search

NBCLab / power-replication

A replication and extension of Power et al. (2018)
https://www.overleaf.com/read/swgjxcjqytxg
Apache License 2.0
2 stars 0 forks source link

Code review on distance-dependent motion-related artifact plots #2

Closed tsalo closed 3 years ago

tsalo commented 6 years ago

A test notebook is here and the relevant functions are here. The plots look alright, but I don't know if they're implemented correctly. A code review would be nice at some point.

tsalo commented 5 years ago

The significance tests weren't making sense before when run on the 40-subject ADHD dataset downloaded using nilearn, but I think I figured out the bug (I wasn't calculating the slope correctly). Now, the slopes are generally significant (i.e., there is distance-dependent impact of motion on functional connectivity), although the intercepts are not (i.e., there is a general effect of motion on functional connectivity). Given that we expect for the slope to be significant, regardless of preprocessing pipeline, in single-echo data, that part makes sense to me. I also don't know what preprocessing is applied to the dataset, but it could hypothetically include CompCor or GSR, which would explain why the intercepts aren't generally significant.

While I would still like to get a second (and even third) pair of eyes on the code itself, I think it's pretty solid (though really inefficient) at this point.

Here are example figures, if anyone wants to take a look:

** Update: I'm moving the motion analysis code to a separate repository (tsalo/ddmra), so that I can make it public.