search

PennLINC / xcp_d

Post-processing of fMRIPrep, NiBabies, and HCP outputs
https://xcp-d.readthedocs.io
BSD 3-Clause "New" or "Revised" License
75 stars 26 forks source link

Scaling data dramatically affects ALFF #1032

Open tsalo opened 9 months ago

tsalo commented 9 months ago

Summary

Just to be clear, this documents a problem with the ALFF metric, not with XCP-D.

ALFF is dependent on the scale of the data, so the scaling step I added in #1020 dramatically changes the ALFF outputs. Unfortunately, it's not just a change in the ALFF scale, since some testing by @joellebagautdinova shows that the ALFF maps produced from scaled data don't show any of the spatial patterns that ALFF maps produced from unscaled data do.

The spatial patterns in the unscaled (i.e., original) ALFF maps are very similar to the mean BOLD maps from the preprocessed data (and probably the intercept image from the denoising step), so ALFF itself may be problematic, but it's also clear that the scaled ALFF looks bad.

Additional details

tsalo commented 9 months ago

The scaling difference is handled by #1033, but my concerns about ALFF remain unresolved.

tsalo commented 6 months ago

Should we do what other ALFF implementations have done and scale by the mean ALFF value in the brain?