Hi,
I've had a look at your data and there are a couple of things that might explain your registration problems:
1 - your fieldmap is noisy at the edge
2 - your brain extraction is removing parts of the brain, especially in the frontal lobe
The first one is the most serious and is easy to fix by eroding the fieldmap magnitude brain image before putting it into fsl_prepare_fieldmap. This should remove the noisy edge of the fieldmap, and that will have a significant impact on the edge of the brain in the registrations.
The second one you can improve on by trying different options in BET.
So try these and see if the registration is fixed. If not, let us know.
All the best,
Mark
On 30 May 2013, at 17:08, Benjamin Philip <[log in to unmask]> wrote:
> Certainly! The structural images are already in that zip file I uploaded yesterday (sub03_05_t1 and _t1_brain.nii). The new zip file includes the original EPI (run001.nii.gz), denoised EPI (denoised_data.nii.gz, which is what I used as input for my main 1st-level analysis), fieldmaps (brain-extracted magnitude image, and fieldmap created via fsl_prepare_fieldmaps), and the complete "reg" and "reg_standard" directories from the first-level analysis.
>
> I had trouble putting such a big file onto your server, so I've used dropbox again: https://www.dropbox.com/s/pd1qqiexy2n3qym/regTestPhilip2.zip
>
> Thanks,
>
> Benjamin Philip
>
> On Thu, 30 May 2013 09:55:40 +0000, Mark Jenkinson <[log in to unmask]> wrote:
>
>> Dear Benjamin,
>>
>> I'm afraid there really isn't enough information in these png files for me to determine the fine details of the registration performance. The brain extraction looks fine though.
>>
>> Can you please upload the nifti files, including the original fmap, EPI and structural images (as entered into the FEAT GUI) plus the registration output nifti files (e.g. example_func2highres.nii.gz, highres2standard.nii.gz, etc.). Our upload website is:
>> https://oxfile.ox.ac.uk/oxfile/work/extBox?id=68312615463381F4C
>>
>> From this I should be able to work out what can be improved.
>>
>> All the best,
>> Mark
>>
>>
>> On 29 May 2013, at 17:12, Benjamin Philip <[log in to unmask]<mailto:[log in to unmask]>> wrote:
>>
>> I have some analyses with poor standard-space registration outcomes. "Poor" not "failed"; there are no overt errors on the registration-summary pages, but especially on second-level analysis I see swathes of fMRI-activated voxels outside the brain.
>>
>> What I've done:
>> 1) BET, an imperfect but reasonable job (as good as I could get it).
>> 2) First-level pre-stats separately: including motion-correction, B0 unwarping, and its own pass through registration. I used the output of this for ICA denoising.
>> 3) I ran first-level stats/post-stats on ICA-denoised data. BBR registration to main structural image, 12dof nonlinear (10mm) to standard space.
>> 4) I ran a second-level analysis taking a simple average across 4 runs. No clear differences between runs, so I've only included first-level stuff for one run.
>>
>> What I see:
>> 1) At the first level stats, a little bit of "leakage" beyond the edge of the brain, but not too bad.
>> 2) At second level, huge swathes of activation outside the brain.
>>
>> What I've included in the linked file:
>> 1) t1 and t1_brain files, illustrating BET
>> 2) .fsf files for prestats, stats, and second-level analysis
>> 3) All the PNGs from the registration pages at both levels
>> 4) Sample stats (thresh_zstat and rendered_thresh_zstat) images from both levels.
>>
>> Everything appears okay until the second-level stats images, which have huge non-brain activations. Judging by masksum_overlay at the second level, the problem seems to come from registration to standard space... but I'm at a loss how to improve it. Any suggestions or troubleshooting steps would be greatly appreciated!
>>
>> Files at https://www.dropbox.com/l/KVztbYxsJojB7NQBys3kM9
>>
>> Thanks,
>>
>> -----
>> Benjamin Philip, Ph.D.
>> Postdoctoral Fellow - Freylab
>> University of Missouri
>> 206 Melvin H. Marx Bldg
>> 1416 Carrie Francke Drive
>> Columbia, MO 65211
>> [log in to unmask]<mailto:[log in to unmask]>
>>
>>
|