Optimal Tortoise 3.1.4 setting of <correction_mode>?

Submitted by ceierud on Wed, 04/10/2019 - 12:11

Dear Tortoise Team,

Thanks for sharing your great software!
Looking to upgrade our IDL pipeline to Tortoise 3.1.4 we wonder what the optimal setting of the <correction mode> parameter is in DIFFPREP's reg_settings file? We do *NOT* have any sagittal images that needs to be registered to axial. After tests comparing the different settings we get identical results for both cubic and multistart, but different for quadratic, in the resulting files ending with *_DMC.nii (comparing DTIs with 45 directions). Comments from a default reg_settings file for the correction mode parameter states the following:

<!-- Correction mode. Specifies which undesirable effects will be corrected-->
<!-- off: no DWI registration-->
<!-- Predefined motion and eddy distortion correction optimization settings. Each setting points to a file in the softwares settings folder-->
<!-- rigid:  . Corrects only motion..-->
<!-- quadratic:  . Eddy currents are modeled with quadratic functions.. Quadratic model is sufficient 99% of the time-->
<!-- cubic:   Eddy currents are modeled with cubic functions..-->
<!-- multistart:   Eddy currents are modeled with quadratic functions.. The rigid search space is  VERY LARGE. This can be used to register sagittal images to axial for example..-->
<!-- You can also create a custom settings file in ~/DIFF_PREP_WORK and put its name here..-->

Submitted by nayaka on Wed, 04/10/2019 - 17:11



Can you please describe if the results you are seeing with quadratic is better than when the data is run with cubic? The default setting is quadratic as it usually works efficiently most of the time to give better results. To state optimal setting, that would depend on your data. On Certain occasion cubic may work better and usually the processing can be run with this setting, if default (quadratic) setting does not perform the correction well.



Submitted by irfanogl on Thu, 04/11/2019 - 14:02


Just to pitch in,

Quadratic would work with 99.9% of the data. There were only very few extreme datasets that I had to use cubic on.

Additionally, these settings are not hardcoded into the executables. They are actually settings files that are located in DIFFPREP/Assets/mecc_settings  folder.

If you want to see what the differences are between these settings, you can open them in a text editor and examine them as well.


Submitted by ceierud on Tue, 04/16/2019 - 12:33


We found only very subtle differences between the quadratic and cubic methods and we could not distinguish which method was superior for our data (after screening). We have decided to evaluate the quadratic processing since it is the default option of Tortoise 3.1.4 and is faster to process compared to the cubic method.​ Thank you very much for both answers above, which were helpful for us to continue.