Diffusion MRI fiber tractography of the brain

In the paper

“Filtering the tractogram makes the distribution of endpoints more uniform, but does not prevent the many implausible terminals (middle).”

In the paper

“In particular, trajectories end prematurely in regions containing complex white matter configurations (green arrows) or at the cerebrospinal fluid interface (cyan arrows).”

In the paper

“while the white matter”

In the paper

“Note that Euler integration is a first-order integration method and assumes that the orientation v(r i ) is constant at the length scale of step size Δ, which will make this method highly susceptible to overshoots in highly curved regions, especially for larger step sizes (see Figure 3).”

In the paper

“Although there seems to be a growing consensus in the dMRI community that diffusion tensor tractography should be deprecated, 6,34,37,38 this wisdom does not seem to have been fully translated into clinical practice.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [6]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [34]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [37]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [38]

In the paper

“Higherorder numerical integration schemes, such as Runge-Kutta integration, take into account the variations of v between r i and r i + 1 and are much less susceptible to these integration errors.23,39”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [23]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [39]

In the paper

“It should be clear at this point that, even when using higher-order integration methods, the sheer step-by-step nature of such a procedure makes streamline fiber tracking extremely susceptible to error propagation: indeed, even small local errors can accumulate along the way, causing tracks to veer off course and jump to unrelated adjacent tracts or to stop prematurely.”

In the paper

“Recent studies have shown that DTI-based tractography can result in systematically unreliable and clinically misleading information, while higher-order tractography approaches, using the same dMRI data, clearly demonstrate fiber trajectories that are biologically much more plausible.32,36”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [32]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [36]

In the paper

“It should be noted that even the most advanced higher-order models are still only models, i.e. simplified approximations of the physical reality, and that some of the assumptions behind these models might not always hold.”

In the paper

“even the most advanced tracking algorithms are bound to produce only very crude approximations to the actual connectome.”

In the paper

“In voxels of complex fiber architecture (e.g.crossing fiber populations or partial volume effect between adjacent fiber populations) the diffusion tensor is often a poor representation of the underlying fiber orientations, [28][29][30] causing false negatives, in which tracking can terminate prematurely, 31,32 or false positives, in which tracking can switch to an unrelated adjacent tract.32,33”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [28]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [29]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [30]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [31]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [32]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [33]

In the paper

“Indeed, it is easy to see that, as tracks grow longer, the chance of encountering local tractography errors as well as the extent of error accumulation increases, making it more likely for trajectories to terminate prematurely.”

In the paper

“While this latter approach is particularly appealing as it allows for correlation analyses between structural and functional connectivity, it may suffer from ill-defined fiber orientations present in gray matter regions.”

In the paper

“By treating the problem in a probabilistic fashion, it also becomes possible to track through regions of high uncertainty, where deterministic techniques would usually stop, acknowledging, however, that the probability of”

In the paper

“As previously suggested by Behrens et al., 100 combining the two approaches could potentially provide both types of information simultaneously.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Behrens et al., 100

In the paper

“These local fiber orientations are subject to noise and model errors, resulting in a distribution of possible fiber orientations also known as the uODF.”

In the paper

“While this technique is very powerful, it relies heavily on prior anatomical knowledge.”

In the paper

“Specification of prior knowledge that is too strong might lead to inconsistencies with the actual data, potentially resulting in non-existent fiber trajectories.On the other hand, specification of prior knowledge that is too weak might lead to fiber tracks that perfectly match the underlying data, but that make little sense anatomically.”

In the paper

“These methods construct globally optimal pathways from a given seed point, which do not necessarily always align with the locally estimated fiber orientation.However, they do not consider the pathways from all seed points simultaneously, setting them apart from the purely global methods.”

In the paper

“It is also easy to see that increased architectural complexity will increase the local fiber orientation uncertainty, resulting in fewer tracks making it from start to finish.”

In the paper

“These errors are especially important in the context of streamline tractography, as measurement uncertainty can propagate.”

In the paper

“As a consequence, diffusion from closely aligned fiber orientations will be blurred together, implying that only a single fiber population is present.”

In the paper

“TDI can provide maps with high anatomical contrast, and when combined with a super-resolution voxel grid can potentially resolve sub-voxel anatomical details. 102”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [101]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [102]

In the paper

“Both issues are not only of practical importance when trying to qualitatively reconstruct a particular tract; they are also a confounding factor when it comes to quantification of brain connectivity.”

In the paper

“It should be noted at this point that, while probabilistic uODF tracking gives an indication of the precision of the tracking result, it says nothing about accuracy: even the most reproducible fiber pathways can be highly inaccurate.10”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [10]

In the paper

“In general, this problem has many solutions and selecting a suitable one requires prior knowledge about the expected properties of the trajectories such as local smoothness.”

In the paper

“However, convincing results based on this combination have not been published to date.”

In the paper

“Typically, probabilistic tractography algorithms build on the deterministic streamline approach described in the previous section, and as such are subject to the same limitations.”

In the paper

“Global tracking promises a better stability with respect to noise and imaging artifacts and a better agreement with the actual dMRI data that was acquired.”

In the paper

“Another, lesser-known downside is that streamline tractograms are typically a very poor predictor of the dMRI data that was actually measured, resulting in fiber tracks that have little to no quantitative or biological meaning (this issue is further discussed in section 5.4).”

In the paper

“Typically, these ROIs are defined by the user.This task requires anatomical knowledge and is subject to inter-operator variability.”

