Steve,

Thanks for your response. When I run
fslstats mean_func.nii.gz -M <returns> 7679.119706
fslstats mean_func.nii.gz -k mask.nii.gz -p 50 <returns> 9348.549805
Since mean_func.nii.gz is masked with mask.nii.gz, it seems the 4D mean in the brain is not that close to 10000, but is between 7000 to 8000. The median is fairly close to 10,000 though. After looking at the code, I think I know what is happening. The mask used for determining the median is overwritten with a dilated mask. Then after SUSAN is run (which does some kind of smoothing of edge brain voxels I am guessing?), the data is masked again with this dilated mask, this dilated mask ends up includes some voxels outside of the brain that have had data smoothed into them with much lower intensities, this drives the mean down to about ~7600 and the median down to about ~9300, for this example subject.

I think we would like to save all the intermediate "prefiltered" output but I can't find the line in the feat code that I see in report_log.html (/bin/rm -rf prefiltered_func_data*?) I think I would just comment this out. And maybe edit the code to save the undilated mask. I will search for the other posts on using mean_func.nii.gz for % BOLD normalization. Thanks again.


###start of code

/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/fslmaths mask -dilF mask

/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/fslmaths prefiltered_func_data_mcf -mas mask prefiltered_func_data_thresh


/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/fslmaths prefiltered_func_data_thresh -Tmean mean_func

/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/susan prefiltered_func_data_thresh 531.74926725 2.54777070064 3 1 1 mean_func 531.74926725 prefiltered_func_data_smooth


/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/fslmaths prefiltered_func_data_smooth -mas mask prefiltered_func_data_smooth

/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/fslmaths prefiltered_func_data_smooth -mul 14.1043917912 prefiltered_func_data_intnorm


/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/fslmaths prefiltered_func_data_intnorm filtered_func_data

/home/bfs-raid3/ryanm/DOWNLOADS/fsl-4.1.6-centos5_64/bin/fslmaths filtered_func_data -Tmean mean_func


/bin/rm -rf prefiltered_func_data*

###end of code

On Tue, Feb 1, 2011 at 3:12 AM, Stephen Smith <[log in to unmask]> wrote:

Hi - actually it's a bit more objective / consistent than that - within the masked/thresholded brain image, the median intensity is found and used to determine the global (4D) intensity scaling - so in fact the mean intensity inside the brain should be pretty close to 10000 for all subjects.

However if you want to find voxelwise % signal change then you can use the mean_func image for normalisation - see previous posts on that.

Cheers.

On 31 Jan 2011, at 21:28, Christopher Bell wrote:

Feat now normalizes each filtered_func_data to a different 4D mean when running pre-stats, usually the 4D mean ends up being somewhere between 7000 to 8000. I wish to find the % signal change using featquery. How would featquery know what each subject's image has been scaled to and find the appropriate % bold change?

Chris Bell
University of Minnesota

---------------------------------------------------------------------------
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
---------------------------------------------------------------------------