DIFFPREP oddities

Submitted by sidhuj2 on


I have a series of questions concerning the output of the DIFFPREP GUI. My group is using Tortoise version 2.5.0 for DTI analysis. Currently, I am trying to discern the baseline for rerunning tortoise on different terminals but on the same data set. I am not using a structural image for fitting; I have read online that this means the software utilizes a B0 image from the diffusion sequence as the structural image. If this is true, then the structural image and the template should be the same (apart from scaling?). However, after running tortoise, I inspect these two (structural.nii and template.nii) and found that there is a slight shift between the two images. What's more interesting is that the output files differ slightly between two different terminals (running the diff command comparing the structural and template files respectively shows that the files differ).

I have checked some things between the two terminals and found the following:

                                         fmri1                                                                                                                  comp-01

CentOS-release                6.6                                                                                                                    6.10

output of uname -a      Linux fmri1 2.6.32-504.30.3.el6.x86_64 #1 SMP Wed Jul 15 10:13:09 UTC       Linux is-comp01 2.6.32-754.12.1.el6.x86_64 #1 SMP Tue Apr 9 14:52:26

                                     2015 x86_64 x86_64 x86_64 GNU/Linux                                                            UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

processor info              Intel(R) Xeon(R) Silver 4114 CPU @ 2.20GHz                                                    Intel(R) Xeon(R) CPU           E5530  @ 2.40GHz


I was hoping that you could help me understand why I am getting different outputs on different terminals. Any help would be appreciated.

Submitted by irfanogl on Mon, 08/23/2021 - 12:26



First of all, I would like to state that TORTOISE 2.5 generation is no longer supported. The V3.x.x generation is a superset and superior to 2.5 generation in every way, including speed and ease of use.

Regarding your questions:

In 2.5 generation, the template image, in terms of the matrix size and the resolution is an that is different than both the b=0 image and the structural image, even when the structural image is provided. This is to overcome some limitations in the mutual information metric local minima. So it is expected that the template image that is saved will be the same image as a b=0 image, however, again with different resolution and matrix size.  Just to clarify, the template image in this context is a temporary image and has no effect on the output. It is only used to improve the optimization in the registration and serves no other purpose.  For instance, because the optimizer in the 3.x.x generation is already more robust, the concept of a template image is omitted and it does not exist anymore.

However, running it on different terminals should generate the same results (minus noise).  Again, to improve the registration accuracy, the DIFFPREP V2.5 generation intentionally adds random noise to the template image.  This noise is random and will differ from run to run. However, the two images should be slightly different even from the same terminal. IF they are identical in the two runs from the same terminal, the issue is something else and needs to be further investigated.

Could you try to run it again from the same terminal twice and let us know if they are identical or different? Also how are the two images from two terminals different? Can you do a subtraction and see if it just looks like a noise image?