Deep Learning to extract biomarkers for the diagnosis and personalized treatment of neuropsychiatric disorders

Improving the diagnosis and treatment strategies for individuals with neuropsychiatric disorders has widespread impact on the individual and our society. Increased quality-of-life of the patient and their relatives, lower economic burden on healthcare and a better understanding of psychiatric disorders are few of the benefits. Currently, diagnosis of psychiatric disorders is based on a clinical interview. This is known to lead to variability in diagnoses given by different clinicians. Moreover, personalized clinical prognosis and the prediction of treatment outcome is currently impossible due to the absence of accurate indicators, and treatment is therefore performed on a trial-and-error basis.

In this interdisciplinary proposal, we aim to investigate the feasibility of using advanced neural networks on neuroimaging data to extract neuroimaging biomarkers that (I) diagnose individuals with various psychiatric disorders and (II) predict whether a patient will respond to treatment. We are at a time in history where large datasets (>10000), advanced algorithms and cheap computational power have given us the opportunity to tackle a problem of this magnitude and importance. Large public datasets will be used to develop the algorithm and a biomarker for diagnosing depression. We will subsequently use those methods to develop a biomarker for the response to electroconvulsive therapy (ECT) in an international cohort of 300 severely depressed patients. The prediction of ECT response will provide proof-of-concept for our approach, which can readily be extended to all other psychiatric disorders and thus pave the way for personalized treatment in psychiatry.