Same here Mary! and do need to thank again Alfonso for detailed explanation and the wonderful tool.
So I used to do scrubbing procedure following Power's paper,removing those bad volumes after all preprocessing (so won't have a problem of bandpass with continuous data), and trying to keep all subjects with same number of remaining volumes (so to avoid different length of  data across subjects by throwing out also some good volumes for some subjects), the minimum length of data would be 4minutes or higher can be set to maximize the possible number of subjects to include as well as a criteria to exclude subjects (so then we don't have to throw out subjects that only have couple of big spike that would usually need to be excluded by the criteria Mary mentioned).
Anyway, I do think the way ART implemented has its advantage like Alfonso explained, and I hope in future version of CONN can add a choice to disregard/remove the bad volumes identified in the art_outlier.mat file at the first level analysis step (not physically remove but just in calculation step) so it won't have the issue in bandpass etc that requires continuous data. It'll be wonderful! :)
Hope my thoughts are conveyed clear. Thanks!
Xiaozhen