Hi - The mean_FA and all_FA appear to be changing depending on the subject subset even with the -T flag (although the mean_FA skeletons are identical to each other--both are the FMRIB58 template skeleton). I sent Matthew what I could regarding the issue, will follow up here with his reply for anyone who digs up this thread in the future.  Inspecting the all_FA file, the individual subjects seem like more voxels are masked out the more subjects that are run, possibly a step to ensure all the same voxels are present for each subject in later statistical analysis? Just a guess on my part.  Thanks again for all the help!

On Thu, Mar 6, 2014 at 2:36 AM, Stephen Smith <[log in to unmask]> wrote:

Hi - if you are sure that your mean_FA is being derived from the standard space template and not (different subsets) of your study data - which should happen if you use the -T option for stages 2 and 3 - then you should get the same deprojected data for a given subject, regardless of what other subjects are in the analysis.   If you are sure this is not working right, then please send the data to Matthew to check out - see the FSL wiki for data upload instructions.

Cheers, Steve.

On 5 Mar 2014, at 18:43, Benjamin Trotter <[log in to unmask]> wrote:

Hello again, following up on my last email.  I think it better describes the specific issue I'm experiencing than the original email.  The issue remains unresolved on my end.  I am wondering if you have any insight regarding why I am finding the discrepancies in DTI measures for the same ROI in the same subject depending on what subject subset the particular subject was processed as a part of--despite the fact that, for either subset, we are deprojecting the same subject's data from the same template according to the same atlas to get the ROI values.  It seems like it should be independent from which other subjects were processed with the particular subject. As always, thanks for the assistance.

On Mon, Feb 24, 2014 at 9:30 AM, Benjamin Trotter <[log in to unmask]> wrote:

Hi - I was hoping to account for the different mean_FA skeletons resulting from running different subsets by using the template option in both tbss_2_reg and tbss_3_postreg.  If I use the -T flag for these steps, and then run the command "tbss_deproject <atlas> 2 -n", can I expect that the ROI defined by the atlas will have the same diffusion measure values regardless of the subset of subjects that was processed? This result is not what I am finding.  It is a good suggestion, to start with the full analysis, but the research project is on-going and more subjects are still being added to the study.  We hoped that by using the template and its derived skeleton that we would be able to include these new subjects in analyses as their data comes in without having to episodically re-run every subject, which as of now seems like it would only result in a (for the most part slightly) new dataset anyways. Thanks again for your assistance, hoping to get to the bottom of this.

On Sat, Feb 22, 2014 at 3:52 AM, Stephen Smith <[log in to unmask]> wrote:

Hi - if you run "full" TBSS analyses with different subsets of subjects, you will get a slightly different mean_FA and hence skeleton each time - which will then change individual subjects' results (hopefully only slightly).

Would that explain it?   I think it would be easy enough to avoid this by starting with a full analysis of all subjects, and then only rerunning sub-parts of the analysis for subsetes of the data, making sure to always base things on the same mean_FA and skeleton?

Cheers.

On 21 Feb 2014, at 17:18, Benjamin Trotter <[log in to unmask]> wrote:

Hi Stephen,

Thanks for your response, I think I need to clarify the inconsistency that's troubling me.   I believe the all_FA and warp files are all in order. My expectation is that, regardless of how many subjects are processed at once, the DTI measures for a given ROI for the same subject should not change. For example, if I process 150 subjects on one day, and then 200 subjects a different day, and then compare the value I get out for FA in a given tract for the same subject (after deprojection to that subject's native space) that the FA value will be consistent between processing-days/#subjects-processed. I'm finding that their is a discrepancy here.  Can you think of what may be causing this inconsistency?

Thanks again!

On Fri, Feb 21, 2014 at 2:17 AM, Stephen Smith <[log in to unmask]> wrote:

Hi - the output of the deprojection is dictated by various files, such as the ordering of subjects in all_FA (and the warp files from the first stage of the registration) - so you need to make sure these all match your expectations.  I can't think why there would be (eg) any inconsistency between the set of subjects inside all_FA and the outputs from the deprojection - 

Cheers

On 20 Feb 2014, at 19:50, Benjamin Trotter <[log in to unmask]> wrote:

I am running through the TBSS processing stream and am finding that after deprojection according to a segmentation atlas (tbss_deproject <atlas> 2 -n) to a subject's native space that the ROI values I get out for the DTI measures using asegstats2table change depending on how many subjects I am running at a time.  As a fix, I've tried using the template flag for both the tbss_2_reg and tbss_3_postreg steps, so that in theory the same skeleton voxels will be deprojected regardless of which subjects are run at once.  However, even using the template I am still getting differences between runs depending on which subjects I run.  It is probably useful to note that I get no differences between runs when I run the same set of subjects for each run. Has anyone else experienced this issue? I'm wondering if there are processing steps happening "behind the scenes" in TBSS that could result in these differences even if the template is used.  

The differences between runs are generally minor, with most across-run tract DTI measures being well correlated, however I would like to better understand the source of these discrepancies and whether there is anything I can change in my processing to avoid them.  

Thanks in advance for any and all assistance!

---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve

---------------------------------------------------------------------------

Stop the cultural destruction of Tibet

---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve

---------------------------------------------------------------------------

Stop the cultural destruction of Tibet

---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve

---------------------------------------------------------------------------

Stop the cultural destruction of Tibet

--

Ben Trotter

Health Science Specialist

VA Boston Healthcare System

Translational Research Center for Traumatic Brain Injury and Stress Disorders (TRACTS)