JISCMail - FSL Archives

Dear all,

I somehow missed Anderson's reply on the 25th - sorry for not acknowledging your help and thank you very much for your email then! I think that model sounds overall quite sensible, but presumably (like I think Shane was getting at) I would be looking for the group variable to be significant in spite of the core RT/AD correlation likely also being significant? i.e. the RT/AD relationship should be considered the "nuisance" one and the patient v. control grouping as the one of primary interest?

Baixiong - It's difficult to say exactly about your data; there's a lot of different things that can happen and it's hard to judge over the internet by text alone! Your pre-processing steps seem reasonable to me as they've been typed (though I tend to use fslsplit to get the b0 from the 4D dataset, I'm less familiar with fslroi and am assuming it was implemented correctly for you). Your steps miss out the dtifit command to actually calculate all your DTI maps though, but I'm guessing this was done after the eddy_correct stage and before you averaged L2/L3 to get RD?

If you want to know for definite what the underlying FA values are in each of your groups, rather than relying on the idea of which TFCE stats map is showing significance, then you could threshold your TFCE stats maps at the 0.95 value, and then use those thresholded images as masks to extract the underlying FA mean / SD from each subject's FA map (the ones which have been transformed into the experiment's standard space). This would let you see the values which are contributing to the difference directly for each subject, and know 100% which group is higher.

What I would say (assuming that the preprocessing is correct) is that though decreases in FA is the expected change, raised FA values can also occur in cases of damage. This is a situation I've encountered a handful of times myself. Things like immune processes etc. are thought to contribute to FA increases (one paper I've leant on a bit in the past showed astrocyte invasion to sites of damage to lead to AD increases, and thus FA increases, because the way astrocytes line up along damaged nerve fibres is very unidirectional). You could think about if there's anything "unusual" about your patient population; like were the images captured during an MS attack whereas a lot of literature is perhaps studying chronic MS patients where (at the time of the scan) there may not be any immune processes going on? etc.

All the best,

Iain

On Fri, 7 Jun 2019 at 10:47, baixiong xiao <[log in to unmask]> wrote:

HI, lain,
I also finished TBSS, but the result is bad. It is opposite from the
literature. This is the steps I used. Could you find something wrong
in it?

I compared MS(multiple scelrosis) patients with normal controls with
TBSS, they are named in a logical order, as follows
CON01_FA.nii.gz
CON02_FA.nii.gz
CON03_FA.nii.gz
...
CON14_FA.nii.gz
MS01_FA.nii.gz
MS02_FA.nii.gz
MS03_FA.nii.gz
...
MS17_FA.nii.gz.

I used the scripts as follows:

Pre-process:
fslroi MS01.nii.gz b0 0 -1 0 -1 0 -1 0 1;
Bet2 b0.nii.gz nodif_brain –m –f 0.3;
eddy_correct MS01.nii.gz data 0(time-consuming);
Get L23: fslmaths L2_image -add L3_image -div 2 L23;

TBSS-process:
tbss_1_preproc *.nii.gz;
tbss_2_reg -T;
tbss_3_postreg -S;
tbss_4_prestats 0.2;
cd ../stats
design_ttest2 design 14 17;
randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d
design.mat -t design.con -n 5000 --T2;

Then I used this script -----------fslview_deprecated
$FSLDIR/data/standard/MNI152_T1_1mm mean_FA_skeleton -l Green -b
0.2,0.7 tbss_tfce_corrp_tstat1 -l Red-Yellow -b 0.95,1------------- to
display the result.

But what is beyond me is that there is no significant voxels. This is
impossible. Because MS patients have obvious lesions, they must have
some microstructual changes compared to NC.

Then I upload tbss_tfce_corrp_tstat2 -l and set the display rang
0.95:1, now I can find some significnat regions. What I cannot
understand is I think I should find significant voxels in test1,
because as the official website says: contrast 1 gives the
control>patient test and contrast 2 gives the control<patient test. So
in my case, I should have got result from ttest1, this shows the
control > MS patients in FA metric(because the FA values of MS
patients should decrease). But in fact, the result is contrary, why is
that? Did I make a mistake in certain steps? Or should I change the
display range?

Thank so much.

Baixiong

On Fri, Jun 7, 2019 at 12:51 AM Shane Schofield
<[log in to unmask]> wrote:
>
> Hi Iain
>
> Just to chime in if you don’t mind!
>
> Would the intercept be a column of 1s?
> However, this analysis won’t tell us whether the relationship is specific to the patients, right?
>
>
>
>
> Sent from Yahoo Mail for iPhone
>
> On Saturday, May 25, 2019, 10:33 pm, Anderson M. Winkler <[log in to unmask]> wrote:
>
> Hi Iain,
>
> You can do the TBSS analysis in which you investigate the association between RT and AD, while having the groups (patients and controls) as nuisance. The model would be:
>
> EV1: RT
> EV2: groups (coded as +1/-1)
> EV3: intercept.
>
> All the best,
>
> Anderson
>
>
> On Wed, 22 May 2019 at 07:02, Iain Croall <[log in to unmask]> wrote:
>
> Hello experts,
>
> I've got some DTI data that I've done a TBSS experiment with, comparing a patient population against matched controls. This shows some widespread significant differences in axial diffusivity. There's also cognitive data for all subjects, from which I know the patients are impaired in a reaction time task compared to the controls.
>
> I'd like to correlate the TBSS with the reaction time task. My question at the heart of this would be to ask if the physiological difference between groups (the TBSS AD change) is a driver of the cognitive difference between the groups (the cog. functioning reaction time difference). I'm conscious that simply including the reaction time values in a correlation with TBSS data will undoubtedly produce many significant results simply by virtue of this being linked to WM tract health in any situation. i.e. there will be normal "control" correlations there, and I want to separate that out to see if the damage has led to the loss of reaction time.
>
> A) One idea would be to restrict a TBSS correlation analysis between the AD maps and cog. scores to just the patients, and then try and match up significant locations with where I know AD has changed. But I feel this still doesn't address the issue of how much those relationships were there "anyway".
>
> B) Is there some kind of GLM model mixture that would more efficiently answer the question?
>
> Thanks in advance for any help,
> Iain
>
> ########################################################################
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
>
>
> ________________________________
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
>
>
> ________________________________
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1

########################################################################

To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1