Dear Jessica,
> I'm attempting a 2nd order analysis to determine differences between two groups
> of subjects on a particular task. (It's MRI data.) I selected a two-groups
> t-test on the basic models, selected the con*.imgs, etc. I selected no mean
> scaling, and proportional thresholds at 0.8, no explicit masking, and the
> global calculation to be the mean--taking the defaults wherever possible,
> essentially.
>
> When the analysis is done, I looked at mask.img to see what parts of the brains
> were being included--and mask.img only shows a very small blip of included
> brain, right above the cerebellum. The con*.imgs are almost full-brain, so why
> would the mask.img take everything out? My hypothesis is that the 0.8 is
> somehow not right, but I don't know what it should be otherwise.
For the 2nd level, you have contrast images as input. Here you can use a
SPM feature called 'implicit masking'. Implicit masking is a function of
the data, in contrast to 'explicit masking', where you would input an
additional mask. One thing that the implict masking does for you is to
remove any voxel time series from the analysis that has at least one
'NaN' (not a number) in it. Contrast images have NaN at voxels, where
the 1st level didn't estimate parameters.
Therefore, a solution for analyzing contrast images at the 2nd level is
to not use an analysis threshold. This will make SPM analyze all voxels,
where there is data for all contrast images, but remove any voxel from
the analysis, where there is at least one 'NaN'. This will give you the
same mask as for the analysis at the 1st level.
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Imaging Neuroscience
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
|