Potential Pitfalls of Using Fractional Anisotropy, Axial Diffusivity, and Radial Diffusivity as Biomarkers of Cerebral White Matter Microstructure

In the paper

“it is now widely recognized that the majority of cerebral white matter voxels contain multiple fiber populations with different trajectories”

What the sources say

• Partly supported

  The Abstract says that multiple fibre populations occur across large regions of white matter. But Results (Local fitting of the complex fibre model; page unavailable) reports that only one third of voxels with fractional anisotropy above 0.1 supported more than one orientation in the 60-direction dataset. This supports the presence of multiple populations, not the claim that they occur in a majority of cerebral white matter voxels. (LLM, conf=0.88)

  cites Behrens et al., 2007 sourceExperimental — accuracy not yet measured
• Supported

  In Vivo Estimation of Fiber Orientations (Table I) reports two or more fiber orientations in approximately 90% of white matter voxels using CSD and 63% using bedpostx. Both estimates support the claim that a majority contain multiple fiber populations with different trajectories. (LLM, conf=0.94)

  cites Jeurissen et al., 2013 source

In the paper

“and is more likely somewhere between 60 and 90% (Jeurissen et al., 2013)”

What the sources say

• Supported

  The In Vivo Estimation of Fiber Orientations section (Table I) reports multiple fiber orientations in approximately 90% of white-matter voxels using CSD and complex configurations in approximately 63% using bedpostx. The Discussion repeats both estimates, supporting the cited 60–90% range. The percentages are method-dependent estimates, not a single measured proportion. (LLM, conf=0.94)

  cites Jeurissen et al., 2013 source

In the paper

“with many regions thought to contain as many as 3 or more intersecting fiber bundles with different trajectories (Figure 1)”

What the sources say

• Supported

  The In Vivo Estimation of Fiber Orientations section reports large clusters of voxels with three fiber populations, including a region where three named bundles interdigitate. It also reports that CSD detected three or more orientations in about 45% of white-matter voxels, supporting the claim's hedged wording. (LLM, conf=0.93)

  cites Jeurissen et al., 2013 source

In the paper

“the proportion of cerebral white matter voxels containing complex fiber geometries and/or multiple fiber bundles in different orientations (i.e., crossing fibers, kissing fibers, etc.) is now thought to be at least 33% (Behrens et al., 2007)”

What the sources say

• Partly supported

  The Results (Local fitting of the complex fibre model) and Discussion report detectable complex fibre architecture in approximately one third of voxels with fractional anisotropy greater than 0.1 using a 60-direction dataset; page numbers are not supplied. This supports the approximate proportion, but not a lower bound of at least 33% for all cerebral white matter voxels. (LLM, conf=0.88)

  cites Behrens et al., 2007 sourceExperimental — accuracy not yet measured

In the paper

“in the case of ihMT and MWI likely also provide more myelin-specific information than can be obtained with more general measures based on DTI or T1w/T2w ratio metrics (Mädler et al., 2008; Ercan et al., 2018; Uddin et al., 2019).”

What the sources say

• ?Not checked — no full text of the sourcecites Mädler et al., 2008 source
• ?Not checked — no full text of the sourcecites Ercan et al., 2018 source
• Partly supported

  The Introduction calls MWF a myelin-specific measure and DTI metrics general measures of tissue microstructure; the Conclusions also say T1w/T2w ratios should not be interpreted as myelin-specific in subcortical structures. The paper mentions an earlier ihMTR comparison but does not establish that ihMT is more myelin-specific than DTI or T1w/T2w. Page numbers are unavailable in our copy. (LLM, conf=0.86)

  cites Uddin et al., 2019 sourceExperimental — accuracy not yet measured

In the paper

“as pointed out by Jeurissen et al. (2013), even regions with crossing fibers, many observed differences/changes in FA, AD and/or RD are likely to reflect actual anatomical and biophysical phenomena, as long as other pitfalls in the acquisition and analysis have been avoided (Jones and Cercignani, 2010).”

