One would expect the regression matrices to be identical for the two cases (though with a degree of freedom difference, as one bigger matrix might be broken into two), and assuming that makes this more of a question of registration, given that the subject left the scanner between the 2 sessions. And note that there will be distortion differences between the EPI time series. Whether data was collected to reduce distortion effects can be left as a separate consideration.
The basis registration question here is how to reasonably align the pre- EPI data with the post- EPI data. Some choices (from the perspective of afni_proc.py) include:
1a. Do a single analysis, with no special correction.
This is undesirable, as the cross-session EPI data will surely not register well enough with 3dvolreg.
1b. Do a single analysis, where each run is registered to its own MIN_OUTLIER volume, and where the per-run MIN_OUTLIER bases are also registered to the global MIN_OUTLIER volume.
This might work reasonably well. It relies on 3dAllineate to align EPI from different sessions.
2a. Do 2 analyses, using the same anatomical reference.
The MIN_OUTLIER EPI reference volume would be aligned to the anatomical reference (before possibly being taken to standard space - a separate topic), separately for each session’s analysis.
This might also work reasonably well. It relies on the cross-session EPI registration of each session’s MIN_OUTLIER volume being aligned to the anatomy.
2b. Do 2 analysis, using different anatomical references.
This is similar to 2a, but aligns each session’s MIN_OUTLIER to that session’s anatomy, and relies on consistent registration of the multiple anatomies to the template. It assumes use of a standard space.
This should work reasonably well too, but particularly for non-linear registration to the template, it relies on that working well. Normally small problems there would affect the registration of one subject to the group. Here it would also affect the registration of one subject across sessions.
Each of those seems reasonable, but they are different. It is not clear which way to advise.