Minimally preprocess BOLD phase data

[tsalo]

What would you like to see added in fMRIPrep?

I think phase reconstruction is going to become more popular over time as tools become available to leverage it. It would be great if fMRIPrep could apply minimal preprocessing to these data. The following steps would probably be enough:

1. Phase unwrapping, preferably leveraging multi-echo information when available, using ROMEO.
2. Motion correction (no STC necessary, AFAIK)
3. Distortion correction

Do you have any interest in helping implement the feature?

Yes

Additional information / screenshots

This is related to #3308 (NORDIC uses phase data, when available), nipreps/sdcflows#36 (dynamic distortion correction include phase unwrapping), and vanandrew/warpkit#12 (warpkit will hopefully output unwrapped phase time series soon).

Some potential applications of the phase data:

• phase regression (e.g., Curtis et al. (2014))
• phase jolt time series (OHBM 2024 poster, code)
• improved T2* estimation (Cohen-Adad et al. (2012))
• complex-valued ICA (e.g., Adali and Calhoun (2007))
• complex-valued modeling (e.g., Yu et al. (2022))

I think that most processing could be pulled from fMRIPrep derivatives (e.g., transforms) and applied to the phase data in an fMRIPost workflow, but the one application I'm concerned about is the T2* estimation, which has implications for the optimal combination done within fMRIPrep.

I have created an fMRIPost repo (nipreps/fmripost-phase) to handle at least some of the processing/applications I've proposed.

[tsalo]

The more I think about it, the less I worry about the optimal combination. fMRIPost-Phase and/or fMRIPost-Tedana can recalculate the T2* using phase data and re-generate optimally combined data in boldref space. Even if optimally combining the data using the complex-based T2* map is "better", I don't think fMRIPrep's coregistration (the main thing driven by the optimally combined data) would be adversely impacted much.

I think I'll just close this issue in favor of working on fMRIPost-Phase. If there's some kind of massive shift in the field and phase data become as common as multiband, or even multi-echo, then we can revisit this.