The design you showed has 40 rows...
On 30 April 2018 at 11:17, Tudor Popescu <[log in to unmask]> wrote:
> Thanks Anderson for your kind reply! Indeed, I had in the meantime figured
> that the .con problem was in the header incongruency. But am now still
> stuck with the "number of rows" error pasted in my earlier email! Using
> fslinfo on my diff_s2 image, as you suggest, confirms that indeed the
> number of volumes ("dim4") is 20 - the same as the number of rows that I
> have in (one of the) designs that I tried.
>
> Cheers,
> Tudor
>
> On 30 April 2018 at 15:45, Anderson M. Winkler <[log in to unmask]>
> wrote:
>
>> Hi Tudor,
>>
>> Please see below:
>>
>> On 23 April 2018 at 08:11, Tudor Popescu <[log in to unmask]> wrote:
>>
>>> Hi Anderson,
>>>
>>> Thanks for getting back to me. Let me reply to your points below.
>>>
>>> I think so, although for the voxelwise you might want to start with the
>>>> smoothed one.
>>>
>>> In the commands above, I had first computed the difference 4D image,
>>> then smoothed this image and taken it randomise. Do I understand well that
>>> I need to reverse this order, i.e. smooth the starting images
>>> (*_struc_GM_to_template_GM_mod), compute the difference 4D image from
>>> these, and take *that *to randomise?
>>>
>>
>> Sorry, I thought you meant you started randomise with the non-smoothed.
>> Computing the difference before or after smoothing doesn't make any
>> difference.
>>
>>
>>
>>>
>>> I'm a bit worried with the use of FNIRT above. So the masks aren't in
>>>> standard space yet?
>>>
>>> The masks were initially created in standard space, but with the FNIRT,
>>> I was trying to align them to the study-specific (VBM) template space. Is
>>> that wrong? To me the masks look fine, although I am not sure how I can
>>> check which space they are really in!..
>>>
>>
>> For FNIRT, you need to align a brain to another brain, not a mask to a
>> brain. You can apply previously computed transformations to a mask, though
>> (with applywarp).
>>
>>
>>>
>>> Also, having run randomise as
>>>
>>> *randomise -i diff_s2.nii.gz -m GM_mask -o diffs -d diffs.mat -t
>>> diffs.con -T -n 5000*
>>>
>>> I get the error
>>>
>>> *Loading Data: ERROR: Program failed. **An exception has been thrown: **diffs.con
>>> has insufficient data points*
>>>
>>> even though my .con and my .mat files (attached) both have the same
>>> number of columns. As instructed here
>>> <https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Single-Group_Paired_Difference_.28Paired_T-Test.29>,
>>> I defined my design with one EV as FFX (pre-to-post difference) and the
>>> remaining EVs as RFX coding for subject identity; and accounted for each EV
>>> in the contrast vector.
>>>
>>
>> At top, you .con file says there are 40 lines, but inside you use only 2.
>>
>> All the best,
>>
>> Anderson
>>
>>
>>
>>>
>>> Cheers,
>>> Tudor
>>>
>>>
>>> On 23 April 2018 at 05:18, Anderson M. Winkler <[log in to unmask]>
>>> wrote:
>>>
>>>> Hi Tudor,
>>>>
>>>> Please see below:
>>>>
>>>> On 16 April 2018 at 11:09, Tudor Popescu <[log in to unmask]> wrote:
>>>>
>>>>> Dear FSL experts,
>>>>>
>>>>> I have pre and post-intervention data for a single group of subjects,
>>>>> and I'd like to use VBM to compare the post-pre difference against zero, at
>>>>> the whole-brain & ROI-level. To that end, I've merged all "pre" files into
>>>>> a single 4D image; same for "post"; defined their difference as a new 4D
>>>>> image; and smoothed it:
>>>>>
>>>>> *fslmerge -t all_pre_struc_GM_to_template_GM_mod.nii.gz
>>>>> ./struc/??_1_struc_GM_to_template_GM_mod.nii.gz*
>>>>> *fslmerge -t all_post_struc_GM_to_template_GM_mod.nii.gz
>>>>> ./struc/??_2_struc_GM_to_template_GM_mod.nii.gz*
>>>>> *fslmaths all_post_struc_GM_to_template_GM_mod.nii.gz -sub
>>>>> all_pre_struc_GM_to_template_GM_mod.nii.gz
>>>>> all_diff_struc_GM_to_template_GM_mod.nii.gz*
>>>>> *fslmaths all_diff_struc_GM_to_template_GM_mod.nii.gz -kernel gauss 2
>>>>> -fmean all_diff_s2_struc_GM_to_template_GM_mod.nii.gz*
>>>>>
>>>>>
>>>>> For whole-brain, I then typed:
>>>>>
>>>>> *randomise -i all_diff_s2_struc_GM_to_template_GM_mod.nii.gz -m
>>>>> GM_mask -o analysis1 -d design.mat -t design.con -T -n 5000 *
>>>>>
>>>>>
>>>>> For the ROI analysis, I know that randomise can also be used together
>>>>> with the relevant ROI mask, but for now I extracted the average values of
>>>>> grey matter volume from my ROI across all subjects:
>>>>>
>>>>> *fslmaths /masks/F3 -mul /FAST/MNI_2mm_GM /masks_2mm/masks/F3_GM -odt
>>>>> float*
>>>>> *fnirt --ref=/stats/template_GM.nii.gz
>>>>> --in=/masks_2mm/masks/F3_GM.nii.gz --iout=/masks/F3_GM_template.nii.gz*
>>>>> *fslmaths all_diff_s2_struc_GM_to_template_GM_mod.nii.gz -mul
>>>>> /masks_2mm/masks/F3_GM_template finalImage -odt float*
>>>>> *fslstats -t finalImage -M*
>>>>>
>>>>>
>>>>> I have the following doubts:
>>>>> 1) Have I started from the correct files (*_struc_GM_to_template_GM_mod)
>>>>> to compute my difference image?
>>>>>
>>>>
>>>> I think so, although for the voxelwise you might want to start with the
>>>> smoothed one.
>>>>
>>>>
>>>>> 2) Should mean GMV values really be extracted from the *smoothed*
>>>>> difference image, since no voxel-wise stats are actually being done? It
>>>>> might be that smoothing averages values in a similar way anyway, but even
>>>>> so, do we not need to extract values from the maximum-resolution
>>>>> (unsmoothed) image?
>>>>>
>>>>
>>>> I would use the non-smoothed, otherwise voxels from outside the mask
>>>> contribute to what you find inside. Having said that I'm a bit worried with
>>>> the use of FNIRT above. So the masks aren't in standard space yet? To align
>>>> them, need to have a GM image in that same space as the mask, then in a
>>>> second stage, use "applywarp" to align the mask. I think your
>>>> /masks/F3_GM_template.nii.gz file may not be as correct as expected.
>>>>
>>>> All the best,
>>>>
>>>> Anderson
>>>>
>>>>
>>>>>
>>>>> Thanks in advance for any help!
>>>>>
>>>>> Cheers,
>>>>> Tudor
>>>>>
>>>>
>>>>
>>>
>>
>
