 |
 |
> 1) For normalization, do I need to have the entire skull as well as the > brain? While trying to learn SPM, I applied the normalization routine to > the anatomicals as a quick test on a small dataset (I realize this isn't > something you'd normally do). The normalized anatomicals appeared > distorted anywhere where the image was cropped or the skull was missing > (at the top of the image where the skull was out of the coil range; and at > the bottom -- cerebellum and inferior frontal -- where there was full > brain but possibly not enough skull). In those locations, the brain > appeared stretched out to where the skull ought to be. The spatial > normalization output ps image was similarly distorted, but I was told this > is "normal". If you find that the spatial normalization is distorting your images un-necessarily, then simply reduce the number of parameters so that the distortions do not occur. Spatial normalization matches images to the template by minimizing the residual sum of squared differences between the images. If the template looks nothing like your images, then you will need to create a new one. All the voxels within the image volume are used in the matching, so if for any reason, the top or bottom of the image has been set to zero (eg. after reslicing), then you are likely to have problems. If the top and bottom planes contain poor data, then it may be a good idea to chop these planes from the image volume (possibly by setting the the OFFSET so that it SPM ignores the first few planes, and reducing DIM(3) to a more appropriate value). There are sometimes artifacts near the top and bottom of the FOV. These are because of the smoothing that is performed prior to estimating the spatial normalization parameters. Voxels near the edges can not be smoothed properly (because the smoothing algorithm does not know what lies outside the field of view). This is fixed in the next release. > 2) Do I need to have the functional data include the anterior commissure > to normalize? Our protocol includes both functional slices that cover the > superior part of the brain (sup. parietal and sup. frontal) and anatomical > slices that cover most of the brain. The slices are chosen such that the > functional slices are a subset of the anatomicals (same orientation and > planes) but the anats are thinner and cover a larger extent. Thus I can > select the AC in the anats but not the functionals (planes don't go down > that far). For non-normalized data, I can select a non-AC reference > point (though Glass brains are way off). Is this data SPMable? Providing the ORIGIN refers to the same point in both images, the planes are parallel in both sets of images, and the voxel sizes are filled in correctly, then SPM should consider your images to be in register. However, if your ORIGIN is a long way from the AC, then you may need to alter the affine starting estimates for the spatial normalization accordingly. The ORIGIN need not necessarily be the co-ordinate of a voxel within the image volume. It is perfectly allowable for it to reflect a voxel co-ordinate outside the image. The only problem is that a bit of care is needed to compute the appropriate co-ordinate. More advanced users can provide ".mat" files for the structural and functional images. These files can provide more information about the relative positions and orientations of the images than the ORIGIN and VOX fields alone. Good luck, -John %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%