Improving therapy outcome prediction in major depression using multimodal functional neuroimaging: A pilot study

Abstract

Mounting evidence emphasizes the usefulness of imaging biomarkers for predicting therapy outcome in major depressive disorder (MDD), in particular building on functional imaging studies of task-based responses to emotional face stimuli and resting state-related connectivity patterns. To explore the possibility that prediction accuracy even in small patient samples would significantly gain from integrating data from different imaging modalities, we acquired functional neuroimaging data both at-rest and during exposure to emotional faces from 21 MDD patients before and 7 weeks after treatment-as-usual, as well as from 20 age- and gender-matched control participants assessed at similar intervals. As expected, MDD patients showed disturbed pre-treatment responses to emotional faces, including left amygdala hyperactivation. Therapeutic outcome correlated with pre-treatment activation, with subgenual cingulate response to emotional faces yielding best results (r values ranging from 0.4 to 0.66). A support vector machine classifier trained on task-based or resting-state data predicted responder status, with the right dorsolateral prefrontal cortex connectivity pattern yielding best accuracy (88.9%). Crucially, combining task-based with resting-state data increased prediction accuracy by 6.5–7.7 percentage points on average. From this pilot study, we conclude that multimodal functional imaging has the potential of improving therapy outcome prediction even in small MDD sample sizes, resulting in about one additional correct classification every 15 patients. The present results inform future studies which are needed to consolidate imaging approaches as a means of establishing precision medicine in psychiatry.