Hi Devin,
You get the holes in your mask.img (I assume this is the file that you are talking about) because SPM uses another threshold to determine what the final mask.img looks like, not the -inf that you used to create a binary version of the brainmask.img.
To see this threshold, go to spm_defaults.m and search for defaults.mask.threshold. The default value is .8, which is a stringent threshold (why you see the holes). When you specify an explicit mask, it merely tells SPM to search within the voxels designated as 1s in the mask. SPM reads the raw data in your functional images at this voxel and uses defaults.mask.threshold to determine what the final mask.img looks like.
For the Imcalc step, you may want to use the formula 'i1>threshold' where threshold is not -inf but a value that will result in a good mask where all brain voxels are set to 1 (check reg the binary image with brainmask to determine what this value is). Then modify the defaults.mask.threshold to be more liberal (e.g -inf) and then run the first-level analyses. You should get a better mask from this method.
I posted a similar response to another post. https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1009&L=SPM&P=R36441&1=SPM&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4
Hope that helps,
Vy T.U. Dinh
Research Assistant, Neurological Sciences
Rush University Medical Center
Phone: (312) 563-3853
Fax: (312) 563-4660
Email: [log in to unmask]
________________________________________
From: SPM (Statistical Parametric Mapping) [[log in to unmask]] on behalf of Devin Sodums [[log in to unmask]]
Sent: Tuesday, October 25, 2011 12:08 PM
To: [log in to unmask]
Subject: [SPM] Holes in brainmask even after running explicit mask in model estimation
Dear SPM users,
I have been trying to correct for some signal drop out (holes in the brainmask.hdr that is outputted from first level model estimation) and can't seem to resolve the problem. My steps are the following
1. Created binary version of brainmask.nii image in apriori folder provided by spm8 using ImCalc with a threshold of -infinity
2. Included this mask under the explicit mask option and ran first level model estimation again
As I understand it, with a mask that was binarised to negative infinity, there shouldn't be any voxels left out since it is such a liberal threshold. However, there are still holes (I'm assuming because of signal dropout) in the OFC and other areas. Any help/direction would be greatly appreciated.
Cheers,
Devin Sodums
Research Assistant, Cognitive Neuroscience Unit
Montreal Neurological Institute
|