Different results with different versions of DSI Studio

392 views
Skip to first unread message

Jonas W.

unread,
Mar 17, 2022, 4:49:17 AM3/17/22
to DSI Studio
Hey Frank,
sorry for bothering again, but I face another problem and I don't know what's the reason. I updated DSI Studio a few days ago. Before that I was using a version max. a few weeks old.  
Before the update, in my connectometry analysis I found only negatively correlated tracts.
With the updated version I reran the QSDR reconstruction, created a new database and ran the correlational Tractography analysis again. Now the tract recognition works fine from the beginning, but most importantly, I also get positively correlated results and different tracts also for the negatively correlated. I uploaded the db files to Dropbox. Could you please have a look on them and help me what's the difference between those two? I hope that my new analysis is correct now.
Thanks for your good help and have a great day

Frank Yeh

unread,
Mar 17, 2022, 11:51:20 AM3/17/22
to dsi-s...@googlegroups.com
I am very sorry for the inconsistency. The normalization routine in DSI Studio is under revision.
I would use the V1_connectometry.qa.db.fib.gz, because it has a smoother signal appearance.
Is it from more recent version or the earlier? 
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 on the web visit https://groups.google.com/d/msgid/dsi-studio/33c09125-9fe9-4537-aadc-69f9230e5a00n%40googlegroups.com.

Jonas W.

unread,
Mar 17, 2022, 5:16:40 PM3/17/22
to DSI Studio
Hey Frank,
it is from the older one. The reconstructed fib files of the subjects were comparable. Also my results look more consistent with my hypothesis using the second approach and not the one you suggested. Where could that difference come from?
thanks for helping!
Best 
Jonas 

Frank Yeh

unread,
Mar 17, 2022, 7:45:20 PM3/17/22
to dsi-s...@googlegroups.com
The spatial normalization parameters are different.
There is a recent update to DSI Studio to fix possible misalignment errors.
If possible, could you use the latest version to test again?
I will help see if the quality is better.
Frank

On Thu, Mar 17, 2022 at 5:16 PM Jonas W. <jomi...@gmail.com> wrote:
Hey Frank,
it is from the older one. The reconstructed fib files of the subjects were comparable. Also my results look more consistent with my hypothesis using the second approach and not the one you suggested. Where could that difference come from?
thanks for helping!
Best 
Jonas 

Frank Yeh schrieb am Donnerstag, 17. März 2022 um 16:51:20 UTC+1:
I am very sorry for the inconsistency. The normalization routine in DSI Studio is under revision.
I would use the V1_connectometry.qa.db.fib.gz, because it has a smoother signal appearance.
Is it from more recent version or the earlier? 
Frank 


On Thu, Mar 17, 2022 at 4:49 AM Jonas W. <jomi...@gmail.com> wrote:
Hey Frank,
sorry for bothering again, but I face another problem and I don't know what's the reason. I updated DSI Studio a few days ago. Before that I was using a version max. a few weeks old.  
Before the update, in my connectometry analysis I found only negatively correlated tracts.
With the updated version I reran the QSDR reconstruction, created a new database and ran the correlational Tractography analysis again. Now the tract recognition works fine from the beginning, but most importantly, I also get positively correlated results and different tracts also for the negatively correlated. I uploaded the db files to Dropbox. Could you please have a look on them and help me what's the difference between those two? I hope that my new analysis is correct now.
Thanks for your good help and have a great day

--
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/33c09125-9fe9-4537-aadc-69f9230e5a00n%40googlegroups.com.

--
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.

Jonas W.

unread,
Mar 18, 2022, 8:01:35 AM3/18/22
to DSI Studio
The QSDR is still running but I already see that the R2 values are around 5 lower than before.
Best, Jonas 

Frank Yeh

unread,
Mar 18, 2022, 8:57:42 AM3/18/22
to dsi-s...@googlegroups.com
How about the R values of the previous two versions?
Frank

Jonas W.

unread,
Mar 18, 2022, 9:07:15 AM3/18/22
to DSI Studio
they were much better, in both versions (around +5)

Frank Yeh

unread,
Mar 18, 2022, 9:25:42 AM3/18/22
to dsi-s...@googlegroups.com
Thank you very much for the information.

The first two versions did not impose a requirement for inverted mapping, whereas the updated version does and thus reduces the R2 value.
What is the connectometry result compared with the first two? 

If possible, please also send me the connectometry database, and I will evaluate the misalignment.

Thanks a lot for your help!

