Context: Although psychiatric disorders are, to date, diagnosed on the basis of behavioral symptoms and course of illness, the interest in neurobiological markers of psychiatric disorders has grown substantially in recent years. However, current classification approaches are mainly based on data from a single biomarker, making it difficult to predict disorders characterized by complex patterns of symptoms.
Objective: To integrate neuroimaging data associated with multiple symptom-related neural processes and demonstrate their utility in the context of depression by deriving a predictive model of brain activation.
Design: Two groups of participants underwent functional magnetic resonance imaging during 3 tasks probing neural processes relevant to depression.
Setting: Participants were recruited from the local population by use of advertisements; participants with depression were inpatients from the Department of Psychiatry, Psychosomatics, and Psychotherapy at the University of Wuerzburg, Wuerzburg, Germany.
Participants: We matched a sample of 30 medicated, unselected patients with depression by age, sex, smoking status, and handedness with 30 healthy volunteers.
Main outcome measure: Accuracy of single-subject classification based on whole-brain patterns of neural responses from all 3 tasks.
Results: Integrating data associated with emotional and affective processing substantially increases classification accuracy compared with single classifiers. The predictive model identifies a combination of neural responses to neutral faces, large rewards, and safety cues as nonredundant predictors of depression. Regions of the brain associated with overall classification comprise a complex pattern of areas involved in emotional processing and the analysis of stimulus features.
Conclusions: Our method of integrating neuroimaging data associated with multiple, symptom-related neural processes can provide a highly accurate algorithm for classification. The integrated biomarker model shows that data associated with both emotional and reward processing are essential for a highly accurate classification of depression. In the future, large-scale studies will need to be conducted to determine the practical applicability of our algorithm as a biomarker-based diagnostic aid.