Scalar metric-weighted tract endpoint mask

[Valerie Sydnor]

Hi Frank!

I am working with a custom thalamocortical autotrack tract atlas from our paper https://www.nature.com/articles/s41593-025-01991-6 :) The atlas comprises streamlines connecting the thalamus to cortical regions.

My goal is to 1. identify which voxels in the cortex are connected to the thalamus by streamlines in the atlas, and 2. to weight each cortical voxel by the average <scalar metric> of all streamlines that end in that voxel, with <scalar metric> being e.g. QA, MD, ISO, etc. Ideally, the output is a nifti file where each cortical voxel that has streamlines connecting it to the thalamus has a scalar value that is derived by averaging metrics of interest across those streamlines; voxels with no connectivity to thalamus have a 0. I'm thinking of this as a scalar metric-weighted track endpoint image.

I know that step 1 can be accomplished easily in DSI Studio, for example by saving a track density image of tract endpoints. However, I'm not sure how to accomplish step 2, which requires knowing which atlas streamlines end in each voxel and what their average <scalar metric> is. 

Is there any way to accomplish this in DSI Studio? Thanks for your help and guidance here!

Valerie

[Frank Yeh]

I will figure out the best solution for this task and get back to you.
Best,
Frank

> --
> You received this message because you are subscribed to the Google Groups "DSI Studio" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to dsi-studio+...@googlegroups.com.
> To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/d87b019b-8804-49c6-ba21-ec1bf9d99c8bn%40googlegroups.com.

[Valerie Sydnor]

Amazing, thank you in advance!

Valerie

[Frank Yeh]

Hi Valerie,

Thanks for the reminder—!.

This is a more sophisticated problem than working at the voxel level.
I would suggest that first use HCP-MMP regions as the unit, which
simplifies the analysis before going down to voxel resolution. I would
also estimate the pathways at the population level:

For each subject, run --action=atk or trk to map corticothalamic
pathways using each HCP-MMP region as an ROI, and then collect the
tract statistics using --export=stat. This would give us a population
sample of metrics estimated for each HCP-MMP region.

Afterward, the results could be aggregated fairly easily into a NIFTI
file using Python.

Just my two cents.

Best,
Frank

> To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/0ea65bc6-bb65-4481-845f-5ecf4ae6b254n%40googlegroups.com.

[Valerie Sydnor]

Hi Frank,

Thanks for thinking about this and getting back to me! 

I have indeed started with HCP-MMP regions as the unit and created the thalamocortical connections at the population level using the HCP1065.1.25mm.fib.gz diffusion template. I have ~230 individual thalamocortical connections in the initial population tractography atlas. So I could indeed use the --export=stat approach mentioned above to get average scalar metrics in tracts connecting the thalamus to each individual HCP-MMP region at the group level (I don't need individual-level metrics here). I agree this could be a good starting point.

The problem is that I am now filtering all of the streamlins (across the 230 connections) in the thalamocortical atlas, and only a subset is retained. I am intersted in the connectivity profile of the retained streamlines (which cortical/thalamic voxels do they connect to?), which I can obtain from a binarized tract density endpoint image. However, I am also interested in a weighted connectivity profile of the retained streamlines (which cortical/thalamic voxels do they connect to, and how "strongly"? -- based on a microstructural diffusion metric that proxies connectivity stength). This is why I am wondering if there is a way to weight voxel-wise tract enpoint images by a scalar metric in connected streamlines, as opposed to streamline density. Does that make sense? Happy to elaborate more if you think that this may be achievable through DSI Studio. 

Thank you!!!

Valerie 

[Frank Yeh]

I see.
I would break each HCP-MMP region into one-voxel region as NIFTI files
and use the same pipeline.
Gemini AI should be able to write a quick Python script to generate them.
The only challenges is that the computation may take a lot of time...

On Tue, Mar 10, 2026 at 12:57 PM Valerie Sydnor

> To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/b08e5025-929b-464b-83dc-90009bba4acfn%40googlegroups.com.

[Valerie Sydnor]

Thank you for the help here, Frank. I was able to successfully accomplish this at the regional level using thalamic connections to HCP-MMP regions and I am building out code to now do it voxel-wise. Your idea to consider each voxel a "region" and use --export=stat was very helpful!

Best,

Valerie