We are working with a group of patients that have a focal neurdegenerative dementia that targets the left hemisphere language network. Using FreeSurfer were are able to locate a significant cortical thinning predominantly located within the left hemisphere language regions. We are interested in correlating the degree of thinning, caused by disease, with neuropsychological performance in the patient group (n=20).
From a statistical standpoint its unclear which approach would be best:
1) Should we limit our correlation analysis to the regions that demonstrate significant thinning? If so, what threshold or FDR correction is most appropriate? For example, we can create group thickness maps (Controls vs Patients) using qdec and make our FDR correction extremely stringent at 0.0001 and then create ROI labels based on the thinning that survives (see attached picture). The mean thickness value from each ROI can then be used in a correlation analysis. Does this make sense from a statistical standpoint? or we should we use a less stringent FDR and have a larger region of atrophy? 2) or Should examine correlations across each vertex of the hemisphere using qdec? If so, what threshold or FDR correction is most appropriate?
Thanks in advance for your help. Emily