non-parametric vertex-wise statistics

[vsteiger]

hi all,

I'm using RMINC for vertex-wise analyses of shape data. So far I have only applied vertexLM but I would like to run a more robust non-parametric approach. Is there a non-parametric approach (e.g. randomize or TFCE from FSL, monte carlo simulation from FreeSurfer) implemented in RMINC or soon to come?

Best,

Vivian

[cfhammill]

Hi Vivian, yes I just wrote functionality for doing randomization tests for TFCE and linear models. Should be releasing soon.

[vsteiger]

hi chris, great news. I'd be happy to (beta)-test...just in case ;-)

[cfhammill]

Excellent, thanks!

You can try it out now if you do

devtools::install_github("Mouse-Imaging-Centre/RMINC", ref = "develop")

Usage is something like:

lmod <- mincLm(files ~ covs, data = frame)
randomization_dist <- mincRandomize(lmod)
randomization_dist_tfce <- mincTFCE(lmod)

thresholds(randomization_dist)
thresholds(randomization_dist_tfce)

Both results objects store the standard object the mincLm or mincLm after tfce, the randomization distribution of maximal statistics, and the args used to call the function. Documentation for these functions can be found with the usual ?

This will take a long time without parallelization - to parallelize locally pass parallel = c("local", <ncores>)

If you have a cluster set up you can run parallel = c("", <a big number>)

[cfhammill]

Whoops just re-read your question - currently we don't have TFCE implemented for vertex data, but I could trivially add randomization for vertexLm objects. I'll keep you posted

[gdevenyi]

+1 for vertex TFCE!

[cfhammill]

I'll open that as a new issue

[cfhammill]

the bones of vertexTFCE have been written, should have a rough version available for testing early next week.

[cfhammill]

Released