Best regards,
Frank

Frank

Jonas W.

unread,
Mar 19, 2022, 5:30:49 PM3/19/22
to DSI Studio
Hi Frank,
the connectometry negative tt results are reeelatively similar to the second approach (the one that you said had a less smooth signal appearance (reconstructed in March 14 build), except that there are no positively correlated tracts anymore. The oldest results are however pretty different from the second, and now the third approach. How can I fix it? Could you also provide some details on fixing the inverted mapping issue? this is my code for the reconstruction: 
/Applications/dsi_studio.app/Contents/MacOS/dsi_studio --action=rec --source=PA_eddy_unwarped_${file}.src.gz --mask=PA_brain_mask_${file}.nii.gz --cmd="[Step T2a][Remove Background]" --method=7  --param0=1.25 --record_odf=1

Best,
Jonas 

Frank Yeh

unread,
Mar 20, 2022, 5:02:50 PM3/20/22
to dsi-s...@googlegroups.com
Hi Jonas,

Sorry for causing a lot of hassle. The 1.7mm seems to be a
challenge for DSI Studio. I built a version that hopefully addresses
the inverted mapping issues here.

Please update DSI Studio again (dated March 19 or 20) at
dsi-studio.labsolver.org and using the following command to
reconstruct data to the 2-mm space:

/Applications/dsi_studio.app/Contents/MacOS/dsi_studio --action=rec
--source=PA_eddy_unwarped_${file}.src.gz
--mask=PA_brain_mask_${file}.nii.gz --cmd="[Step
T2][Edit][Resample]=2" --method=7 --param0=1.25 --record_odf=1

I can help check the constructed connectometry database to see if
the alignment are better.

Best regards,
Frank
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/c7c82fbd-e541-41e9-999e-3ba7cf6e4e1dn%40googlegroups.com.

Jonas W.

unread,
Mar 21, 2022, 4:46:41 AM3/21/22
to DSI Studio
Hey Frank,
the R2 values are marginally better but still reduced compared with the results from the March 14 and earlier version (R2 around 4 lower than e.g. in the March 14 version.) 
Best, Jonas

Frank Yeh

unread,
Mar 21, 2022, 9:10:32 AM3/21/22
to dsi-s...@googlegroups.com
R2 does not quantify smoothness and may not always indicate good
reproducibility.
The database can be visually inspected at Step C2a to confirm the quality.
Frank
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/2bce5c99-4f61-4610-a46f-888fd042f172n%40googlegroups.com.

Jonas W.

unread,
Mar 21, 2022, 10:09:07 AM3/21/22
to DSI Studio
Hi Frank,
many many thanks for helping me out. Thanks for the advice, now I can also inspect them on my own! I also see the case that poorly reconstructed fibs can still produce a smooth database. Still I am worried because of the low R2 values ( from 0.43 to 0.63) that showed up just because of updating DSI Studio. As you said in the documentation they definitely have to be adjusted. What can I still do?
Best, 
Jonas 

Frank Yeh

unread,
Mar 21, 2022, 10:22:20 AM3/21/22
to dsi-s...@googlegroups.com
The only effective approach here is to smooth the raw data.

The acquisition of 1.7 mm in your data has lower SNR and thus much
noisier output in the anisotropy map.

I am going to open a new interface with --aciton=src to allow you to
smooth the data (something like --smoothing=2).

Hope this helps address the lower SNR problem in the connectometry analysis.

I will touch base again once the interface is ready.

Best,
Frank
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/8ca0aab8-a7a5-4aa7-a0fa-1d540563f0c3n%40googlegroups.com.

Frank Yeh

unread,
Mar 21, 2022, 11:41:50 AM3/21/22
to dsi-s...@googlegroups.com
Hi Jonas,

The update is under build at
https://github.com/frankyeh/DSI-Studio/actions/runs/2017094376

Once the Mac build is ready, download the new version and run the
following command for reconstruction:

/Applications/dsi_studio.app/Contents/MacOS/dsi_studio --action=rec
--source=PA_eddy_unwarped_${file}.src.gz
--mask=PA_brain_mask_${file}.nii.gz --cmd="[Step T2][Edit][Smooth
Signals]+[Step T2][Edit][Resample]=2" --method=7 --param0=1.25
--record_odf=1

Thanks again for your patience and look forward to the results.

Best,
Frank

Jonas W.

unread,
Mar 23, 2022, 9:53:23 AM3/23/22
to DSI Studio
Hi Frank,
thank you so much for updating the package in this short time! Indeed, the R2 values are excellent now and also the connectometry db looks much smoother than any of the previous versions. The results are still kind of confusing. As before, I don't get any positive correlations with QA, and the negative results have over 20,000 tracks (in many major tracts, that is not plausible to me as I am investigating a major depression sample and my VOI is a treatment outcome) with a FDR < 0.05, T-threshold 2.5 and 20 voxels length. I am now rising the T-threshold and maybe the voxel threshold to get more interpretable results. The question to you would be if this is a correct way to do, or if we could have missed something in between, that could affect the results. 
Another question is very simple: is there a way of telling the CLI to use a parametric correlation? also my flag --selection=deltaMADRS>-1 caused problems (Warning: --selection is not used. Please check command line syntax.) 
Best, 
Jonas

Frank Yeh

unread,
Mar 23, 2022, 10:06:17 AM3/23/22
to dsi-s...@googlegroups.com
Thanks a lot for testing all these versions. You may also increase the
permutation count until the results become more consistent.
It is likely that the subject have substantial variation, and you
would find results different by just adding 100 more permutation
count. If this is the case, keep adding permutation count until the
results converges.

