t-statistics tract rendering

[Mark Vernon]

Hello,

After performing a correlational tractography analysis with a "T Threshold" of 2.5, I loaded the t_statistics.fib.gz file and opened the inc.tt.gz tract file then selected "Tract Rendering" -> "Color" -> "Local Index" and selected "inc_t" as the Index.  

After rendering the t-statistics values onto the inc.tt.gz tracts, I would expect that only values greater than 2.5 (the threshold I set) would be shown as only fibers with a correlation which yields a t-stat greater than 2.5 would be used in the subsequent tractography.  But after setting the Local Index to "inc_t", the Max value to "2.5", and the Min value to "0.0", I see tracts being rendered with what appear to be values less than 2.5.  I've attached a screenshot for reference.

I think that I'm missing something in how the t-stat threshold is used in the tractography after the correlational analysis is completed.

Thanks so much,

Mark

[Frank Yeh]

I will check and look into this problem.

--
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 on the web visit https://groups.google.com/d/msgid/dsi-studio/8f662960-242f-430e-a52f-db25742655e8n%40googlegroups.com.

[Mark Vernon]

Would it be that there are dispersed fibers which have a correlation factor equal to or greater than the t-stat threshold  (in this case 2.5), and then the tractography attempts to connect these fibers but there are still sections of the tracts between these greater-than-threshold fibers with fibers that were below the t-stat threshold?  Or is the tractography algorithm set to only trace fibers which had t-stat correlation above threshold?  The practicum website (https://practicum.labsolver.org/ct.html) states that the termination criteria for fiber tracking is determined by anisotropy threshold, angular threshold, and the t-stat threshold.  

Thanks for all of your help understanding this,

Mark

[Frank Yeh]

Sorry for my delayed checking this issue.

I hope to investigate this and get back to you soon.

Frank

To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/3d4df08f-2ac2-46e0-8700-df9017b3f52dn%40googlegroups.com.

[Wenjun Huang]

hello, Frank

I had the same problem, I specified T-threshold=2.5.  I loaded the t_statistics.fib.gz file and opened the inc.tt.gz tract file then selected "Tract Rendering" -> "Color" -> "Local Index" and selected "inc_t" as the Index.  I got the following chart:

and if I set min value to 2.5, the tracts become dark blue.

Other than that, I have a couple of questions about correlational tractography that I'd like to ask
1. When doing correlational tractography between QA and behavioral variables，the value of fdr is reported in the web report only if fdr is not set in the parameters. If I set fdr=0.05 and the web report only has whether fdr is greater or less than 0.05, how do I get the value of fdr in this case?

2. what exactly does t-threshold mean? I find in the official website that Higher values will map tracks with a more substantial correlation effect, whereas lower values for a weak correlation. But I don't fully understand it yet. Is it considered statistic? Since it is doing a spearman correlation itself, does it have anything to do with the correlation coefficient?
3. what exactly does fdr mean in the web report? Why is there only one fdr for so many bundles? I tried to find the answer in the relevant literature, but I'm sorry I didn't understand much about the methodology related.
4. after getting significantly different *.tt.gz, i used recognize and cluster, how do i get the fdr value for each cluster? I see that it is not reported in the general literature, only in this article I see the fdr value for each cluster: https://www.sciencedirect.com/science/article/pii/S092549272200052X. Because of our boss's request, I would like to ask how should I get the fdr value for each cluster's fdr value? As well, since it's a spearman correlation analysis, will I be able to get the correlation coefficients for each cluster?
5. finally, why use spearman for correlation? Is it related to correlational tractography itself or is it related to my own behavioral variables? I don't see where I can change it, for example I changed it to pearson related

Best,

Wenjun

[Frank Yeh]

I had the same problem, I specified T-threshold=2.5.  I loaded the t_statistics.fib.gz file and opened the inc.tt.gz tract file then selected "Tract Rendering" -> "Color" -> "Local Index" and selected "inc_t" as the Index.  I got the following chart:

and if I set min value to 2.5, the tracts become dark blue.

This is a known issue, and I am yet to find the bug (hope to identify and fix soon)

 

Other than that, I have a couple of questions about correlational tractography that I'd like to ask
1. When doing correlational tractography between QA and behavioral variables，the value of fdr is reported in the web report only if fdr is not set in the parameters. If I set fdr=0.05 and the web report only has whether fdr is greater or less than 0.05, how do I get the value of fdr in this case?

You may check out the length-fdr table. The longer tract usually has a lower FDR.