What the sources say

• Partly supported

  In “Implications for DTI” (page not provided), the paper says that, if known analysis pitfalls are avoided, an observed FA change is highly likely to reflect a true underlying biophysical phenomenon—even though its meaning is ambiguous in crossing-fiber regions. It discusses axial and radial diffusivity as similarly difficult to interpret, but does not explicitly extend that likelihood statement to changes in AD or RD. (LLM, conf=0.86)

  cites Jeurissen et al., 2013 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Jones and Cercignani, 2010 source

In the paper

“ex vivo MRI and histological studies of the macaque brain have revealed that white matter crossing fibers become even more apparent upon higher resolution and more detailed examination”

What the sources say

• Supported

  The Introduction describes ex vivo MRI of a macaque brain followed by histological analysis of the same specimen. The Conclusion reports that crossing fibers became more frequent at higher spatial resolutions in both MRI and histology, supporting the citing sentence’s paraphrase. (LLM, conf=0.94)

  cites Schilling et al., 2017 source

In the paper

“DKI data acquired using conventional hardware and reasonable scan times can resolve crossing fibers significantly better than conventional DTI (Glenn et al., 2016)”

What the sources say

• Partly supported

  The Conclusion (page unavailable) says DKI directly resolves crossing fibers and improves tractography accuracy relative to DTI; the Methods report use of a 3T MRI system. However, the Discussion says this study’s scan time was relatively long compared with typical clinical protocols. It cites shorter DKI protocols as feasible but does not establish the same crossing-fiber performance at those scan times. (LLM, conf=0.88)

  cites Glenn et al., 2016 sourceExperimental — accuracy not yet measured

In the paper

“suggesting that this is a fundamental problem that cannot be overcome by scanning at higher resolution to reduce partial volume effects (Schilling et al., 2017).”

What the sources say

• Supported

  The Discussion says the crossing fiber problem is a fundamental limitation of diffusion MRI that cannot be overcome by increased spatial resolution. It also reports that crossing fibers remained in more than 50% of white-matter voxels at 32 µm, despite reduced partial-volume averaging. (LLM, conf=0.94)

  cites Schilling et al., 2017 source

In the paper

“other quantitative MRI techniques such as calibrated T1w/T2w ratio mapping (Ganzetti et al., 2014; Uddin et al., 2018), inhomogeneous magnetization transfer (ihMT) imaging (Manning et al., 2017; Swanson et al., 2017), and T2 relaxation-based myelin water imaging MWI (Prasloski et al., 2012; Lee et al., 2018) are likely less affected by fiber orientation”

What the sources say

• Supported

  In the Discussion section on myelin-related information, the paper reports that T1-w/T2-w scores were more uniform than fractional anisotropy (FA) across white-matter structures. It attributes some FA variability to crossing fibers with different orientations, supporting the hedged claim that the ratio measure is less affected by fiber orientation than FA. The cited paper does not address the other techniques listed in the sentence, which have separate citations. (LLM, conf=0.83)

  cites Ganzetti et al., 2014 source
• ?Not checked — no full text of the sourcecites Uddin et al., 2018 source
• ?Not checked — no full text of the sourcecites Manning et al., 2017 source

In the paper

“In fact, increased diffusion anisotropy in the presence of Wallerian degeneration due to chronic lacunar infarcts has been reported (Pierpaoli et al., 2001)”

What the sources say

• ?Not checked — no full text of the sourcecites Pierpaoli et al., 2001 source

In the paper

“disproportionate atrophy or degradation of one or more fiber bundles – along with the relative preservation of other fiber bundle(s) – could result in a seemingly paradoxical increase in FA, despite an actual decrease in local fiber density, myelination, etc., (Figure 2A).”

In the paper

“this makes the conventional “if some is good, more is better” interpretation of FA values much more problematic than originally thought.”

