Most functional neuroimaging studies have very small sample sizes, and don't have the power for us to pay much heed to an association, let alone make causal inferences, most findings are not consistently replicated, no findings are diagnosis specific, and the methodologies for both structural and functional neuroimaging studies are rather arbitrary and try and make something out of small differences the significance of which is unknown. In psychiatric genetic studies, the associations tend to be transdiagnostic and thus do not support the existence of our current mental disorders as diseases, and with exception of the older candidate gene studies and some of the GWAS studies, most studies fail to explore functionality of particularly SNPs or CNVs associated with psychopathology again making it hard to make sense of the biological plausibility of these associations.
Although the NIMH is moving away from DSM-based diagnoses to RDoC, it is laughable that for long people have assumed that diagnoses like 'schizophrenia', 'bipolar disorder' and 'depression' represent 'diseases' that are natural kinds and for which biomarkers can be found for. Psychiatric diagnoses require functional impairment or significant distress and thus are socially constructed and determined more by social and psychological factors that have nothing to do with underlying biology of the phenomena. Whether someone is significantly impaired or distressed by, say auditory hallucinations, which is what we require in order to say this is part of an illness, has little to do with neurobiology of auditory hallucinations themselves.
I am not disputing that psychiatry will not at some point progress to a phase where there are identifiable 'diseases' (though I am not sure how necessary or desirable this actually is), but I am disputing that our current concepts of mental disorder equate in any way, shape or form, to 'diseases' - of the brain or otherwise.