Tract-based spatial statistics: Voxelwise analysis of multi-subject diffusion data
Introduction
The diffusion of water in brain tissue is affected by the local tissue microstructure; for example, it diffuses more easily along the major axis of a white matter fibre bundle than perpendicular to it (Moseley et al., 1990). Magnetic resonance diffusion tensor imaging (DTI) is sensitive to water diffusion characteristics (such as the principal diffusion direction and the diffusion anisotropy) and has therefore been developed as a tool for investigating the local properties of brain tissues such as white matter tracts (Le Bihan, 2003). There has also been a great deal of interest in using diffusion anisotropy as a marker for white matter tract integrity, for example, for disease diagnosis, tracking disease progression, finding disease sub-categories, studying normal development/aging, and as complementary information to investigating normal brain function (Horsfield and Jones, 2002, Lim and Helpern, 2002, Moseley, 2002, Neil et al., 2002, Pagani et al., 2005).
Diffusion anisotropy describes how variable the diffusion is in different directions and is most commonly quantified via a measure known as fractional anisotropy (FA) (Pierpaoli and Basser, 1996). It is highest in major white matter tracts (maximum theoretical value 1) and lower in grey matter while approaching 0 in cerebro-spinal fluid. As a marker for tract integrity, FA is a useful quantity to compare across subjects as it is computable voxelwise and is a scalar value that is independent of the local fibre orientation (and therefore a relatively objective and straightforward measure to compare across subjects). Some researchers have simply summarised diffusion characteristics globally (for example, histogram-based summary measures of fractional anisotropy (Cercignani et al., 2001, Cercignani et al., 2003)), in order to compare different subjects. However, most recent work has been interested in spatially localising interesting diffusion-related changes. Many studies have, to this end, followed similar approaches to voxel-based morphometry (VBM, originally developed for finding local changes in grey matter density in T1-weighted structural brain images (Ashburner and Friston, 2000, Good et al., 2001)). In VBM-style FA analysis, each subject's FA image is registered into a standard space, and then voxelwise statistics are carried out to find areas which correlate to the covariate of interest (e.g., patients vs. normals, disability score, age).
There has been much debate about the strengths and limitations of VBM (Bookstein, 2001, Ashburner and Friston, 2001, Davatzikos, 2004, Ashburner and Friston, 2004). Some researchers continue to doubt the general interpretability of the results from this approach, primarily because there can be ambiguity as to whether apparent changes are really due to change in grey matter density or simply due to local misalignment, though it does seem that through careful application and validation, structural imaging studies using VBM can draw valid conclusions (e.g., Watkins et al., 2002). However, the potential problems with VBM-style approaches for data such as multi-subject FA images have not yet been investigated fully. In particular, this use raises a serious question, which has not yet been satisfactorily answered: how can one guarantee that any given standard space voxel contains data from the same part of the same white matter (WM) tract from each and every subject? In other words, how can we guarantee that registration of every subject's data to a common space has been totally successful, both in terms of resolving topological variabilities and in terms of the exact alignment of the very fine structures present in such data? A second problem relates to the standard practice of spatially smoothing data before computing voxelwise statistics — the amount of smoothing can greatly affect the final results, but there is no principled way of deciding how much smoothing is the “correct” amount (Jones et al., 2005). (Smoothing also increases effective partial voluming, a problem with VBM-style approaches particularly when applied to data such as FA; see Discussion for more comment on this.)
In this paper, we propose an approach to carrying out localised statistical testing of FA (and other diffusion-related) data that should alleviate the alignment problems. We project individual subjects' FA data into a common space in a way that is not dependent on perfect nonlinear registration. This is achieved through the use of (a) an initial approximate nonlinear registration, followed by (b) projection onto an alignment-invariant tract representation (the “mean FA skeleton”). No spatial smoothing is necessary in the image processing. We refer to this new approach as Tract-Based Spatial Statistics (TBSS). In the next section, we discuss, in slightly more depth, VBM-style approaches, and review some alternative approaches published to date. In following sections, we describe the proposed approach in detail, giving various example images illustrating the different analysis stages involved. Finally, we present example TBSS results from several DTI-based imaging studies.
