Dear Mike,
During single-subject analyses a mask is generated based on intensity values of the time series. The idea is to exclude low-intensity background voxels irrelevant for the analysis, although the procedure will also exclude some of the boundary voxels. The default settings usually work reasonably well (one would expect problems for data with large intensity inhomogeneities like in multi-coil data); you lose some of the "rim voxels", but as one usually has forwarded smoothed images the lost ones tend to correspond to voxels which fall outside the unsmoothed brain anyway.
Group analyses are restricted to those voxels which contain data across all single-subject models / beta or con images. Thus in your case, the overlaid activations do not extend towards the surface of the right hemisphere as the voxels were not part of every single-subject model. This is no problem for the analysis though.
Also make sure that functional and structural data were coregistered properly, that the normalisation/segmentation worked well, and that the same normalisation parameters / flow fields were used when normalising the EPI series and the structural volumes; if any of those fails and/or if you use different normalisation strategies (e.g. EPI onto EPI template, structure onto T1 template) you will introduce distortions and/or biases.
Best
Helmut
|