comparing fMRIPrep and HCP Pipelines: parcelwise GLM statistics

As explained in the previous post (the introduction is here) with the target knots and contrasts for each task defined, and with the Schaefer parcellation to identify comparable sets of voxels and vertices, we're now able to see if there's a difference in the GLM between the four preprocessing combinations (surface, volume, fMRIPrep, HCP pipelines).

 

In the previous post I showed some of the group-average TENT curves. Above are the coefficients going into those curves, but for all 13 test people - the four green dots for AX-CPT parcel 90 above are (almost) the same means as the four lines at knot 4 (shaded) in the previous post ("almost" since the green dots here are normal means, while the previous ones were robust; the individual participant's coefficients are the same). A unique plotting symbol is used for each participant, and the four columns of points at each parcel are the four preprocessings. Plots with the distributions for all 400 parcels are here (source here).

For a simple way to describe the distribution at each parcel I used t-tests: is the across-subjects mean different than zero? These plots have each parcel colored by t, HCP in the first row, then fMRIPrep, then their difference. The differences are HCP - fMRIPrep, so cool colors indicate parcels with larger ts when preprocessed with fMRIPrep. There are a lot of cool colors in the difference images, but it's still a lot of data to absorb; the next (final?) post in this series will summarize these statistics for just 20 of these parcels.