Tortoise 3.1.4 EstimateTensorNLLSRESTORE

Submitted by ceierud on Thu, 04/25/2019 - 18:34

Dear Tortoise Team,

Thanks again for your great product and support that we want to run the latest version of.
Noticing that Tortoise 3.1.4 has the command EstimateTensorNLLSRESTORE, which we tried to run, we believe we are stuck with mask problem. First of all we do not believe we have any masks from Tortoise 3.1.4 DIFFPREP. We tried to created a mask using .../TORTOISE_V3.1.4/DIFFCALC/DIFFCALCV25/DIFFCALC/diffcalc_main/calcvm using the [mask raw images] button (with options "Threshold..."=0 and "f bet..."=0.25) and the [save mask] button, creating a mask that does not seem to be in the same space as the other Tortoise 3.1.4 brain volumes including subjID_DMC.nii. The result (subjID_DMC_R1_DT.nii) were provided without errors in about a minute of processing, but subjID_DMC_R1_DT.nii had what looked like a mask problem and the results were unusable after running the following:
EstimateTensorNLLSRESTORE -i ~/subjID_DMC.list -m ~/subjID_DMC_OF_USER_SAVED_MASK.nii -n -v

Thinking we may need to create a mask manually we tried the following command lines (to create a new mask called mask-f.nii):
cd ~
fslsplit subjID_DMC.nii vol
bet vol0000.nii.gz mask -f 0.25 -g 0 -m #knowing first volume is our B0
fslmaths mask_mask.nii mask-f -odt float
gzip mask-f.nii.gz -d

After the "manual" mask Tortoise 3.1.4 seem to get stuck at the "Computing R1 RESTORE Tensors...", keeping processing even after 24 hours when we finally gave up this run, using the following syntax:
EstimateTensorNLLSRESTORE -i ~/subjID_DMC.list -m ~/mask-f.nii -n -v

Of course we know we have the option to reside back to Tortoise 2.5 for the DIFFCALC processing, but we are all ears if you have any suggestions of how to run Tortoise 3.1.4 EstimateTensorNLLSRESTORE with more success and which package (2.5 or 3.1.4) you suggest for DIFFCALC including RESTORE.

Best Regards,
Cyrus Eierud, PhD
Rush Alzheimer's Disease Center
1750 W Harrison Street, Suite 1021N
Chicago, IL 60612

Submitted by irfanogl on Fri, 04/26/2019 - 11:35



The EstimateTensorNLLSRESTORE executable is admittedly still in beta phase. Even though we tested it on several datasets and it worked fine, there might be cases where it doesnt work as expected.

Before I go into the executable, this is normally how I generate my mask:

ExtractImage -i  subjID_DMC.nii -v 0

bet2  subjID_DMC_V000.nii mask -m -f 0.3

Then the file mask_mask.nii.gz is the one I use.

For RESTORE, the noise pat (-n -v) is  not very much tested. So I suggest to use

EstimateTensorNLLSRESTORE -i ~/subjID_DMC.list -m ~/mask_mask.nii.gz 


Let me know if this way works. 

ABout 3.1.4 vs. 2.5 generation. THe 2.5 is very well tested. algorithmically, 3.1.4 should be almost identical to 2.5 (if it doesnt run 24 hrs :)) . However, conceptually, both of the might have caveats with some assumptions, such as noise distributions, parallel imaging related spatial noise patterns and etc. We are working on a completely new approach for this but it will take a while to make it user-ready.


Submitted by ceierud on Mon, 04/29/2019 - 12:14


Thank you for fast reply and interesting info about the ExtractImage command. That the difference EstimateTensorNLLSRESTORE results between 3.1.4 and 2.5 generation may not be that different leads us to believe there is no loss to keep running the DIFFCALC using the 2.5 generation at this point. We did try the exact EstimateTensorNLLSRESTORE commands you suggested, but still we get a very long processing time and stopped it after 24h. However, we can run the DIFFCALC part fairly fast (1h per subject) in the 2.5 generation and may do that. We are still interested in more information about EstimateTensorNLLSRESTORE that I believe at some point will be an improvement over the 2.5 generation and we are then all ears about it and you may then contact us at cyrus_eierud@rush.edu about this matter.

Best Regards,
Cyrus Eierud, PhD
Rush Alzheimer's Disease Center
1750 W Harrison Street, Suite 1021N
Chicago, IL 60612