Header
  1. Home
  2. Robust tensor fitting

Robust tensor fitting

Submitted by vanania on
Category
DIFFCALC

Dear Tortoise community,

I am a new Tortoise user and downloaded version 3.2.0 for Linux.

In this moment, I would need to estimate the diffusion tensor using the robust RESTORE approach. I have a few questions about how to correctly run the command EstimateTensorNLLSRESTORE:

  • Does the RESTORE implementation available in TORTOISE correspond to what is described in the paper "Chang, Lin‐Ching, Lindsay Walker, and Carlo Pierpaoli. "Informed RESTORE: a method for robust estimation of diffusion tensor from low redundancy datasets in the presence of physiological noise artifacts." Magnetic resonance in medicine 68.5 (2012): 1654-1663”?
  • At the moment, I already have a pre-processed DWI subjID_prep.nii as well as bvals, bvecs and a pre-computed mask brain_mask.nii. Is it correct to apply the RESTORE algorithm in the following way:

            ImportNIFTI -i  ../subjID_prep.nii -b subjID.bval -v subjID.bvec -p vertical

            EstimateTensorNLLSRESTORE  -i ../subjID_proc/subjID.list -m ../brain_mask.nii

If this is the correct the way to go, 

  • What is the signal standard deviation used for performing RESTORE? Is it the same approach as in the paper mentioned above?

  • What do the following outputs exactly represent?

  1. ../subjID_R1_DT.nii: to my understanding, this is the Diffusion Tensor (Dxx, Dyy, Dzz, …). But what is the unit of measurement in which the data is stored?
  2. ../subjID_R1_OUT.nii
  3. ../subjID_R1_VOUT.nii —> it looks like the outlier map but why isn’t it binary (0-1)?
  4. ../subjID_R1_AM.nii

Thanks a lot for your precious help, looking forward to your reply!

Vincenzo Anania

Submitted by vanania on Tue, 03/10/2020 - 08:29

Permalink

Dear Tortoise team,

Do you have any updates about my question on EstimateTensorNLLSRESTORE?

Thanks again for your attention,

Vincenzo 

Submitted by irfanogl on Thu, 03/12/2020 - 15:50

Permalink

First of all apologies for the delay. The web system currently has issues and does not let us know about the posted questions. we are working on a fix.

 

Coming back to your question...

1) No. The RESTORE implementation actually corresponds to:

https://www.ncbi.nlm.nih.gov/pubmed/15844157

not the iRESTORE paper, which has different assumptions.

 

2) 

            ImportNIFTI -i  ../subjID_prep.nii -b subjID.bval -v subjID.bvec -p vertical

            EstimateTensorNLLSRESTORE  -i ../subjID_proc/subjID.list -m ../brain_mask.nii

If this is the correct the way to go?

 

Yes it is.  

 

3) What is the signal standard deviation used for performing RESTORE? Is it the same approach as in the paper mentioned above?

Yes it is the approach described in the RESTORe paper.

4) 

  1. ../subjID_R1_DT.nii: to my understanding, this is the Diffusion Tensor (Dxx, Dyy, Dzz, …). But what is the unit of measurement in which the data is stored?
  2. ../subjID_R1_OUT.nii
  3. ../subjID_R1_VOUT.nii —> it looks like the outlier map but why isn’t it binary (0-1)?
  4. ../subjID_R1_AM.nii

 

Yes the DT image is the tensor image with order (dxx,Dyy,Dzz,Dxy,Dxz,Dyz).  The unit is 10-6 mm^2/s   Or in other words micro meter ^2 /s.

OUT.nii is the cumulative outlier map.

VOUT .  yes the outlier map. it should be binary.   Even if it is saved as a float image, it should only contain zeros and ones.  The save format will be changed soon.

AM. The estimated b=0 image.

 

Hope this helps.