On 18 May 2009, at 18:51, Marc Dubin wrote:
Could it be that the previous version of tbss and the present one scale FA differently? When I ran the series of tbss scripts in the previous version, there was a step that rescaled everything from 0->10000 and that is no longer done in the current version of tbss. Could that somehow be leading to the sign error that I'm getting?
Best,
Marc
On Thu, May 14, 2009 at 1:34 PM, Marc Dubin <[log in to unmask]> wrote:
Hi Steve,
You might think so, but I have also tried changing the FSL design, not de-meaning EV2 and getting rid of the constant EV1 (and making the cog non-demeaned score the only EV). This is the same as the ROQS analysis and I get almost the same exact confusing reults.
Best,
Marc
On Thu, May 14, 2009 at 10:26 AM, Steve Smith <[log in to unmask]> wrote:
Hi,
On 14 May 2009, at 17:54, Marc Dubin wrote:
Hi Steve,
Thanks so much for getting back to me...
The randomise command that I used is:
randomise -i all_FA -o tbssFA_Elision_PN -m mean_FA_mask3000 -d designElision_PN.mat -t designElision_PN.con -n 5000 -V -x -D
I have used this with and without the -1 option and with and without the -D option and get the same results.
Because the design matrix includes a mean EV you shouldn't demean the data (so don't use -D option) and the -1 option also doesn't make sense in this scenario. The rest looks fine. I take it you are comparing the second contrast [0 -1] with the 'hand' approach below.
I have also attached the design files.
By manual, I mean the ROQS automatic segmentation technique developed by Niogi and McCandliss (Neuroimage, 2007, 35:166:174). I am using this to automatically isolate the left SCR by finding its boundary and then taking the average FA inside this boundary. This single number average correlates positively with my cognitive score.
I assume you are doing equivalent modelling - i.e. either demeaning the data extracted in this way, or including a mean EV in your regression?
However, when I run the above FSL analysis, I get no positive correlation with any of the voxels in the left (or right) SCR. However, when I modify the FSL/TBSS analysis so the Elision contrast is (0 -1) instead of (0 1), I get voxels of significance throughout the entire left SCR , as expected, however these voxels are negatively correlated with the cognitive score (Elision). There seems to be a sign error somewhere.
Seems like there must be a dodgy step in this "hand" analysis....
Cheers.
Any thoughts you have on what I'm doing wrong would be most appreciated!
Best,
Marc
On Thu, May 14, 2009 at 9:18 AM, Steve Smith <[log in to unmask]> wrote:
Hi, back from Hawaii now....catching up....
No - this sounds straightforward so there shouldn't be this inversion. The randomise-based modelling looks simple and correct - maybe you should tell us exactly what syntax you used for the randomise command - and also explain in more detail what you mean by "more direct, manual, FA analysis"?
Cheers.
On 8 May 2009, at 14:58, Marc Dubin wrote:
Dear All,
I have a TBSS & linear regression question.
I have run a simple linear regression with one independent (cognitive test score) and one dependent variable (fractional anisotropy) that I set up in GLM_gui.
EV1 = constant
EV2 = cognitive test score (which is demeaned by GLM_gui)
contrast of interest (contrast 2) is (0, -1)
The problem is I get almost the same exact region of significance as I get in a different, more direct, manual FA analysis. However, with the manual approach, the correlation is positive, not negative. Does anyone have ideas about what might be giving rise to this sign error?
Thanks in advance!
Best,
Marc
On Wed, Jan 21, 2009 at 8:44 AM, Steve Smith <[log in to unmask]> wrote:
Ah, good point. I was thinking more of the options where other programs were run _before_ bet2 as opposed to _after_.
It _might_ be good enough to run bet2 with the mesh output option turned on, using as input the brain-extracted output from the first run of bet with the -S option. ??
Cheers.
On 20 Jan 2009, at 18:11, Marc Lalancette wrote:
Thanks for your prompt reply.
I suspect that for you the easiest thing
would be to amend the -S option so that it gave the surface outputs
you need, and then run the betsurf second-stage stuff separately
afterwards."
I looked at the script, but I don't know how to do that. The -S option
first runs bet2, then applies corrections to the masks with fslmaths calls.
If I add the option to output the surface meshes in the bet2 call, I would
get the same mesh as without the -S option.
Is there a tool I can use to convert a mask to a mesh directly? Or is that
code only in bet2?
Cheers,
Marc Lalancette
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
--
Marc Dubin, MD PhD
Clinical Research Fellow
Brain Imaging Laboratory
NYSPI & Columbia University
Office: 212-543-6702
Mobile: 646-831-8886
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
--
Marc Dubin, MD PhD
Clinical Research Fellow
Brain Imaging Laboratory
NYSPI & Columbia University
Office: 212-543-6702
Mobile: 646-831-8886
<Archive.zip>
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
--
Marc Dubin, MD PhD
Clinical Research Fellow
Brain Imaging Laboratory
NYSPI & Columbia University
Office: 212-543-6702
Mobile: 646-831-8886
--
Marc Dubin, MD PhD
Clinical Research Fellow
Brain Imaging Laboratory
NYSPI & Columbia University
Office: 212-543-6702
Mobile: 646-831-8886
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