In the paper

“Note that whole-brain tracking can be achieved either by seeding from all white matter voxels or by seeding solely from those voxels that are close to the gray matter-white matter interface, as this is where fibers are known to originate from (see Figure 5).”

In the paper

“Local methods are fast and used widely but have important drawbacks.The most apparent is probably that minor errors in the local fiber orientations can accumulate and significantly affect the final result.”

In the paper

“While these frameworks are often mathematically elegant, they are not widely adopted in software packages and their use in practical studies has been limited.”

In the paper

“The overrepresentation of long bundles can be reduced by seeding at the gray matter-white matter interface; however, it is not guaranteed that these trajectories will be able to cover the whole of the white matter, as tracks typically end prematurely due to errors.”

In the paper

“There are many misconceptions that surround fiber tracking, and not recognizing its many technological pitfalls can easily lead to big mistakes, in particular when it comes to interpretation of fiber-tracking results.”

In the paper

“In a sense, this approach can be regarded as providing more biophysically meaningful tractograms, as it is driven by anatomical dispersion rather than noise characteristics and model inaccuracies.”

In the paper

“However, by introducing anatomical constraints during fiber tracking, the resulting fiber trajectories no longer stop prematurely within white matter, but evenly cover the white matter-gray matter interface.”

In the paper

“Also, when studying connection strengths for the entire brain (aka connectomics), whole-brain seeding will potentially overrepresent long fibers, which will result in biased connectivity matrices with little biophysical meaning.”

In the paper

“However, this approach does not necessarily yield accuracy, and some structures observed in TDI have been shown to be caused by noise rather than anatomy.103,104”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [103]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [104]

In the paper

“Finally, and most importantly, the seeding procedures associated with whole-brain tracking are known to cause overrepresentation of long fiber tracks and underrepresentation of short fiber tracks.45,105”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [45]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [105]

In the paper

“While there is no information in the voxel-wise data to disambiguate these local geometries, it is possible to test whether they are supported by the local neighborhood and use this information to make better-informed decisions during fiber tracking.26”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [26]

In the paper

“More specifically, fiber-tracking algorithms have a hard time penetrating into the gyri, and if they do manage they tend to propagate more often towards the crowns of the gyri than towards the gyral walls.88”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [88]

In the paper

“When adopting the fODF approach, at least in theory, we are closer to an actual measure for quantification; however, in practice we still have to curb our enthusiasm, as pretty much exactly the same limitations that plagued local track counts come into play.”

In the paper

“This requirement can be met either by using a global tracking approach, 71,73 which satisfies this requirement by its very definition, or by applying filtering approaches that will remove or weight fiber tracks, such that the sum of their contributions fits the measured dMRI signal best.45,[74][75][76][77]”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [71]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [73]
• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [45]

In the paper

“What is alarming is that these artificial bundles approximate true anatomical structures very well: they connect cortical areas and appear to be dense and well structured.”

In the paper

“First, as TDI is based on fiber tracking, it is very sensitive to noise, the accumulation of noise-induced errors, and model imperfections.”

In the paper

“Matters become even more complicated in the case of fanning fiber configurations, as one should also take into account the polarity of the fanning.”

In the paper

“In general, techniques that disambiguate local fiber geometries do not seem to have found their way into popular fiber-tracking software yet and remain inaccessible to most researchers.”

In the paper

“In theory, global tractography methods should already be able to resolve these local ambiguities by their very definition, as they search for the set of continuous trajectories that best explain the full dMRI data, in particular when they are combined with microstructural continuity priors.113”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [113]

In the paper

“However, each of these geometries may require different decisions to be taken during track propagation, and failing to distinguish between these cases could adversely affect fiber-tracking outcomes.”

In the paper

“While most high-order local fiber models can provide high-quality fiber orientation estimates in voxels containing pure white matter, in voxels partially containing grey matter tissue these models may no longer be appropriate and are known to produce unreliable, noisy fiber orientation estimates.47”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [47]

In the paper

“It should be clear that fiber tracking is a complicated endeavor involving many steps (local fiber orientation modeling, interpolation, integration/ propagation, seeding, masking, stopping criteria, etc.), each incurring potential errors.”

In the paper

“Such effects can potentially veer a track off course and create completely artificial fiber trajectories.”

In the paper

“By using a proper high-order model, by applying realistic anatomical priors, and by ensuring fidelity of the tractograms to the data, the biological accuracy can be vastly increased and one could potentially start to derive useful connectivity information.”

In the paper

“Without a proper high-order model, without anatomical constraints, and without enforcing fidelity of the fiber tracks to the raw dMRI data, quantifying connections using fiber tracking is the equivalent of opening Pandora's box: the resulting”

In the paper

“Experimental evidence suggests that the angular frequency content of the dMRI signal in the brain white matter levels off at a b-value of approximately b = 3000 s/mm 2 , suggesting that this b-value achieves the highest practically available angular resolution.122”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites [122]

In the paper

“Increasing spatial resolution is challenging, as it typically goes hand in hand with either a significant drop in SNR or a significant increase in acquisition time, resulting in either poor-quality data or impractically long scan times.”

In the paper

“As a consequence, even with state-of-the-art high-angularresolution diffusion imaging and cutting-edge fiber orientation estimation techniques, inter-fiber angles below 30° are rarely resolved, and are a potential source of inaccurate and/or completely artifactual fiber trajectories.”