Calculating average FBA metrics of specific tracts

[smeisler]

Hi,
Sorry to revive an old thread. The prior posts here are great but I just had a question with a single point:

@ThijsDhollander

Threshold this fixel-wise TDI image using mrthreshold with -abs ... and some number of minimal streamlines that need to have passed through the fixel for the fixel to be “in” the binary fixel mask. This is the step where you can play with this threshold to get a nice continuous, clean and solid “bundle” in your final binary fixel mask, even when there were a few spurious false positive streamlines before. This will then output the binary fixel mask (similarly as above in the example of thresholding the p-value fixel image).

Is there a problem with thresholding at 1? I like my track segmentations from TractSeg and don’t see what look like false positives. Since TractSeg seems to be the primary go-to segmentation method accompanying FBA, I was hoping someone can share their approach.

Thanks,
Steven

[rsmith]

The process by which TractSeg generates streamlines, involving both a voxel propagation mask and dual endpoint masks, means that stray streamlines are precluded from the outset. So yes, in that case you’re likely fine using a threshold of 1 streamline.

[adsouza]

Hi Thijs,
Thanks for sharing this explanation. I tried implementing the following on my dataset:

mrstats subject001.mif -mask tractOI_binary.mif

which provides an error, because the mask file has dimensions 204 x 204 x 204 while subject001.mif has dimensions 571734 x 1 x 1.

(here, subject001.mif is found in the FD_smooth directory, although the same could be said if using any of the subject images in the FD directory).

I also tried using mrgrid to match the dimensions of tractOI_binary.mif with subject001.mif, and this resulted in an image containing only zeros…

Am I using the correct input images? And if so, what can be done to get a sensible result?

Cheers,
Arkiev

[adsouza]

I think I found my error. The -mask should use the output from tck2fixel (and not the output from tckmap)…

Thanks again!

Cheers,
Arkiev

[hui_hong]

@KJue

Hi Thijs,

What’s the best value for mrthreshold? -abs what value? I could not find any articles mentioned this.

[hyting]

Hi, Arkiev

I encountered the same error as yours, the dimensions between my mask and FD.mif are mismatched. The details I have pasted in here. But I seem to use the mask from tck2fixel already(I am not that sure). Could you please help me check?

Thanks in advance!

Best,
Yuting

[adsouza]

Hi Yuting,

Thanks for sharing the link. The syntax of tck2fixel seems sensible. But I’m not sure if the application here is necessary…

I would like to clarify the objective… it looks like a whole brain FBA analysis has been completed, including stats (FD anova). Significant fixels in the whole brain analysis have been found, and the goal is to calculate the proportion of a given tract that contains significant differences in FD as determined from the whole-brain FBA…is this correct? If so, I think this could be done using voxels, for example, mrthreshold the FWE_1mp at 0.95 to create a binary image showing significant voxels. Then do mrstats on this image using -mask (here the mask would be a binary image from the tract of interest, in voxel coordinates) to get the number of voxels within the tract that contain significantly different voxels from the whole-brain FBA.

This is different to asking whether there are differences in FBA metrics in a given tract of interest. To do this, run fixelcsfstats using the -mask option (here the output of tck2fixel would be used as the mask). This will run stats but only within the tract of interest.

I should point out that I am relatively new to fixels…I’d be keen to hear what others have to say about this suggestion

Cheers,
Arkiev

[hyting]

Hi, Arkiev,
Thanks for your quick reply.

@adsouza

I would like to clarify the objective… it looks like a whole brain FBA analysis has been completed, including stats (FD anova). Significant fixels in the whole brain analysis have been found, and the goal is to calculate the proportion of a given tract that contains significant differences in FD as determined from the whole-brain FBA…is this correct?

Yes, that’s exactly what I want to do. But I’m doing this to locate the significant fixels in tracts so that I can get the fixel mask of these tracts to calculate the mean FBA metrics of these tracts. With the mean FBA metrics, I can do the post-hoc test between groups. Following your suggestion, I can also get the significant tracts, then I should do voxel2fixel to get their fixel mask, am I understanding correctly? Does this make sense?

Cheers,
Yuting

[hyting]

Hi Arkiev,

I just found out that @smeisler has answered my confusion on GitHub. The syntax of tck2fixel is correct in my code. It turns out that I can simply ignore the mismatch with index.mif and directions.mif, and only focus on the FBA metrics.

Thanks for your help!

Best wishes,
Yuting

[adsouza]

That’s great, Yuting - thanks for letting me know.

Cheers,
Arkiev