such a mask may be defined simply by applying a threshold to the diffusion images to separate tissue from background. These masks however suffer from the poor spatial resolution of the diffusion images, and the fact that the contrast of these images is optimised for estimating the orientations of the underlying fibres, rather than differentiating between tissue types.

As these criteria for streamline termination are imperfect, it is beneficial to make use of prior anatomical knowledge and/or structural information to guide the tractography process; for instance, using regions of interest based on anatomy (e.g. Catani et al., 2002, Hagmann et al., 2003, Wakana et al., 2007), or restricting propagation to the volume of the white matter (and optionally also grey matter) based on segmentation of an image with high anatomical tissue contrast and spatial resolution (e.g. Guevara et al., 2011, Iturria-Medina et al., 2007, Sotiropoulos et al., 2010). However the use of such information to define simple binary masks for tracking does not make full use of the anatomical information available. For instance, if the segmented white matter image is used as a streamlines propagation mask, the streamlines will be spatially confined to the volume of the white matter; but this constraint does not fully exploit the prior information we have regarding the nature of the anatomical connections that we aim to reconstruct using streamlines tractography.

Here we propose a novel and modular addition to streamlines tractography, which makes more effective use of the information provided by segmentation of an anatomical contrast image to apply biologically realistic priors. Rather than simply defining a binary mask within which to confine our streamlines, we consider the biological properties of the tissue types and fluid within the brain, as well as the nature of the axons of the white matter we are attempting to reconstruct, and tailor our streamlines criteria to match these known properties. We refer to this as “Anatomically-Constrained Tractography” (ACT). Because of the way in which we construct these priors, they can be applied independent of the particular diffusion model or streamlines algorithm used. Here we outline the processing steps necessary to obtain the relevant anatomical information, describe the anatomical criteria we place on the streamlines, and investigate the effects of these priors on the construction of the tractogram for a number of different diffusion models and streamlines algorithms.

Section snippets

Method

The processing steps we have chosen to enable Anatomically-Constrained Tractography are described in full here; Fig. 1 presents these steps visually. Note however that alternative and/or additional software, processes or algorithms could be chosen instead, as long as the information required for Anatomically-Constrained Tractography can be provided in the appropriate format. For all processing steps where the relevant software is not explicitly stated, processing was performed using MRtrix (

Results

Fig. 4 presents the whole-brain fibre-tracking results for the four streamlines tractography algorithms employed here, using Directionally-Encoded Colour (DEC) Track Density Images (TDI) (Calamante et al., 2010) overlaid on the subject's segmented anatomical image. For each algorithm, results are presented for: ‘Conventional’ (i.e. non-ACT) tracking using a DWI-derived tracking mask; ‘conventional’ tracking using the white matter mask from FAST segmentation of the subject's anatomical image as

Discussion

We have presented a new addition to diffusion MRI streamlines tractography, which makes effective use of the information available from anatomical-contrast images. We apply constraints on the streamlines generated that are based on the anatomical knowledge of the biological structures being reconstructed using this framework; as such, these reconstructions should be more realistic with respect to the underlying biology.

Although there has been some previous experimentation with the use of white

Conclusion

We have introduced a novel framework for incorporating prior anatomical information into the diffusion MR streamlines tractography process, which makes effective use of the information available from high-resolution anatomical MR images to improve the biological plausibility of the trajectories generated. It can be applied to any streamlines tractography algorithm in a modular fashion. We have presented preliminary results on whole-brain tractography reconstruction suggesting that seeding from

Acknowledgments

We are grateful to the National Health and Medical Research Council (NHMRC) of Australia, the Australian Research Council, Austin Health, the Dowd Foundation Research Scholarship for Neuroscience, and the Victorian Government's Operational Infrastructure Support Program for their support.

References (49)

  • S. Jbabdi et al.

    A Bayesian framework for global tractography

    NeuroImage

    (2007)
  • B. Patenaude et al.

    A Bayesian model of shape and appearance for subcortical brain segmentation

    NeuroImage

    (2011)
  • G.R. Ridgway et al.

    Issues with threshold masking in voxel-based morphometry of atrophied brains

    NeuroImage

    (2009)
  • S.M. Smith et al.

    Advances in functional and structural MR image analysis and implementation as FSL

    NeuroImage

    (2004)
  • S.N. Sotiropoulos et al.

    Brain tractography using Q-ball imaging and graph theory: improved connectivities through fibre crossings via a model-based approach

    NeuroImage

    (2010)
  • J.-D. Tournier et al.

    Diffusion-weighted magnetic resonance imaging fibre tracking using a front evolution algorithm

    NeuroImage

    (2003)
  • J.-D. Tournier et al.

    Robust determination of the fibre orientation distribution in diffusion MRI: non-negativity constrained super-resolved spherical deconvolution

    NeuroImage

    (2007)
  • D.J. Tozer et al.

    Linking white matter tracts to associated cortical grey matter: A tract extension methodology

    NeuroImage

    (2012)
  • M.P. van den Heuvel et al.

    Small-world and scale-free organization of voxel-based resting-state functional connectivity in the human brain

    NeuroImage

    (2008)
  • S. Wakana et al.

    Reproducibility of quantitative tractography methods applied to cerebral white matter

    NeuroImage

    (2007)
  • Y. Zhang et al.

    Atlas-guided tract reconstruction for automated and comprehensive examination of the white matter anatomy

    NeuroImage

    (2010)
  • J. Ashburner et al.

    Rigid Body Registration

  • P.J. Basser et al.

    In vivo fiber tractography using DT-MRI data

    Magn. Reson. Med.

    (2000)
  • P. Guevara et al.

    Accurate tractography propagation mask using T1-weighted data rather than FA

  • Cited by (747)

    • White matter and literacy: A dynamic system in flux

      2024, Developmental Cognitive Neuroscience
    View all citing articles on Scopus
    View full text