Comments and ControversiesWhite matter integrity, fiber count, and other fallacies: The do's and don'ts of diffusion MRI
Introduction
Diffusion weighted MRI (DW-MRI) (Behrens and Johansen-Berg, 2009, Jones, 2010a, Le Bihan and Breton, 1985, Le Bihan et al., 1986) is currently the only method capable of mapping the fiber1 architecture of tissue (e.g., nervous tissue, muscle) in vivo and, as such, it has triggered tremendous hopes and expectations. As the technique has matured, an increasing number of software packages have been developed that allow such data to be analyzed in a push-button manner — sometimes to such an extent that the end-user need not know anything about the underlying physics, and yet are still able to derive a p-value which can be interpreted according to the hypotheses being tested. There are, however, a substantial number of pitfalls associated with these methods (see, e.g., Jones, 2010b, Jones, 2010c, Jones and Cercignani, 2010, Le Bihan et al., 2006), which can lead to biased or, in some cases, completely fallacious conclusions being drawn. What is not in question, however, is that DW-MRI carries invaluable in vivo information about tissue microstructure, but in order to extract this information in the most efficient and unbiased way, it is important to make the right choices for the acquisition and analysis of these data, and, even more importantly, for the interpretation of the results.
It is in this context that this article focuses on three issues. The first of these is: What exactly are we measuring with DW-MRI, i.e., what is the immediate meaning of the data we get from the technique? The second is: What questions are we trying to answer on the basis of DW-MRI? The link between these two issues is the basis for the third issue, interpretation. Our main focus is on measurements within white matter of the live human brain, although many of the issues discussed here are equally relevant to pre-clinical studies in animal models, and of tissues other than brain.
The target audience of this article is the typical ‘end user’ who has access to diffusion-weighted MR sequences, provided by the MR scanner manufacturer, and uses ‘push-button’ software packages to analyze their data to look for group differences or structure–function correlations. It is our opinion that, without basic insight into the fundamental principles of the method and, most importantly, its limitations and pitfalls, misunderstandings, misconceptions and misinterpretation will be perpetuated. Our aim is to provide this grounding to the aforementioned target audience.
Section snippets
What does diffusion-weighted MR imaging actually measure?
Diffusion-weighted MRI measures just one thing — the dephasing of spins of protons in the presence of a spatially-varying magnetic field (‘gradient’). The mechanism of interest here is the phase change resulting from components of incoherent displacement of spins along the axis of the applied field gradient, which changes their Larmor frequency. The longer the protons are allowed to diffuse (the ‘diffusion time’, Δ) and the higher the mean squared displacement per unit time of the molecules
What questions do we ask of the DW-MRI data?
There are, of course, myriad questions that are asked of DW-MRI data — but these can be grouped into classes.
One important class is concerned with the trajectory of fiber pathways and their interpretation in terms of anatomical connectivity, and includes questions of the form: “Which gray matter regions are inter-connected by white matter fibers?”; “Where do these fibers pass?”; and “How strong are these connections?”4
Interpretation
The interpretation of DW-MRI data is essentially a model based procedure, even if no formal, mathematically described model is invoked, i.e., the measured data are combined with a number of assumptions about the underlying processes and structures. These model assumptions always represent a simplification of reality, i.e., they neglect certain aspects of the true generative mechanism of the data. For the choice of the model, three aspects are important: (a) the quality and quantity of the
Conclusions
As we stated at the outset, the only thing that that we can say with any certainty in diffusion MRI is that we measure a signal change when a motion-sensitizing gradient is applied along a given axis. Inferring anything else is dependent on the quality of the model and the quality of the data. There are many mechanisms by which the diffusion weighted signal can be modulated. This includes but is not limited to, the myelination, the axon density, the axon diameter, the permeability of the
The do's
- a.
Carefully consider the question(s) to be asked of the data and consider whether the data acquisition/analysis allows you to answer these questions.
As most of the recommendations given below (small voxels, many directions, high diffusion weighting, high SNR) are in mutual competition, the user has to decide where to invest the precious acquisition time. For simple questions such as unspecific white matter differences between two groups, there are minimal demands on the data acquisition and
The don'ts
- a.
Don't assume that the principal eigenvector of a diffusion tensor is a good indication of the actual fiber orientations in all voxels.
Although in a limited set of places (where the bundle-to-voxel size is favorable and all fibers are highly parallel in the voxel), the principal eigenvector may do a good job, it is unsafe to use this simple model throughout the whole brain.
- b.
Don't assume that tractography using a single diffusion tensor will be adequate for all fiber trajectories in the brain.
Most
References (193)
- et al.
Orientationally invariant indices of axon diameter and density from diffusion MRI
NeuroImage
(2010) - et al.
Composite hindered and restricted model of diffusion (charmed) MR imaging of the human brain
NeuroImage
(2005) - et al.
Non-mono-exponential attenuation of water and N-acetyl aspartate signals due to diffusion in brain tissue
J. Magn. Reson.
(1998) - et al.
Dementia induces correlated reductions in white matter integrity and cortical thickness: a multivariate neuroimaging study with sparse canonical correlation analysis
NeuroImage
(2010) - et al.
MR diffusion tensor spectroscopy and imaging
Biophys. J.
(1994) - et al.
Estimation of the effective self-diffusion tensor from the NMR spin echo
J. Magn. Reson. B
(1994) - et al.
Probabilistic diffusion tractography with multiple fibre orientations: what can we gain?
NeuroImage
(2007) - et al.
White matter abnormalities revealed by diffusion tensor imaging in non-demented and demented HIV + patients
NeuroImage
(2009) - et al.
Atypical development of white matter microstructure in adolescents with autism spectrum disorders
NeuroImage
(2010) - et al.
Does diffusion kurtosis imaging lead to better neural tissue characterization? A rodent brain maturation study
NeuroImage
(2009)