I’m trying to perform afd measurement on a given .tck extracted from a whole brain tractography, by using the command afdconnectivity.
I lauched this command two times for the same .tck, with and without the option -wbft.
Besides the extremely different time of processing (it was very long when using the option -wbft), results were totally different (for example 10.7 without -wbft and 0.003 with -wbft).
I would like to ask more clarification about how -wbft command works and I would like to understand if this big difference can be considered expected when applying the entire whole brain dataset on a given .tck.
My dataset consists on whole brain tractography (10 million streamlines) obtained after conventional pre-processing steps (motion correction, etc.).
Thank you in advance.
In case you missed it, you can find more information on the
afdconnectivity tool here, and in a previous thread.
As suggested in the documentation, if you have a way to perform EPI distortion correction, then it’s recommended you try SIFT. From a theoretical point of view the
afdconnectivity tool can be thought of as a poor mans SIFT. However practically it has been quite useful in several studies, including this recent paper.
Basically the -wbft option tries to make afdconnectivity more SIFT-like (by factoring out the AFD in each fixel that may be contributed from streamlines in merging fibre pathways, which are co-located but you are not interested it (i.e. they are not part of the streamlines in your tract-of-interest)). However without using SIFT, the proportion of the AFD in each fixel that is assigned to each pathway is not free from tractography biases. In fact we recently used the
-wbft option on a study and saw an unexplained increase in a control population. This increase went away without the
-wbft option, which we think was explained by a particular tractography bias in the clinical population. So personally I prefer the more straight forward interpretation of
afdconnectivity without the
-wbft option, even if this means the measurement may be influenced by other pathways occupying the same fixels. However Rob may disagree with me on this one.
It’s also worth noting even though SIFT is more ideal theoretically, tractograms are still full of false-positive streamlines after SIFT, which will affect the connectivity estimates from counting streamlines. And so depending on how accurate the whole-brain tractography is around your tract-of-interest, taking a more simple direct measurement of the AFD within your tract-of interest may be just as (or possibly more) informative in terms of detecting group differences. However, these are just my thoughts, and I have absolutely no data to back them up.
Just a word of warning, before using
afdconnectivity you need to make sure the DWI images are bias corrected and intensity normalised as explained here. These instructions work well, however we hope to release a new command to improve on these steps soon. So keep an eye on the announcements.
Besides the extremely different time of processing (it was very long when using the option -wbft) …
Yes, processing time is necessarily longer with the
-wbft option, since an entire whole-brain tractogram must be mapped to the image, not just the streamlines corresponding to your pathway of interest.
… results were totally different (for example 10.7 without -wbft and 0.003 with -wbft). I would like to ask more clarification about how -wbft command works and I would like to understand if this big difference can be considered expected when applying the entire whole brain dataset on a given .tck.
The derived connectivity values will most certainly not be comparable between using and not using that option. The latter will always be substantially smaller:
-wbft option, all of the AFD in every fixel traversed contributes to the total pathway fibre volume.
-wbft option, only a fraction of the AFD in each fixel is added to the pathway fibre volume. That fraction depends on the fraction of the tractogram streamlines density in that fixel that belongs to your pathway of interest.
I’m hoping to publish a technical note at some point explaining how all this works, what each technique provides, and why SIFT / SIFT2 is in fact the preferable solution (in an ideal world where streamlines tractography actually works ).
Hi David and Robert,
thank you very much for your clear explanation.
I’ll try to do some tests and, if something interesting were to come out, I’ll tell you.