I would like to calculate AFD values for several tracts. I know there are a lot of discussion on the forum already, but I was hoping someone could tell me specifically how to calculate AFD, as I am still not 100% sure about this. I ran the intensity normalisation across subjects, did the bias field corrections and calculated a group average response function. My questions is after running tcksift2, what exactly do I then do to get the AFD values? Do I just sum the cross-section multipliers for each streamline that are in the output textile or do I add them up and then divide by streamline length?
At that point, just summing of the weights does the job (or just track counting in case of SIFT1; those all have the same weight in that case). The crucial insight to gain is that the streamline length is, at that point, not weighting the result itself (and hence should not be divided by any more). SIFT(2) makes the streamline counts/weights match the underlying AFD, but there’s only 1 streamline/weight per streamline, however short or long that streamline is. The streamline, after SIFTing, tries to fit, on average the underlying AFDs; not the sum of them along the streamline length.
Not sure if that’s a clear explanation… @rsmith may be able to explain it from other points of view.
Great, that’s all I needed to know. Thanks for clarifying this!
@ThijsDhollander got it, but just a warning on nomenclature: AFD is a measure of fibre volume, whereas the streamlines count after SIFT / sum of streamline weights after SIFT2 is a measure of fibre cross-sectional area. This factoring out of the streamlines length is important from both physical and biological perspectives, and also in the way the result is described; i.e. I would not refer to such a quantity as “The AFD of the tract”, since that statement would imply a volume measurement. “Connection density” probably suffices, unless we can come up with something better: “Intra-cellular cross-sectional area” doesn’t really roll off the tongue…
I know what you’re referring to of course, but I must say that I’ve found even this terminology has been causing quite a bit of confusion lately (in the way it also pops up in e.g. the description of the
afdconnectivity command: “a measure that is more related to the cross-sectional volume of the tract”).
This confusion, I have found, seems to arise from the (“clash with”) terminology we use for (fixel-)based analysis of fibre density (FD), fibre cross-section (FC) and the combined metric FDC. The way “cross-sectional volume (or area) of the tract” is mentioned in a SIFT / SIFT2 /
afdconnectivity context, seems to refer to FC (because it sounds like it), but its “equivalent” is actually (much closer to) FDC; but then on average for the whole tract; i.e., SIFT(2) factors in both the “FD” per voxel that it (aims to) fit, as well as the “width” (cross-sectional area on a macroscopic scale) of the bundle. Both reductions (at some point along a given tract) of FD and/or FC would cause a reduction of the “connection density” probability that is the output from SIFT(2) for that given tract.
So as the terminology is maturing, I’d personally try to aim for something that sounds a bit more like, or at least drives users’ intuition towards, the “FDC” concept, rather than sounding too much purely like the “FC” one. Or in other words, reserve the terminology of “fibre cross-section(al area) (FC)” for the pure macroscopic effect: this is actually what we’ve been doing now in some recent submissions and publications if we were facing the challenge of describing “FBA of FD and FC” in more general terms for the sake of an abstract (of a paper): using terminology like “microscopic” and “macroscopic” to refer to the specific nature of the 2 separate properties we can extract and compare.
Self-contradiction in 2 words Let’s not base too much off of that description…
This confusion, I have found, seems to arise from the (“clash with”) terminology we use for (fixel-)based analysis of fibre density (FD), fibre cross-section (FC) and the combined metric FDC.
The issue in my eyes is actually the FC metric. This isn’t a “fibre cross-section”, but a “change in fibre cross-section”. I’ve tried to clarify this where I can, and would have preferred if it were part of the “official definition” of the term partly for this reason, but a compromise was needed so as not to bloat the FBA terminology.
So as the terminology is maturing, I’d personally try to aim for something that sounds a bit more like, or at least drives users’ intuition towards, the “FDC” concept, rather than sounding too much purely like the “FC” one.
I’d hesitate about explicitly comparing it with FDC, since the similarity is incomplete and would be of little use to anyone doing connectomics but not FBA. If you have a good understanding of both concepts, then absolutely it’s closest to FDC; but I’m not sure if drawing the parallels helps or hinders education of the concept.
The important points to convey are that with tractography we get a total fibre connection density (as opposed to a cross-sectional change), of the pathway (as opposed to individual fixels), and that this quantity is proportional to fibre cross-sectional area rather than volume (as opposed to FC, which is a change in cross-sectional area that modulates a volume) and is therefore invariant to changes in length.
And all preferably in four words or less.
… Does the forum have a voting feature?
I wouldn’t really let this evolve based on a popularity vote. For those interested, I’ve created a GitHub “issue” about this. It’d be good to get some constructive discussion going, but this thread was probably not the ideal place to kick it off… (my bad ).