I have a mask of a sub-region of the amygdala and I want to determine streamlines between that mask and the some other atlas regions. What is the best way to go about doing this?
Probably start by specifying both the amygdala sub-region mask as well as an atlas region as two -include regions, and repeat for each atlas region. You could also streamline (no pun intended) the process via tck2connectome with all atlas regions + the amygdala sub-region at once, and specify the -out_assignments option to get the assignments. Based on the latter, you could select the streamlines between the amygdala sub-region and each atlas region you’re after. I’m not sure what format the -out_assignments option exactly is in though; @rsmith can probably help you better with that one.
If you specifically want to extract those streamlines from one parcel to all / a large number of other parcels, then manipulating the data as a “connectome” is the more robust & efficient way of doing so.
Say, for instance, you have a parcellation image where the amygdala of interest is node number 41. You then want those streamlines that connect from that region to every other region, such that, for example, connections from the amygdala to the entorhinal area (let’s say, parcel number 5), are stored within one track file; connections from the amygdala to some other parcel are written to another track file, and so on.
tck2connectome tractogram.tck parcellation.mif connectome.csv -out_assignments assignments.csv
will give you the vital file assignments.csv, which encodes which two parcels each streamline was assigned to.
Now, to separate the tractogram into separate files, one for each connection from the amygdala to another parcel:
connectome2tck tractogram.tck assignments.csv edge_tracks_ -nodes 41 -files per_edge
From this, you will get a series of files:
edge_tracks_41_1.tck
edge_tracks_41_2.tck
edge_tracks_41_3.tck
edge_tracks_41_4.tck
edge_tracks_41_5.tck
...
Each of these files contains the tracks from parcel number 41 (the amygdala) to another parcel, where the target parcel is captured within the name of the file. So connections from the amygdala (41) to the entorhinal area (5) would be stored in file edge_tracks_41_5.tck.
connectome2tck is fairly powerful and flexible in how it can capture & store specific tracks of interest based on assignments of streamlines to parcels; but it takes a deep understanding of exactly what you’re trying to achieve with it in order to configure it properly. Though it may well be a bit of fun to just play around with the options and see what it gives you.
Rob
1 Like
Hi @rsmith,
I was wondering about the accuracy of this particular approach, i.e. generating a complete whole brain connectome and then specifically selecting connections from the amygdala to all other nodes using connectome2tck. I experimented with the HCP connectome tutorial on the site that uses the Desikan atlas and in the csv output file, I looked at connections between the amygdala and the lateral & medial OFC regions as part of a quality check. As the uncinate fasciculus is known to connect those regions, I expected to see substantial no. of streamlines. However, while I had substantial results for the left amygdala (~350 to the L_lateral OFC and ~1200 to the L_medial OFC), the right amygdala had 0 streamlines tracked to the R_lateral OFC and 7 to the R_medial OFC.
As a consequence, I am not sure if I trust the connectome produced as a whole. I did 100M streamlines sifted to 10M so I don’t think it is a result of having too few streamlines.
Do you think there may be some explanations for this or perhaps improvements I could try?
Thanks.
I looked at connections between the amygdala and the lateral & medial OFC regions as part of a quality check. As the uncinate fasciculus is known to connect those regions, I expected to see substantial no. of streamlines. However, while I had substantial results for the left amygdala (~350 to the L_lateral OFC and ~1200 to the L_medial OFC), the right amygdala had 0 streamlines tracked to the R_lateral OFC and 7 to the R_medial OFC.
Yep; that happens. However looking at an absence of connections won’t tell you much experimentally. This is where looking at all of the connections emanating from a particular ROI to the rest of the brain may be helpful: You can see if the streamlines you would biologically expect to go from node A to node B are instead going to adjacent node C, or if there’s some node D in between A and B that’s causing the streamlines to terminate prematurely, or if there’s some fault in the image data or tissue segmentation or image alignment that’s preventing streamlines from going where you expect them to, or if there’s some other cause of the unexpected asymmetry between the hemispheres.
As a consequence, I am not sure if I trust the connectome produced as a whole.
There are plenty of articles bashing the fundamental accuracy of streamlines tractography that agree with you. We can only do the best we can with the data we have. 
2 Likes
Hi @rsmith,
Thanks for your reply. I did have a closer look again at the data such as overlaying the images on each other etc. However, as these were all part of the HCP single subject data which was used in the tutorial on the mrtrix website, I don’t think the preprocessing steps were the issue. I followed the steps in the tutorial and am wondering what results people got when they ran this tutorial (or if you happen to have a copy of the output from conducting the workshop back then).
When I looked at the numbers corresponding to nodes of the OFCs and the amygdala, I noted that the left and right lateral OFCs had ~100+ streamlines going across to one another in both hemispheres which was rather weird, the streamlines from the lateral OFCs also seem not to be able to reach areas such as the temporal pole and amygdala for which there is strong biological evidence for. There are streamlines that hit the superior temporal area (~90-100) but that’s about as far as it goes within the temporal lobe. I am now wondering if it is an issue with the tracking parameters since it seems to not be able to reach these temporal areas.
My command for running tckgen was: tckgen WM_FOD.mif 100M.tck -act 5TT.mif -backtrack -crop_at_gmwmi -seed_dynamic WM_FOD.mif -maxlength 250 -select 100M -cutoff 0.06 . Should I be specifying the step and angle to get better results? I am not sure if masks should be specified given this is a whole brain tractography approach.
Would really appreciate any advice I can get regarding this issue.
Should I be specifying the step and angle to get better results?
You can certainly experiment with those: The default of 1/2 voxel step size and 45 degrees per step is fairly short compared to other data due to the smaller voxel size, using 1.25mm step size and 45 degrees per step would be closer to the “standard” “default”. But there really is no correct answer as to what these should be.
I am not sure if masks should be specified given this is a whole brain tractography approach.
For whole-brain tracking with ACT, assuming you don’t want to exclude any white matter, the best option is to not use any mask.
Also, on revision:
I was wondering about the accuracy of this particular approach, i.e. generating a complete whole brain connectome and then specifically selecting connections from the amygdala to all other nodes using
connectome2tck.
My suspicion is that the issue you describe is generic to streamlines tractography, and is not specific to the approach of generating whole-brain streamlines and subsequently selecting tracks. If you were to perform targeted tracking with both seed and target regions, this would simply be obscuring the large number of generated streamlines that connect only one of those two regions, which is precisely what you’re seeing in the whole-brain data.
Rob