Puzzling patterns Exploring the potential of diagnostic and prognostic neuroimaging in mental disorders

Mental disorders are highly prevalent and impose a significant burden on the individual and on society. Their underlying neurobiology and treatment mechanisms remain poorly understood, although this could improve the clinical care of these disorders. This thesis explores whether neuroimaging (examining brain structure and function in-vivo), large study samples, and multivariate methods can aid in the understanding of the diagnosis and prognosis of mental disorders and their treatment. These applications include using neuroimaging for modeling of early detection and progression of Alzheimer’s disease, assessing treatment effects of electroconvulsive therapy on brain structure and function in depression, modeling the effects of deep brain stimulation on brain reward system function, the prediction of cognitive behavioral therapy in OCD, and the diagnosis of psychiatric disorders with deep learning. Across these studies, this thesis shows that neuroimaging may hold potential for understanding disease progression and treatment effects. However, many findings were inconsistent, and not all studies showed a benefit of using neuroimaging for diagnostic or prognostic purposes. This thesis highlights that the subjective nature of psychiatric disorders, clinical heterogeneity, methodological limitations, replication failures, and variability in study designs present large challenges that currently limit the use of neuroimaging in the clinical care of mental disorders, even with multivariate methods. However, they could be overcome in the future with the use of more stringent study designs, collaborative efforts, unsupervised learning models, and different approaches to brain function.