Hi,

If you preprocess your data with FEAT rather than through the melodic GUI then you need to transform your functional data to MNI. That is, your filtered_func_data files are still in individual functional space and you're feeding these data (all in different spaces) into a single melodic run - this will not work properly and will instead produce garbage results....
There is no way to fix this post-hoc (i.e. transforming melodic_IC) - you need to re-run your analysis.

hth
Christian


On 21 Aug 2013, at 06:27, Luigi Angelo Maglanoc <[log in to unmask]> wrote:

> Dear FSL mailing list,
>  
> I am unable to overlay a g-ICA spatial map, and the various outputs
> of randomise onto an MNI-152 template.
> I know this question has been posed before, but I don't seem to be able
> to do what has been suggested several times. Any help would be greatly
> appreciated!
>  
> The raw fMRI data dimensions:
> Voxels: 64 x 64 x 39
> Dimensions: 3.75 x 3.75 x 3mm
>  
> For the preprocessing, the data was run through FEAT GUI, with an initial structural image, a main structural image,
> and registered to standard space (MNI152_T1_2mm_brain non-linear).
> The, the resulting filtered_func_data (which is in standard space, right?) files were used as the input in melodic with the following line:
>  
> melodic -i ${scriptdir}/${filelist} \
>     -o ${workdir}/final/${filelist}_${modelorder} \
>     -v --tr=2.0 --report\
>     --guireport=${workdir}/final/${filelist}_${modelorder}/report.html \
>  
>  
> I tried to add --bgimage=${scriptdir}/bg_image_3mm (where the bg_image_3mm is a resampled MNI_152_T1_1mm_brain),
> but it wouldn't run, with the following error message:
>  
> 
> ERROR:: background image and data have different dimensions
> 
> 
> I tried running it with MNI152_T1_2mm_brain, but that didn't work either
>  
> I have tried resampling the melodic_IC.nii.gz and various dual regression/randomise outputs to
> match an MNI template, with flirt and applywarp, i.e.
>  
> flirt -ref bg_image_3mm.nii.gz -in melodic_IC.nii.gz -out test_melodic   ----> the melodic_IC.nii.gz file is resampled, but it has been spatially shifted and grossly distorted in shape
>  
> flirt -ref $FSLDIR/data/standard/MNI_152_T1_1mm_brain -in melodic_IC.nii.gz -applyxfm -out test_melodic ---> the melodic_IC.nii.gz is resampled, but shifted to the right
> flirt -ref $FSLDIR/data/standard/MNI_152_T1_1mm_brain -in melodic_IC.nii.gz -applyxfm -init melodic2MNI.mat -out test_melodic --->same asthe above (.mat file by using -omat first)
> flirt -ref $FSLDIR/data/standard/MNI_152_T1_1mm_brain -in melodic_IC.nii.gz -applyisoxfm 3 -out test_melodic ---> same as the above
> applywarp -i melodic_IC.nii.gz -o test-r $FSLDIR/data/standard/MNI152_T1_2mm_brain.nii.gz ----> same as above
> 
>  
> I tried following the line fromhttp://fsl.fmrib.ox.ac.uk/fsl/fslwiki/MELODIC/FAQ#How_do_I_transform_the_MELODIC_results_from_a_low-resolution_standard_space_to_a_higher-resolution_one.3F
> 	• flirt -in image3mm -ref $FSLDIR/data/standard/MNI152_T1_2mm -applyxfm -usesqform -out image2mm
> again, the melodic_IC file is resampled, but now it is shifted in a different way.
>  
> When I do
> flirt -in melodic_IC -ref melodic_IC applyxfm 3 -out melodic_IC_3mm ---> it gets resampled to have 3 x 3 x 3 dimensions, but the voxels (80 x 80 x 42) doesn't match
> the bg_image_3mm, and subsequent flirt with -applyxfm shifts the output like the other examples.
>  
> Another issue is that the various atlases for the raw fMRI data, filtered_func_data, to g-ICA to dual regression/randomise output is wrong:
> the hippocamapal area is labeled as being frontal cortext, so it seems that everything has been shifted
> posterior.  Will the labelling have an effect on the actual results?
>  
> I also tried resampling an MNI template to match the dimesions of the fMRI data
> flirt -ref melodic_directory/mean.nii.gz -in $FSLDIR/MNI152_T1_1mm_brain -out resampled_MNI.nii.gz ---> it works, but the image quality is bad
>  
> Is there a way to resample the melodic_IC.nii.gz and dual regression/randomise outputs to an MNI template, or another
> way to overlay them onto an MNI template? Or to resample an MNI template to melodic_IC, but with "better" image quality?
> Or is the better solution to upsample the raw fmri data? The problem with this is that there isn't much disk space available...
>  
>  
> Sorry for the long post, and thank you so much in advance!
>  
> Sincerely
> Luigi
>