Dear Anderson
Thank you for your reply.
The chunks I have created are not of equal size. I have n chunks of 20
subjects each and one chunk with 19 subjects in it. However, I am actually
not using the mean_FA but rather after deriving a study mask I am masking a
target and a target skeleton and creating the individual skeletons on a
subject level rather than as a all_skeletonised 4D image.
Sorry, just to make it clearer, considering that mean_FA is used to derive
the mask, would using unequal chunks affect the mean_FA_mask?
Thanks again.
Best wishes,
Rali
On 13 January 2013 19:03, Anderson M. Winkler <[log in to unmask]>wrote:
> Dear Rali,
>
> Yes, your procedure should work for a mask. However, a procedure like this
> for the mean_FA will only give correct results if all chunks have the same
> number of subjects. If you are unfortunate and have a prime number of
> subjects, then there would be no solution except discard some of them... An
> alternative that doesn't require explicitly dividing the subjects into
> chunks would be to increment the mask and the mean_FA for each subject
> inside a loop. Below is a sketch, you can modify according to your needs
> (there are probably other solutions too):
>
> 1) Threshold the template (or the image of any subject) using a very high
> number, guaranteeing that nothing survives, so it contains only zeroes at
> the beginning of the loop:
> fslmaths template.nii.gz -thr 20000 toincrement.nii.gz
>
> 2) Make a copy, now full of ones (intersection goes with multiplication,
> and 1 is neuter):
> fslmaths toincrement.nii.gz -add 1 tomultiply.nii.gz
>
> 3) Start a counter in case you don't want to count how many subjects you
> have:
> c=0
>
> 4) Loop over all subjects:
> for i in ${list_images} ; do
> fslmaths ${i} -max 0 -bin -mul tomultiply.nii.gz tomultiply.nii.gz -odt
> char
> fslmaths ${i} -max 0 -add toincrement.nii.gz toincrement.nii.gz -odt
> double
> c=$(expr ${c} + 1)
> done
>
> 5) At the end of the loop, the mask is simply the file named
> "tomultiply.nii.gz", and you can rename it (use immv for extension-free
> renaming):
> mv tomultiply.nii.gz mask.nii.gz
>
> 6) The mean FA is the file named "toincrement.nii.gz" divided by the
> number of subjects (counter), with the mask applied:
> fslmaths toincrement.nii.gz -div ${c} -mas mask.nii.gz mean_FA.nii.gz
>
> Note that a for-loop is slower than averaging over the 4D (-Tmean), but
> the benefit is that you can operate in very large datasets without
> splitting the sample in subgroups all of the same size.
>
> Hope this helps!
>
> All the best,
>
> Anderson
>
>
> 2013/1/13 Rali Dimitrova <[log in to unmask]>
>
>> Dear FSL users
>>
>> Let say I was working on a large data set that would require to adjust
>> the tbss_3 script to run chunk-wise. What I actually need as output is a
>> study specific mask (mean_FA_mask). Would the following be correct:
>>
>> 1. Create a mean across all individual in chunk
>> 1.1. merging FA_to_target into n chunks - fslmerge -t $chunk <input>
>> 1.2. creating a mask for each chunk - fslmaths $chunk -max 0
>> -Tmin -bin mean_FA_mask_$chunk -odt char
>> 1.3. applying the mask to each chunk - fslmaths $chunk -mas
>> mean_FA_mask_$chunk $masked_chunk
>> 1.4. creating a mean_FA for each chunk- fslmaths $masked_chunk -Tmean
>> mean_FA_$chunk.nii.gz
>>
>> 2. Merging all chunks - fslmerge -t
>> mean_chunk input<mean_FA_$chunk.nii.gz>
>>
>> 3. binarise the mean_chunk fslmaths mean_chunk
>> -max 0 -Tmin -bin study_specific_mask -odt char
>>
>> Considering that the study_specific mask is a binary file it should not
>> be affected in any way by the order of my images in each chunk I am
>> merging, am I right?
>>
>> Your help will be greatly appreciated!!
>>
>> Best wishes,
>> Rali
>>
>
>
--
Best wishes,
Rali Dimitrova
Division of Psychiatry,
University of Edinburgh