Section snippets
VBM — overview and application to diffusion data
Voxel-based morphometry (Ashburner and Friston, 2000, Good et al., 2001) has been used in many structural imaging studies, looking for localised differences in grey matter density, typically between two groups of subjects. The common approach can be simply summarised:
- •
(Optional) Create a study-specific registration template by aligning all subjects' structural images to an existing standard space template image (such as the MNI152). Average all aligned images to create the new template, and
Overview of TBSS
As discussed above, strengths of VBM-style analyses are that they are fully automated, simple to apply, investigate the whole brain, and do not require prespecifying and prelocalising regions or features of interest. Limitations include problems caused by alignment inaccuracies, and the lack of a principled way for choosing smoothing extent. Tractography-based approaches have fairly complementary advantages and disadvantages. They can overcome alignment problems by working in the space of
Results
In the following sections, we present example results and quantitations from different stages of the TBSS analysis, followed by example results from several diffusion imaging studies. The data generally used to illustrate TBSS are taken from a study of amyotrophic lateral sclerosis (ALS, a progressive neurodegenerative disease most prominently affecting the motor system). The diffusion acquisition parameters for this and all other data used in this paper are given in Fig. 5.
Discussion
In this final section, we discuss some of the limitations of our approach, as well as presenting some potentially interesting areas for future research.
Acknowledgments
We are grateful to Karla Miller for many discussions regarding the TBSS approach, Andreas Bartsch for valuable advice on this paper, Einar Heiervang for supplying the reproducibility data, and David Flitney and Brian Patenaude for work on the figure generation. We gratefully acknowledge funding from EPSRC, BBSRC, MRC, the Wellcome Trust and the Multiple Sclerosis Society.
References (54)
- et al.
Voxel-based morphometry — the methods
NeuroImage
(2000) - et al.
Why voxel-based morphometry should be used
NeuroImage
(2001) - et al.
Estimation of the effective self-diffusion tensor from the NMR spin echo
J. Magn. Reson., B
(1994) “Voxel-based morphometry” should not be used with imperfectly registered images
NeuroImage
(2001)Why voxel-based morphometric analysis should be used with great caution when characterizing group differences
NeuroImage
(2004)- et al.
A voxel-based morphometric study of ageing in 465 normal adult human brains
NeuroImage
(2001) - et al.
Average brain models: a convergence study
Comput. Vis. Image Underst.
(2000) Mapping eddy current induced field for the correction of diffusion weighted echo planar images
Magnetic Resonance Imaging
(1999)- et al.
A global optimisation method for robust affine registration of brain images
Med. Image Anal.
(2001) - et al.
Improved optimisation for the robust and accurate linear registration and motion correction of brain images
NeuroImage
(2002)
Spatial normalisation and averaging of diffusion tensor MRI data sets
NeuroImage
The effect of filter size on VBM analyses of DT-MRI data
NeuroImage
Cingulate fasciculus integrity disruption in schizophrenia: a magnetic resonance diffusion tensor imaging study
Biol. Psychiatry
A method for obtaining tract-specific diffusion tensor MRI measurements in the presence of disease: application to patients with clinically isolated syndromes suggestive of multiple sclerosis
NeuroImage
A comparison of random field theory and permutation methods for the statistical analysis of MEG data
NeuroImage
Spatial normalization of diffusion tensor MRI using multiple channels
NeuroImage
White matter hemisphere asymmetries in healthy subjects and in schizophrenia: a diffusion tensor MRI study
NeuroImage
Volumetric, connective, and morphologic changes in the brains of children with chromosome 22q11.2 deletion syndrome: an integrative study
NeuroImage
Disconnection of speech-relevant brain areas in persistent developmental stuttering
The Lancet
Generative and recognition models for neuroanatomy
NeuroImage
White matter tract alterations in fragile X syndrome: preliminary evidence from diffusion tensor imaging
Am. J. Med. Genet., Part B Neuropsychiatr. Genet.
Probabilistic independent component analysis for functional magnetic resonance imaging
IEEE Trans. Med. Imag.
Non-invasive mapping of connections between human thalamus and cortex using diffusion imaging
Nat. Neurosci.
Characterization and propagation of uncertainty in diffusion-weighted MR imaging
Magn. Reson. Med.
White matter asymmetry in the human brain: a diffusion tensor MRI study
Cereb. Cortex
Mean diffusivity and fractional anisotropy histograms of patients with multiple sclerosis
Am. J. Neuroradiol.
Inter-sequence and inter-imaging unit variability of diffusion tensor MR imaging histogram-derived metrics of the brain in healthy volunteers
Am. J. Neuroradiol.
Cited by (5324)
Microstructural abnormality of white matter tracts in rheumatoid arthritis
2024, Brain ResearchComputational limits to the legibility of the imaged human brain
2024, NeuroImageSex differences in the rodent medial prefrontal cortex – What Do and Don't we know?
2024, NeuropharmacologyThe hindbrain and cortico-reticular pathway in adolescent idiopathic scoliosis
2024, Clinical Radiology