Hi Doug/All,
This is a question about registering several functional scans from the same subject in the same space. The functional scans are from different sessions, so even if the slice recording parameters are identical, the head will inavoidably be in different locations and with different rotation. The data have already been collected, so we cannot use any on-line corrections.
Four different techniques come to mind that could be used to achieve this, but I would expect that the accuracy of the methods varies a lot. Optimally IMHO one should register the different functional runs with respect to each other directly, simply because the EPI images are very similar with respect to each other (or at least their T1 contrast is). The high-resolution structural image is very different compared to the EPI images so functional-to-structural registration is expected to be much more inaccurate.
Here are the four different techniques (please tell me if you can think of more):
1. One possibility would be to use mc-sess for all runs from the same subject (place all the runs from the different sessions in a single "bold" directory and just run mc-sess). Possible problem: If I have understood correctly, mc-sess only corrects for motion within slices. Therefore the different head rotations would make this technique very inaccurate. Or does there exist an updated version of mc-sess that could handle even large displacements and rotations?
2. A second possibility would be to use one of the automatic functional-to-structural registration programs and register each functional run to the same hires structural image. However, based on the discussion above, there are expected to be inaccuracies in the functional-to- structural registration that could be large enough to make comparisons across the functional sessions meaningless (i.e., the differences would mainly be driven by inaccuracies in the registration).
3. A third possibility would be to use the same technique as for registration across subjects, i.e., resampling the different runs from the same subject into a spherical space calculated from this specific subject. Again this is just as variety of registration from functional to structural, and would be expected to provide suboptimal results.
4. A fourth possibility would be to find a method for registering all the functional images in a single space with respect to each other, e.g. by picking one of the functional sessions as the target space and resampling all other functional sessions from this subject into this space (without using the hires structurals at all). I guess this would be kind of motion correction but capable of handling rotations. If such a tool exists, it might be more accurate than any of the methods above, simply because the functional images are more similar compared to each other than compared to the subject's hires structural image.
Any ideas/suggestions which method would give the most accurate result? And what kind of tools currently exist?
Thanks,
Tommi
--- Tommi Raij MGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging Bldg 149, 13th St Charlestown, MA 02129 U.S.A.