Please also check out the connectometry paper at https://pubmed.ncbi.nlm.nih.gov/26499808/

 

2. what exactly does t-threshold mean? I find in the official website that Higher values will map tracks with a more substantial correlation effect, whereas lower values for a weak correlation. But I don't fully understand it yet. Is it considered statistic? Since it is doing a spearman correlation itself, does it have anything to do with the correlation coefficient? 

https://pubmed.ncbi.nlm.nih.gov/26499808/ and you may check out the practicum tutorials at practicum.labsolver.org

 

3. what exactly does fdr mean in the web report? Why is there only one fdr for so many bundles? I tried to find the answer in the relevant literature, but I'm sorry I didn't understand much about the methodology related.

 

https://pubmed.ncbi.nlm.nih.gov/26499808/ and also the video.

4. after getting significantly different *.tt.gz, i used recognize and cluster, how do i get the fdr value for each cluster?

The length-fdr table may provide it.

 

As well, since it's a spearman correlation analysis, will I be able to get the correlation coefficients for each cluster?

There are two ways:

1. The output fib file has inc_t and dec_t and you may get the values of the tracts and convert them back to correction.

2. check out online documentation about how to get the subject's qa or dti_fa values after the analysis, and compute the correlation.

 

5. finally, why use spearman for correlation? Is it related to correlational tractography itself or is it related to my own behavioral variables? I don't see where I can change it, for example I changed it to pearson related

The relation is always nonlinear.

Best regards,

Frank

 

[DSI Studio]

Hi Mark,

I have identified the bug. A quick solution is to use the May version at https://github.com/frankyeh/DSI-Studio/releases/tag/2023.12.06

Future releases will also fix this bug, and my apology for all the hassle.

Best regards,

Frank

[Mark Vernon]

Hello again Frank,

Thank you for all of your hard work with this helpful tool.  For reference, my original run of the connectometry analysis was done with the 2024.02.05 release for Ubuntu_2204.  I've re-run the connectometry analysis using the 2024.05.22 release for Ubuntu_2204.  I created a Region using the Regions Misc -> New Region from Thresholding from the inc_t data, then used the Regions Misc -> Modify Region -> Defragment tool on the newly created inc_t region (a check suggested here:  https://groups.google.com/g/dsi-studio/c/6TQlcRiu4j0/m/pgLaJ3wHAQAJ).  While the region appears to agree with one spot on the visualization of above threshold (t_stat > 2.5), there are still many other segments of the resultant positively correlated tracts which do not appear to have t_stat values above the set threshold (see screenshot below).

Whether the issue is a visualization problem or not, I just want to make sure I understand where the resultant positively correlated tracts are coming from.  As I understand it from the practicum (https://practicum.labsolver.org/ct.html) and the method paper (https://practicum.labsolver.org/Materials/paper/connectometry.pdf):

1. The t-statistics are determined from a Spearman rank-based correlation of QA with the study variable.  This correlation yields local fiber directions (the red sticks in the practicum image below) which correlate QA with the study variable with a t-stat greater than or equal to the t-stat threshold.  

2. Tractography is then run on these local fiber directions (which should all have a t-stat > threshold?).  

3. Tracts with a length less than the Length threshold are then removed.  All that is left should be tracts generated from tractography of local fiber directions with QA which correlates to the study variable at a t-stat greater than or equal to the t-stat threshold and with tract lengths greater than the Length threshold.

4. Randomly permute the local connectome matrix row vectors and repeat steps 1-3.  If a tract derived from the original connectome matrix has a tract Length shorter than its corresponding tract derived from the randomly-permuted matrix, then it is labeled as a false positive and is removed.

I want to make sure that the resulting tracts from the correlational tractography analysis are from a correlation with the study variable at the t-stat threshold I'm stating that they are.  When I create a region from thresholding the inc_t data (without defragmentation) at 2.5 (my t-stat threshold for the correlational tractography) I still see that most of the resultant tracts do not appear to pass through regions of inc_t > 2.5.  (see image below)

I'm not sure if there is an error with how the inc_t is calculated (if there is some kind of subtraction or averaging with dec_t), or if the issue is with how the tractography is done on the local fiber directions in Step 2 above (maybe inc_t and dec_t fibers are being incorporated into the tractography algorithm?), or if I am misunderstanding how these results are calculated.  Should I filter the connectometry results via ROI from the region created from thresholding the inc_t data?

If I can get ahold of a Windows machine, I can try to reproduce the results in a Windows environment in case the issues are due to some weird unix calculation error.

Thanks again for your help with this.  I think dsi-studio is an amazing tool.  I just want to make sure I understand what I'm doing.

~Mark

[Frank Yeh]

This is expected because the tracts were generated from multiple resampling cases on the original data. Thus the inc_t after resampling can be higher than the original one.

This is how connectometry created a distribution of the inc_t.

Frank

To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/6ccf3a87-7690-4562-8c98-122181fca4f0n%40googlegroups.com.