Hi Steve,
Thanks for the great response. Auto-classification of first-level ICA components sounds awesome; would love to test (as no doubt others would as well) as soon as is available.
However, given the tools that we have now (e.g. confound regression), just to drill down to some details, I have two follow up q's:
1) Does spike regression (using something like fsl_motion_outliers) violate any assumptions of tc-GICA? From Mark's earlier posts it does not seem like it was designed for that, so I am not sure if any untoward effects might occur with it's use in another context.
2) What utility would you reccomend for confound regression prior to ICA? It is not obvious to me whether FEAT/FILM,, fsl_glm, or fsl_regfilt would be the right choice. Would this regression benefit from the autocorrelation correction FILM provides? Is there a real difference in how the regression is run between fsl_glm and fsl_regfilt? (I can't seem find anything on the list comparing the two, but I apologize if I've missed it).
Thanks so much for the advice and expertise.
Best,
ted
|