BOLD-to-anatomical coregistration can be made more consistent by creating an average BOLD reference, registering all run references to that average, and coregistering the average to the anatomical. This takes advantage of averaging to improve the anatomical fidelity of the BOLD reference prior to coregistration, and avoids algorithmic non-determinism as a source of variation by only registering once.
In nipreps/nibabies#517, we are introducing --bold-coreg-level {run|session}. After discussion with iProc developers, it would be useful to also include a subject level.
The run-level should be preserved for backwards compatibility, although I think it makes sense to transition to session as a default. Subject-level would enable precision-imaging workflows, where session-to-session differences are considered background noise and more BOLD references improves the spatial SNR.
We probably prefer session to subject as a default aggregation level because that allows us to preserve consistent behavior when processing is split across sessions or data are processed incrementally as sessions are collected.
BOLD-to-anatomical coregistration can be made more consistent by creating an average BOLD reference, registering all run references to that average, and coregistering the average to the anatomical. This takes advantage of averaging to improve the anatomical fidelity of the BOLD reference prior to coregistration, and avoids algorithmic non-determinism as a source of variation by only registering once.
In nipreps/nibabies#517, we are introducing
--bold-coreg-level {run|session}. After discussion with iProc developers, it would be useful to also include asubjectlevel.The run-level should be preserved for backwards compatibility, although I think it makes sense to transition to session as a default. Subject-level would enable precision-imaging workflows, where session-to-session differences are considered background noise and more BOLD references improves the spatial SNR.
We probably prefer session to subject as a default aggregation level because that allows us to preserve consistent behavior when processing is split across sessions or data are processed incrementally as sessions are collected.