I used Mrtrix3 to perform tractography and the command “tck2connectome” to build the connectome of my patient. I generated both connectome using the metric “count” and using “mean_scalar” associated with a FA map.
Out of curiosity I plotted for each connections its value in the connexion matrix using the metric count versus its value in the connexion matrix using the FA mean. In my simplicity I had imagined to obtain an almost linear relation between the two metrics (with of course some noise). But I finally obtained a kind of Gaussian distribution https://drive.google.com/open?id=0B1FyzS22KQJzZkJzOTdMWkVzLW8, I am quite surprise.
Is it normal? What do you think about it?
Thank you for your answer and thank for providing MRtrix3 it is a amazing tool !
Your scatterplot is precisely what I would expect to result from such an experiment. Unfortunately it’s difficult to make the following point in a delicate fashion, but it’s an extremely important fact that I think is worth expressing as clearly and unambiguously as possible:
FA is not proportional to fibre density.
I won’t go into the full details here, since it’s a reasonably lengthy discussion and there are plenty of review papers out there where the point is made with far more clarity than I can churn out in a hurried fashion. I would recommend starting with Tournier et al., 2011 and Jones et al., 2013. Then, once you decide that obtaining a genuine quantitative measure of white matter connectivity is crucial for connectome construction, take Smith et al., 2015a for a spin
Thanks a lot for your answer, I will definitively take a look to these papers. I already use SIFT but I will look at your paper to more into details .
I have one other question, to build the connectome by counting the fibers number, I normalized the count value by the product of the two cortical surfaces, but maybe tck2connectome already implement a kind of normalized value. What is the exact meaning of invnodevolume and invlength_invnodevolume? Sorry, may be it is already documented but I didn’t find it.
Typically the heuristic approach you’re referring to involves scaling the strength of each edge by the reciprocal of the sum of either the two node surface areas or their volumes, as originally described in Patric Hagmann’s 2008 paper. The -metric invnodevolume option will perform the latter scaling. The -metric invlength_invnodevolume option will perform both this scaling, and scaling the contribution of each streamline by the reciprocal of the length of that streamline.
I probably haven’t documented them in detail because I don’t condone their use (I’d say “I really need to write that paper” but I’m starting to sound like a broken record)
Note also that the command-line interface for accessing these functionalities will change in the not-too-distant future. Nothing drastic, just consult the help page if you perform an update and suddenly your command doesn’t work any more.
Sorry for being a bit slow on this: post-ISMRM cold :-/
I’m soon going to add a page to the documentation discussing the heuristic connectome scalings, since the question has come up a number of times recently. That will give more detail than what I would otherwise go into here.
Hopefully you caught the blog post yesterday; I ended up putting it there rather than in the documentation since it’s more of an opinion piece than hard truth. But hopefully that gives enough information for you to ponder.
.
(I’d say “I really need to write that paper” but I’m starting to sound like a broken record)