 |
 |
Hi Tudor, There isn't unfortunately a universally accepted solution to this... One common approach that is valid is to draw ROIs for each participant, as opposed to relying on matching homologies through registration. It is up to you really... All the best, Anderson On 25 April 2018 at 11:59, Tudor Popescu <[log in to unmask]> wrote: > Dear FSL experts, > > I have pre and post-intervention data for a group of patients, and I'd > like to use VBM to look at the post-pre GMV difference. I've computed a 4D > "difference" image with all subjects merged, as suggested here > <https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Randomise_details-11>. > > However, the heterogeneity among these pathological brains made me wonder > whether there'd be any sense to doing a VBM-like analysis *within*-subjects, > since the group-level analysis might mask effects due to e.g. overly-warped > registrations. I had in mind papers that plot a colour-map indexing the *number > of subjects *that show a certain effect at the first-level analysis. However, > I'm not sure in the case of VBM one can actually talk of a first-level > analysis, since there is no statistical significance to compute *within*-subjects, > as one can with time-series data (fMRI). > > Another problem, of course, is that an arbitrary threshold would have to > be defined (in terms of pre-to-post effect size) *within *individual > subjects, to be applied across my 4D "difference" image. Would it make > any sense at all to define such a threshold, one beyond which the > pre-to-post change in GMD within a subject is "non-negligible", or would > that be a complete hand-wave? > > Any thoughts on this would be much appreciated, thanks! > > Tudor >