TDI Unit

When I create a TDI image, I can generate a value, either by cursor placement or via drawing a ROI, for a specific region. What is this generated value. Is it a fibre density or fibre count value which is generated ?

Thanks in advance.


Also, does MR Trix use a deterministic or probabilistic tractography model ?


Hi @sahanca,

Streamline count, which is potentially biologically irrelevant and has a strong seeding bias, if the tractogram is not, e.g., SIFTed. If you’re after the (apparent) fibre density, it should be obtained directly from the FOD from CSD (even better: from multi-tissue CSD). See also this article for more insights: .

Any you wish! :smile: :wink: The default algorithm in tckgen is iFOD2, which is a probabilistic tractography algorithm; but others can be selected as well.


Thanks so much for your help.


1 Like

Also, does MR Trix use a deterministic or probabilistic tractography model ?

Given similar questions have been raised in the past, I went ahead and gave a brief description of each algorithm in the tckgen help page.


I have a question about TDI. I created ROI s using TDI image. I extracted the values of 20 ROI s and want to compare these values between two groups and the same way I want to compare FAvalues in the same ROI s. Is this a good idea to compare TDI between two groups? Tckgen has been run with 5M streamlines. I have created ROIs using JHU tract based atlas. Could someone please explain is this method is valid to obtain good results? These ROI s were created using FSL. I am kind of new for the MRtrix. Also I have a doubt about the units of values I extracted in that way. Does TDI has a unit? I am sorry for the basic questions. But I couldn’t figure out muself. Thank you in advanced.


Hi Shane,

As far as quantitative use of TDI, I would recommend reading this article. Tl;dr: Don’t.

The “units” of a TDI is literally a streamline count: the integer value in each voxel is just the number of streamlines traversing that voxel1. Take the mean value within a ROI, and the value is the mean number of streamlines traversing any voxel in that ROI. So obviously that quantity depends on a great deal of other properties, both visible and latent, and both local and distant, which is part of the reason why it’s not advocated for quantitative use: any statistical difference would be nigh impossible to interpret.


1 Exception is if you use the -precise option in tckmap, which activates the method described in Appendix 3 of this manuscript. In that case, the TDI is calculated in units of mm: it is the sum of the lengths of streamline intersections within any particular voxel that is quantified.

1 Like