Problems with obtaining a Structural Connectome

,
1

Hello, MRtrix support team,
I am working with MRtrix to get a structural connectome based on data that was provided for me by some professionals. What I need is the mean length of fibers between 85 ROIs and the mean number of fibers between them, for which I use the following commands respectively:

tck2connectome -symmetric -zero_diagonal wb.tck parcellation.mif -stat_edge sum number.csv
tck2connectome -symmetric -zero_diagonal wb.tck parcellation.mif -stat_edge mean -scale_length length.csv

My very first question is whether these commands would work for my intention?

Secondly, the results I got from these two commands had a bit of a problem, compared with literature, my numbers were off. My maximum mean number of fibers was a thousand times bigger and the minimum mean length of the fibers six times smaller than the respective values in the literature. I know that working with just one brain, the numbers might slightly differ from working with mean values form many brains but my numbers are orders of magnitudes larger. My question is can you help me find what might be the underlying cause of this problem?

I tried to understand why this phenomenon happened by going one step back and generate my own tractography, I limited the minimum length of the fibers to 10 mm and I found out a region is losing all it’s connections, the temporal pole. I lowered the limit step by step until at 3 mm minimum I got the connection in that region back. The problem is the distance between the center of the mass of this region and the connected regions at this minimum (obtained by mrcenter code I found on this forum) is between 11 to 45 mm and the length of the streamline I get is around 4-5mm. I am wondering why this is happening?
The command I’m using to get my tractography is:

 tckgen fod.mif wb_mn3mx130.tck -act aparc+aseg_5tt_al.nii -seeds 5M -minlength 3 -maxlength 130 -seed_gmwmi gmwmi_5tt_al.nii

I am very confused and I’m not sure what should I do next.
Thank you in advance

2

Hi @Shima_Jalalian,

My very first question is whether these commands would work for my intention?

Yep. The “-stat_edge sum” in the first command invocation is not strictly necessary since sum is the default edge-wise statistic, but given you want to differentiate between the two there’s no harm in being explicit.

My maximum mean number of fibers was a thousand times bigger … than the respective values in the literature

Streamline count is non-quantitative; comparing such across different publications is even worse. If you generate more streamlines in your whole-brain tractogram than somebody else’s experiment, you’ll have larger streamline counts in your tractogram.

Even if you bypass that issue and are looking at e.g. fractions of whole-brain streamlines assigned to a particular pathway, there are still so many variables that may differ between experiments that could drive huge differences in results.

So my general advice here would I think be to validate as far as is possible that your own data are correct, that there are no egregious issues with your reconstructions, and then consider what you can conclude from those data and what you cannot. Comparing connectivity values across publications would require contemplation of every single variable differing between those two experiments.

the minimum mean length of the fibers six times smaller than the respective values in the literature

That’s slightly unusual, but my up-front advice would be the same as above; i.e. rather than looking at the final output values and being concerned by their magnitude, look at the raw data and make sure that there are no fundamental issues with where those values are coming from. An observation like this:

I limited the minimum length of the fibers to 10 mm and I found out a region is losing all it’s connections, the temporal pole. I lowered the limit step by step until at 3 mm minimum I got the connection in that region back.

reinforces the need for such care.

The problem is the distance between the center of the mass of this region and the connected regions at this minimum (obtained by mrcenter code I found on this forum) is between 11 to 45 mm and the length of the streamline I get is around 4-5mm. I am wondering why this is happening?

Because parcels are not infinitesimally small points, there can be streamlines connecting them that are much shorter than the distance between their centres. Indeed in the limit where two parcels are immediately adjacent to one another (which happens at every parcel border in a whole-brain parcellation by definition), if one removed the streamline minimum length criterion entirely, one could potentially generate streamlines containing only 2 vertices, i.e. the length of one streamline “step”, that connects two parcels. This is even more prevalent because you are seeding at the interface between grey and white matter, which tends to slightly over-emphasize reconstruction of such pathways.

Since you are using ACT, the default minimum length is 2 voxels. Other softwares / published works utilise minimum lengths of up to 25-30mm. This may produce nice-looking reconstructions of the major bundles, but is outright precluding the reconstruction of short connections, which, as I have demonstrated (and argued correlates with biological reality), are exceptionally prevalent.

So one could imagine in the most extreme case that the publications that you are citing have reconstructed, for adjacent nodes, only erroneous trajectories that project deep into the white matter, do a 180, and return back to the adjacent node, whereas your data are generating the actual very short superficial white matter connections between those adjacent grey matter areas. After all, one would expect that the brain would want to construct such connectivity using minimal resources and imposing minimal conduction delay, right?

Food for thought!

Rob

1 Like