Process two-echo data like multi-echo without echo combination

[tsalo]

What would you like to see added in fMRIPrep?

#3362 got me thinking that it might be good to process two-echo data similarly to multi-echo data, except without any tedana-based echo combination steps. Or, maybe we could do a basic unweighted average when there's only two echoes?

1. Motion correct using the first echo.
2. Generate the boldref image either from one of the echoes or just take an average.
3. Apply the same motion parameters to both echoes.
4. Output the preprocessed echoes in target spaces as separate files, as if you were running fMRIPrep twice with --echo-idx set to each of the echoes.

I think we could have a new --me-t2s-fit-method option called mean to just take the average across echoes. --me-output-echos might still be applicable, depending on if we wanted to just output the mean combined data or if we wanted to stick to outputting individual echoes.

NB: I'm not actually sure what the use-cases for two-echo protocols are. I know dual-echo denoising (Bright & Murphy (2013)) is a thing, but AFAIK that's only feasible in very limited cases where you can get the first echo early enough to not have any BOLD signal.

Do you have any interest in helping implement the feature?

Yes

Additional information / screenshots

No response

[tsalo]

Based on this NeuroStars topic, it looks like fMRIPrep is randomly selecting one of the echoes to process (or at least the metadata says as much), so I think this should be bumped up in priority. The author of that topic, @jaetzel, would be fine just raising an error when two-echo data are provided, which seems like a reasonable short-term solution until we figure out the best way to process two-echo data.