Hello SPM community,
I'm writing because I’m experiencing some strange issue within SPM8 and I've tried combing the archives - but haven't found anything related to this specific issue (or do not know the correct search terms). Essentially, a subset of my data is getting squashed at some point in the first-level analysis process - resulting in blank beta maps (zeros shaped like a brain embedded in NaNs) for a subset of runs (1 or 2 out of 3) in a subset of subjects (total n=25). While this does seem to be replicable when I rerun the analysis - it seems to choose a different set of runs to squash when I run different models (e.g. using different #’s of indicator or modulator variables), so it must in some way be interacting with the design matrix (an event-related observational learning paradigm). Having dug through the SPM structure I believe the problem is occurring during the whitening step, though I'm not positive. In particular - I think the hyperparameters related to the sphericity (i.e. SPM.xVi.h) of my data might be incorrectly estimated as they are largely different in runs that get squashed versus those that do not. To help with diagnosis, I’m including links to 2 SPM.mat files. 001SPM.mat is the first level analysis for a subject that seemed to work (e.g. there were no visible problems). 003SPM.mat is a subject for whom the second and third runs were squashed. This is readily visible in the SPM.xX.xKXs.X fields among others.
Any thoughts or similar experiences that anyone has had would be extremely helpful.
Thanks in advance!
SPM.mat links:
https://www.dropbox.com/s/8dfz8z9dg2hijt3/001SPM.mat
https://www.dropbox.com/s/rjleojrv59zlmbz/003SPM.mat
Best,
Damian Stanley
California Institute of Technology
|