If you use FDR threshold, the voxel length only serves to ignore
results shorter than the length. You don't have to test different
voxel length.

> Another question is very simple: is there a way of telling the CLI to use a parametric correlation? also my flag --selection=deltaMADRS>-1 caused problems (Warning: --selection is not used. Please check command line syntax.)

try --select==deltaMADRS>-1 and see if it works.

> Best,
> Jonas
>
> Frank Yeh schrieb am Montag, 21. März 2022 um 16:41:50 UTC+1:
>>
>> Hi Jonas,
>>
>> The update is under build at
>> https://github.com/frankyeh/DSI-Studio/actions/runs/2017094376
>>
>> Once the Mac build is ready, download the new version and run the
>> following command for reconstruction:
>>
>> /Applications/dsi_studio.app/Contents/MacOS/dsi_studio --action=rec
>> --source=PA_eddy_unwarped_${file}.src.gz
>> --mask=PA_brain_mask_${file}.nii.gz --cmd="[Step T2][Edit][Smooth
>> Signals]+[Step T2][Edit][Resample]=2" --method=7 --param0=1.25
>> --record_odf=1
>>
>> Thanks again for your patience and look forward to the results.
>>
>> 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 on the web visit https://groups.google.com/d/msgid/dsi-studio/b87190f1-0a98-472d-8e2f-7b4ac859054cn%40googlegroups.com.

Jonas W.

unread,
Mar 23, 2022, 10:28:06 AM3/23/22
to DSI Studio
I will run more than the 4000 permutations used before and see if the results become more consistent.

--select==deltaMADRS>-1 didn't work unfortunately. I see DSI Studio booting but shutting down immediately without producing any output.

Best,
Jonas

Frank Yeh

unread,
Mar 23, 2022, 10:29:40 AM3/23/22
to dsi-s...@googlegroups.com
It is definitely a bug. I will test and fix it asap.
Frank
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/ed5e978f-6409-4764-bc50-64219587f86fn%40googlegroups.com.

Frank Yeh

unread,
Mar 23, 2022, 3:57:43 PM3/23/22
to dsi-s...@googlegroups.com
Sorry my mistake it should be  --select="deltaMADRS>-1" 
See this it works

Jonas W.

unread,
Mar 24, 2022, 5:40:29 AM3/24/22
to DSI Studio
Okay this works now, many thanks! And how is the flag for using a parametric correlation? 
Thank you very much!
Jonas 

Frank Yeh

unread,
Mar 24, 2022, 10:50:40 AM3/24/22
to dsi-s...@googlegroups.com
--nonparametric=1
The default is on. You don't have to specify it unless you want to turn it off.
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/f2aacd57-feb1-4839-b010-9404576c6d18n%40googlegroups.com.

Jonas W.

unread,
Mar 26, 2022, 11:21:47 AM3/26/22
to DSI Studio
Hey Frank, 
thanks for helping me so much. I suppose I stumbled across another bug using the CLI. Whenever I run a --action=cnt command, after the analysis finished, I get the error  "cannot output file to variable.t3.nqa.fdr0,05.report.html" and no other file is saved. I also tried sudo on Mac but it had no effect. I need the CLI bc I have to run many analyses with minimal changes. 
Best, Jonas 

Frank Yeh

unread,
Mar 26, 2022, 11:32:17 AM3/26/22
to dsi-s...@googlegroups.com
You may try specify --output=/some_dir/file_prefix
I am going to modify the default of the --output to the demographic
file so that all results will have the demographic file name as the
prefix.
Hope this helps,
Frank
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/5c6842f7-d3a5-45fb-b84d-bc78331315c8n%40googlegroups.com.

Jonas W.

unread,
Apr 5, 2022, 9:35:00 AM4/5/22
to DSI Studio
Hey Frank,
how do I start a longitudinal correlational tractography analysis from the command line studying just the "intercept"? --voi="Intercept" or --voi=Intercept does not seem to work and always switches to column 0 in my demo file as a voice
Best,
Jonas 

Frank Yeh

unread,
Apr 5, 2022, 9:36:53 AM4/5/22
to dsi-s...@googlegroups.com
Hi Jonas,

I will modify DSI Studio to allow tracking on the intercept and
will soon get back to you.

Best,
Frank
> To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/ac21425f-3331-480c-95d8-7e44daa44bd3n%40googlegroups.com.

Frank Yeh

unread,
Jun 1, 2022, 9:45:22 PM6/1/22
to dsi-s...@googlegroups.com
Hi Jonas,

    DSI Studio is updated to solve this problem. You can now use --voi=Intercept to study the intercept in longitudinal data.

Best regards,
Frank

Reply all
Reply to author
Forward
0 new messages