Hi Chris:
I can think of two possible reasons for the region getting "lopped off"
1) Not all your functional images may have good signal in this region.
Perhaps there is no signal on some images and SPM only includes voxels where
all images have signal.
2) More likely is that it appears that the signal in this area is lower than
in other parts of the brain due to inhomogeneity. This area may therefore
fail the SPM threshold for including voxels, which for fMRI is 100% of the
whole brain mean for that image. Paradoxically, if your mask excludes voxels
that have a lower value than the mean (and are not NaN) this may actually
increase the mean and worsen the problem. Note that the explicit mask does
not "make" spm include the in-mask areas, it only makes spm exclude the out
of mask areas.
You might try modifying the thresholds. These are stored in SPM.xM.TH. You
could try multiplying the values by 0.9 or 0.8. There were some messages on
the list sometime back about this. Just be aware you may include things you
don't want.
Regards,
Darren
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping)
> [mailto:[log in to unmask]] On Behalf Of Christopher Summerfield
> Sent: Monday, February 12, 2007 9:11 AM
> To: [log in to unmask]
> Subject: [SPM] unfortunate masking during estimation
>
> hi experts,
>
> a question about masking during estimation under SPM2...is
> there any reason why the following should be happening? A
> subject whose brain normalizes fairly reasonably to the EPI
> template, gets a portion of the frontal lobe lopped off
> during estimation, as can be see in the mask produced by the
> first-level analysis (I posted the output of 'check reg'
> online at
> http://angora.snv.jussieu.fr/~chris/stuff/wierdmasking.tiff
> ....top left EPI template; top right, smoothed normalized EPI
> image for this subject; bottom left, mask.img created during
> estimation; bottom right, a con* image). As you can see,
> although the smoothed normalized image is a good match to the
> template, the most anterior bain gets masked out. Why is
> this? I tried the following so far:
>
> 1) check realignment params (they are all OK, <1.5mm movement
> across 8 sessions), no bouncing around in the scanner.
>
> 2) check that there is not a misplaced volume (I extracted
> data from a voxel falling within the missing part of the mask
> for each swavol* in each run, and checked for zeros or
> NaNs...there were none. Nor were the values all the same.
> They were all nice reasonably consisten values.
>
> 3) I tried to manually set an explicit mask corresponding to
> the EPI template, by setting SPM.xM.VM=spm_vol('EPI.mnc')
> before analysis, vainly hoping that SPM would let me choose
> what to mask out (it didn't...same old problem).
>
> 4) I tried changing the bounding box to the EPI template
> default, which seems to be a bit larger...this made no
> difference either.
>
> Any ideas? many thanks....
> Chris
>
>
>
>
>
>
> Christopher Summerfield
> [log in to unmask]
>
|