I'm wondering about the relationship between FA and other scalars (AD, MD, RD) in DTI. To my understanding, FA is a summary measure of microstructural changes, while the other scalars are more specific measures of white matter microstructure. Is it possible that one can observe no differences between two groups in FA, but significant differences in the other scalars?
In my study, I compared FA values between two groups, and found no significant differences. Should I then analyze the other scalars as well, or should there not be any differences in these values as well?
All help appreciated!
Hi Lasse,
In my understanding, it is completely possible. The math to calculates the FA considers all three eigenvalues L1, L2 and L3, resulting in a 'dimensionless' value that weights how directional is your voxel. The absolute values (how big or how low) in this case, does not matter, only the relations among the Ls it is important.
The MD, AD and RD. directly quantifies the diffusion (mm²/s) by three simple, direct and distinct ways: MD the mean of all Ls; The AD the single value of L1 (higher L) and RD, the average of the 2 lowers Ls.
In this sense, we can have equals FA between groups, but in the case we have a bigger absolute L1 L2 and L3 for one of them, the MD may differ (but not necessarily). Similar to the RD and AD.
This is what I understand, I hope it helps a bit.
The answer is Yes! To put it in a simple way, we could say that FA represents the disproportion between axial (L1) and radial diffusivity ((L2+L3)/2). So if axial and radial diffusivity increase or decrease in terms of absolute value but their ratio remains the same then FA will not change significantly. My advice is to always look at all four DTI metric (FA, MD, Axial and radial diffusivities), or at least at FA and MD. Because it can happen that FA does not change while MD does or the contrary. It all depends on the kind of microstructural changes you are dealing with,
As the other two posters noted it is entirely possible for FA to not change, while one of the three other values are different.
Dear Lasse,
Not only from a theoretical point of view provide RD/AD and FA different information. Especially neurodegenerative diseases have a strong effect on both RD and AD in areas with predominantly bundled (non-crossing) fibers. This leads to higher changes in RD and AD but only moderate change in FA. In contrast, demyelinating diseases like (early) remitting relapsing multiple sclerosis predominantly change RD but virtually not AD, resulting in a relatively higher effect on FA.
Indeed, I have found this in a number of studies. For example, please see my 2014 publication on this (attached).
Dear Professor Michael Deppe sorry for this basic question I have AD MD RD and AD images. However, I am only sure the values of FA. How can I be so sure about my other scalers' values are not wrong? How can I check this? Best.
Dear Tugce,
you are asking about values that are "not wrong". In stochastic terminology I think you mean "are within a certain normal range". Such a "normal range" can be, for example, the value range between the 5th and 95 percentile of a distribution of FA, RD, AD, etc values acquired on a healthy control group. In Münster, we calculate DTI scalars within certain regions of interest (ROI, e.g. corpus callosum, brain stem, thalamus, ...). In our research it turned out, that the corresponding distributions of normal FA, RD, and AD ROI values are relatively narrow, which means that pathological alterations can be (easily) detected as they are far below (FA) or above (RD, AD) in patients compared to the values assessed from "healthy" controls (for an example for patients with MS, see the paper Deppe et al, 2016, Human Brain Mapping). An alternative approach can be voxel-wise statistics of (spatially normalized!) FA, RD, FA, and MD maps (for example see FSL tools). And there are many more good approaches. A completely different story would be if the values were actually wrong in sense of computational mistakes. But I don't think that this was your question. Did my answer help?
If not, don't hesitate to ask.
Cheers,
Michael
Dear Professor Michael Deppe
This was very helpful and "more" from my question! Thank you so much, I learned much from your answer! And I will definitely read your paper. In fact my other question is much more basic one. I know how a All_FA.nii.gz image look like in FSLview, for instance, I can guess before I put the image on it. However, the other ones are so similar look. When I look first, I can not say this is All_AD.nii.gz or this All_MD.nii.gz or this is All_RD.nii.gz. Do you have a tip to understand the differences among these?
Best wishes,
Tuğçe
Dear Tugce Cabuk ,
Roughly (!), the AD is the water mobility along fiber bundles and RD the mobility perpendicular to this direction. This simplified model assumes that the term "bundle" has some meaning in the tissue of interest. Regions where this model is not too bad are for example the corpus callosum and the brain stem. (For gray matter voxels it is not appropriate to speak about "fiber directions".) Consequently, AD has higher values than RD in tissue with fibers. This comes directly from the definition of AD, RD, and MD (see literature). MD is something between AD and RD because it reflects the mean mobility of the molecules averaged over all directions (= diffusion "constant" in isotropic tissue). Thus in your images the AD should show the highest value followed by MD and RD for a certain voxel or region of anisotropic tissue.
Important to note: In contrast to FA, the scalars RD, MD, and AD have physical dimensions. If I remember correctly the AD in healthy anisotropic tissue is something like 1.5 * 10-3 mm2 s-1 while the RD is around 0.7 * 10-3 mm2 s-1. (Please correct me if my memory is wrong.) For comparison: Free water (eg. CSF in in left and right ventricles) has a self diffusion of 3*10-3 mm2 s-1 at 37°. Thus your diffusivity scalars should be in the range of some 10-3 mm2 s-1 and definitely less than 3*10-3 mm2 s-1. Otherwise the units are arbitrarily scaled and meaningless (software problem ?). You may check this by means of an appropriate color lookup table for your images. Did I answer your question?
Greetings from Münster, Germany
Michael
- Badges
- Science topic