What the sources say

• ?Not checked — no full text of the sourcecites Tuch et al., 2003 source

In the paper

“which renders these metrics highly sensitive to the relative volume fractions of the various fiber populations, the microstructural integrity of each constituent fiber population, and the interaction between these factors.”

In the paper

“For simplicity, we have focused on FA in our examples, but it should be noted that AD and RD are confounded for similar reasons.”

In the paper

“These examples hopefully highlight how biophysical interpretations of FA, AD, and RD values are problematic in the presence of white matter fiber crossings”

In the paper

“the opposite effect can also be true – where FA values could decrease due to a disproportionate increase in one or more of the non-dominant fiber bundles (e.g., during neurodevelopment, neuroplasticity and/or neural repair).”

In the paper

“how interpreting them to reflect similarity, stability, differences or changes in tissue microstructure can potentially lead to Type I (false positive) and/or Type II (false negative) errors.”

In the paper

“real differences could be missed if the underlying fiber bundles in a given region differ in the same way or change at the same rate, relative to each other.”

In the paper

“Indeed, anytime there is a proportional change in the three eigenvalues (λ1, λ2, and λ3), FA will remain constant.”

In the paper

“certain DKI metrics such as kurtosis fractional anisotropy (KFA) and quantitative kurtosis tensor measures (e.g., radial tensor kurtosis) are likely more robust to crossing fibers (Hansen et al., 2016; Hansen and Jespersen, 2016).”

What the sources say

• ?Not matched

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

  cites Hansen et al., 2016
• ?Not matched

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

  cites Hansen and Jespersen, 2016

In the paper

“we would suggest that either the MD (i.e., mean apparent diffusion along the three tensor dimensions; MD = [λ1 + λ2 + λ3]/3) or the Trace (i.e., total apparent diffusion along the three tensor dimensions; Trace = [λ1 + λ2 + λ3]) values are likely the most robust and interpretable metrics for drawing reverse inferences about the underlying tissue characteristics.”

What the sources say

• ?Not checked — no full text of the sourcecites Genc et al., 2017 source

In the paper

“In both of the theoretical thought experiments outlined in Figure 2, MD and Trace values would increase as expected (owing to less restricted diffusion within the voxel).”

In the paper

“if a multiple sclerosis (MS) lesion or traumatic brain injury damages all of the underlying fiber bundles within a particular region equally (e.g., increased membrane permeability and decreased myelination leading to an equal reduction in λ1, λ2, and λ3), FA will appear unchanged despite potentially significant alterations in the underlying tissues (Figure 2B).”

In the paper

“This is also supported by empirical findings that MD is more closely related to neurite density than FA, AD or RD measures (Genc et al., 2017).”

What the sources say

• ?Not checked — no full text of the sourcecites Genc et al., 2017 source

In the paper

“increases, decreases, and indeed even stable FA, AD, and RD measures become difficult (if not impossible) to interpret in terms of the other underlying tissue microstructural properties (e.g., fiber diameter, fiber density, membrane permeability, myelination, etc.).”

In the paper

“This does not, however, mean that all of the previous interpretations using these metrics are necessarily wrong – especially in cases where the findings have been corroborated using other imaging metrics.”

In the paper

“there are obviously certain white matter regions with very few crossing fibers (e.g., corpus callosum, corticospinal tract, spinal cord, etc.), where the interpretation of these metrics is more straightforward.”

In the paper

“the opposite effect could also be true during neurodevelopment, where FA could appear constant if the all of the constituent fiber bundles were to mature at the same rate (i.e., equal increase in λ1, λ2, and λ3).”

In the paper

“Therefore, there is no way of controlling for these effects or estimating the prevalence of Type I and Type II errors (which could even occur simultaneously in different brain regions) inherent in any reverse inferences between FA, AD or RD and the underlying tissue microstructure.”

In the paper

“one might erroneously infer that equal FA values reflect equivalent underlying tissue composition. However, as in the example above, this is not necessarily the case in the presence of crossing fibers either.”