 |
 |
Hi, and thanks a lot for the info. I am doing pretty much the exact same analysis, and I have been trying to figure out my deign matrices for some time now. Would you mind clarifying one point please; >>> ..EV Cd is the control difference (Time2-Time1) and Ed is the experimental difference Is Cd/Ed coded as "-1" for time2, and "1" for time1 (or the opposite)? EB Cd Ed S1 S2 S3 S4 1 -1 0 1 0 0 0 2 -1 0 0 1 0 0 3 0 -1 0 0 1 0 4 0 -1 0 0 0 1 1 1 0 1 0 0 0 2 1 0 0 1 0 0 3 0 1 0 0 1 0 4 0 1 0 0 0 1 Cheers Inge On Thu, May 27, 2010 11:36, Thomas Nichols wrote: > Dear Miguel, > > Sorry, I forgot about the contrasts... > > For these' EV's > Cd Ed S1 S2 S3 S4 > C1: -1 1 0 0 0 0 > C2: 1 -1 0 0 0 0 > C3: 1 0 0 0 0 0 > C4: 0 1 0 0 0 0 > > > You have the contrasts you specified... > > C1 pre-post increase in controls is higher than in experimental subjects > C2 pre-post increase in experimental subjects is higher than in controls > C3 pre-post increment for the control group, and C4 for the experimental > group > > > But note that this F-contrast > 1 0 0 0 > and this F-conrast > 1 1 0 0 > (to see overall group differences in pre-post effects) are totally > equivalent (and so 1st one is better). > > > And the exchangeability block file is plain text, no header (sorry, > inconsistent, I know). > > -Tom > > > On Thu, May 27, 2010 at 9:29 AM, Miguel Burgaleta < > [log in to unmask]> wrote: > >> >> OK Tom, thanks a lot! I assume that the contrast that I set for my wrong >> DM >> can be applied to yours, right? And what info should I include in the >> "header" of the EB file (before /matrix)? >> >> Miguel >> >> >> >>> EB Cd Ed S1 S2 S3 S4 >>> >>> 1 -1 0 1 0 0 0 >>> 2 -1 0 0 1 0 0 >>> 3 0 -1 0 0 1 0 >>> 4 0 -1 0 0 0 1 >>> 1 1 0 1 0 0 0 >>> 2 1 0 0 1 0 0 >>> 3 0 1 0 0 1 0 >>> 4 0 1 0 0 0 1 >>> >>> Where EB is the exchangeability block file, EV Cd is the control >>> difference (Time2-Time1) and Ed is the experimental difference, and >>> S1-4 are >>> the subject dummy (blocking) variables. >>> >>> (And, *yes*, you need the exchangeability block file... you always need >>> it >>> whenever you have repeated measures in a randomise analysis). >>> >>> -Tom >>> >>> My contrast would be: >>>> >>>> -1 1 0 0 0 0 Interaction group x time (if pre-post changes in >>>> group 2 >>>> (experimental) are higher than pre-post changes in group 1 (control)) >>>> >>>> >>>> If this is correct, I recall my last question in my previous post: 3) >>>> Do >>>> I have to use an exchangeability-block file along with the -e flag in >>>> randomise? If yes, should I add a header like in .con file? The >>>> design.grp >>>> would look like this?: >>>> >>>> >>>> 1 >>>> 1 >>>> 1 >>>> 1 >>>> 2 >>>> 2 >>>> 2 >>>> 2 >>>> >>>> >>>> Thanks, >>>> Miguel >>>> >>>> >>>> >>>> >>>> >>>> >>>> On Wed, May 26, 2010 at 9:28 AM, Stephen Smith >>>> <[log in to unmask]>wrote: >>>> >>>>> HI - no, start from the paired t-test design, and then take the >>>>> pre-post >>>>> EV and split this into one EV for controls and one for non-controls, >>>>> padding >>>>> with zeros. >>>>> Cheers. >>>>> >>>>> >>>>> >>>>> On 25 May 2010, at 14:25, Miguel Burgaleta wrote: >>>>> >>>>> Thanks Stephen. I found out the problem after looking into the >>>>> scripts. I was indeed applying the warps (as tbss_non_FA does), but I >>>>> was >>>>> assuming that they didn't include the affine transform... So I was >>>>> passing >>>>> the _to_target.mat twice (in the concatenated matrix and in the >>>>> original >>>>> warp). >>>>> >>>>> My next concern is about setting randomise design and contrast >>>>> matrices >>>>> properly. I posted an attempt to model this but didn't get an answer. >>>>> I have >>>>> found, however, a post where a similar case is presented so I should >>>>> be able >>>>> to adapt it. Could you please say if the following is correct? Let's >>>>> suppose >>>>> that I have 2 controls + 2 experimental subjects, and 2 timepoints. >>>>> My DM >>>>> would be: >>>>> >>>>> con-time1 con-time2 exp-time1 exp-time2 con1 con2 exp1 exp2 >>>>> 1 0 0 0 1 0 0 0 >>>>> 1 0 0 0 0 1 0 0 >>>>> 0 1 0 0 1 0 0 0 >>>>> 0 1 0 0 0 1 0 0 >>>>> 0 0 1 0 0 0 1 0 >>>>> 0 0 1 0 0 0 0 1 >>>>> 0 0 0 1 0 0 1 0 >>>>> 0 0 0 1 0 0 0 1 >>>>> >>>>> I would include these contrasts (not 100% sure about their >>>>> interpretation though --please see below): >>>>> >>>>> C1: -1 1 1 -1 0 0 0 0 >>>>> C2: 1 -1 -1 1 0 0 0 0 >>>>> C3: -1 1 0 0 0 0 0 0 >>>>> C4: 0 0 -1 1 0 0 0 0 >>>>> >>>>> C1 will say where the pre-post increase in FA in controls is higher >>>>> than >>>>> in experimental subjects >>>>> C2 will say where the pre-post increase in FA in experimental >>>>> subjects >>>>> is higher than in controls >>>>> C3 will show the significant pre-post increment for the control >>>>> group, >>>>> and C4 for the experimental group >>>>> >>>>> I would also ad an F-test to see the overall change across groups: >>>>> >>>>> 1 1 0 0 >>>>> >>>>> My specific questions are: 1) How do my design + contrast look? 2) Is >>>>> my >>>>> interpretation correct? 