I’m using FMRIPREP for pre-processing and SPM for GLM estimation. I estimated a GLM on the surface, using the fsnative outputs from FMRIPREP. The beta weight maps output by the GLM contain many NANs, mostly in areas where there is signal-dropout, but also in smaller chunks throughout the cortex, which is problematic.
I’ve checked the preprocessed fsnative .gii files, and they contain no NANs. Where there is signal dropout, the raw values are 0 or near 0. In the beta map, it seems these same areas with near 0 values correspond to the areas where the beta map is nan.
It seems to me that the beta values should converge to 0 where there is signal dropout, not NAN. Any ideas why I might be getting NAN beta values in the GLM?