Nicreap-
I just want to make sure I'm reading this correctly. When you ran align_epi_anat.py on the original data, there was good alignment, but you saw artifacts around the Ventricles? And when you deoblique first and then run align_epi_anat.py you get a cutoff and bad alignment?
I suspect the problem with the second (deoblique and then alignment) is that align_epi_anat.py does deobliquing within itself. You can turn this off and it may help (though I don't think it will make much impact on the artifacts). What is the rest of your processing pipeline? Are you using afni_proc.py?
In your sparse (clustered) sampling, are you doing anything to establish stead-state images?
-Peter