3) Do I have to use an exchangeability-block >>>>> file >>>>> along with the -e flag in randomise? If yes, should I add a header >>>>> like in >>>>> .con file? The design.grp would look like this: >>>>> >>>>> 1 >>>>> 1 >>>>> 1 >>>>> 1 >>>>> 2 >>>>> 2 >>>>> 2 >>>>> 2 >>>>> >>>>> Thanks a million for your time >>>>> Miguel >>>>> >>>>> >>>>> >>>>> >>>>> >>>>>> Hi - it looks like maybe you're not using the nonlinear (warp) >>>>>> transformations derived from the PRE images? Have a look at the >>>>>> scripts to >>>>>> see how these come into this - I guess you'll need to copy those >>>>>> across from >>>>>> PRE to POST to combine with the concatenated affine transform >>>>>> >>>>> >>>>>> Cheers. >>>>>> >>>>>> >>>>>> On 24 May 2010, at 16:53, Miguel Burgaleta wrote: >>>>>> >>>>>> Hello FSLers, >>>>>> >>>>>> I am using TBSS to process a dataset of subjects scanned at 2 >>>>>> timepoints (PRE and POST). At this point I am trying to do the >>>>>> intra-subject >>>>>> registration, but the result doesn't look good. This is what I have >>>>>> done so >>>>>> far: >>>>>> >>>>>> 1. Apply the full TBSS pipeline to PRE subjects, generating FA, >>>>>> origdata and stats directories with their respective files. Visual >>>>>> inspection of all_FA.nii.gz shows very nice non-linear registration >>>>>> to the >>>>>> default template. >>>>>> >>>>>> 2. Apply tbss_1_preproc to POST subjects, and FLIRT the output to >>>>>> the >>>>>> PRE FA images (in native space): >>>>>> >>>>>> *flirt -dof 6 -ref PRE_FA.nii.gz -in POST_FA.nii.gz -out >>>>>> POST_FA_to_PRE.nii.gz -omat POST_FA_to_PRE.mat* >>>>>> >>>>>> POST_FA_to_PRE.nii.gz looks nicely aligned to PRE images. >>>>>> >>>>>> 3. Concatenate this transformation matrix with that resulting from >>>>>> applying tbss_2_reg to PRE subjects (_FA_to_target.mat), and >>>>>> overwrite the >>>>>> original matrix (after backup ;): >>>>>> >>>>>> *convert_xfm -omat PRE_FA_to_target.mat -concat PRE_FA_to_target.mat >>>>>> POST_FA_to_PRE.mat* >>>>>> >>>>>> 4. Prepare my POST images to run tbss_non_FA on them. For this, I >>>>>> first >>>>>> create a POST directory where I processed my PRE data with TBSS >>>>>> (where >>>>>> origdata, FA and stats folders are) and copy the POST_FA.nii.gz >>>>>> files (after >>>>>> tbss_1_preproc) to that location. >>>>>> >>>>>> When I run tbss_non_FA POST, the result is not OK (POST images >>>>>> occupy >>>>>> the same exact area than PRE images, but brain structures don't >>>>>> match...). >>>>>> However, if I instead feed tbss_non_FA directly with the output from >>>>>> 2 (POST >>>>>> FA images flirted to PRE images), then the result is gorgeous. The >>>>>> problem >>>>>> is, POST images would 'suffer' 2 interpolations and PRE's only one >>>>>> (actually, POST images look a bit blurry with the latter approach). >>>>>> >>>>>> Any idea of what I am missing? >>>>>> >>>>>> Thanks a lot in advance! >>>>>> Miguel >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> --------------------------------------------------------------------------- >>>>>> 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 >>>>>> >>>>>> --------------------------------------------------------------------------- >>>>>> >>>>>> >>>>>> >>>>>> >>>>> >>>>> >>>>> >>>>> --------------------------------------------------------------------------- >>>>> 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 >>>>> >>>>> --------------------------------------------------------------------------- >>>>> >>>>> >>>>> >>>>> >>>> >>> >>> >>> -- >>> ____________________________________________ >>> Thomas Nichols, PhD >>> Principal Research Fellow, Head of Neuroimaging Statistics >>> Department of Statistics & Warwick Manufacturing Group >>> University of Warwick >>> Coventry CV4 7AL >>> United Kingdom >>> >>> Email: [log in to unmask] >>> Phone, Stats: +44 24761 51086, WMG: +44 24761 50752 >>> Fax: +44 24 7652 4532 >>> >>> >> > > > -- > ____________________________________________ > Thomas Nichols, PhD > Principal Research Fellow, Head of Neuroimaging Statistics > Department of Statistics & Warwick Manufacturing Group > University of Warwick > Coventry CV4 7AL > United Kingdom > > Email: [log in to unmask] > Phone, Stats: +44 24761 51086, WMG: +44 24761 50752 > Fax: +44 24 7652 4532 >