One, a study with more than 32 people. Two, replications with equally strong observations.
I've read a lot of research on fMRIs and mental disorders. Something like this pops up every other day. Not a single one has really study the test of time ion terms of diagnostic utility.