Dynamic Causal Modelling Applied to Functional MRI of Depression: An Objective Diagnosis

The diagnosis of depression is based on a subjective interpretation of reported symptoms rather than an objective test based on a measurement. The limitations of the current diagnostic practice of neuropsychiatric disorders have been acknowledged and it is time to shift the paradigm toward a diagnosis based on an objective test following a measurement. In this study, effective connectivity between pairs of resting-state networks obtained from independent component analysis of functional magnetic resonance imaging is estimated using Friston's Dynamic Causal Modelling. Chi-Square feature selection and classification algorithms are used to classify between depressed and control individuals on a case-control dataset consisting of 51 depressed individuals and 21 healthy controls. This resulted in 77 % leave-one-out cross-validation accuracy using eight effective connections. The most discriminative effective connections are interpreted from a clinical view. Besides clinical interpretation, the application of Friston's biologically-inspired model for es-timating effective connectivity is also compared to data-driven methods of estimating causality on the same dataset. Aberrations in the effective connections within the default mode network and between the default mode network and the dorsal attention network are identified using the proposed methodology. This can be linked to the symptoms of depression. If the identified effective connections prove reproducible on other datasets, this is a stride toward a biomarker for depression.