Pattern Classification of Sad Facial Processing: Toward the Development of Neurobiological Markers in Depression

Background: Methods of analysis that examine the pattern of cerebral activity over the whole brain have been used to identify and predict neurocognitive states in healthy individuals. Such methods may be applied to functional neuroimaging data in patient groups to aid in the diagnosis of psychiatric disorders and the prediction of treatment response. We sought to examine the sensitivity and specificity of whole brain pattern classification of implicit processing of sad facial expressions in depression.

Methods: Nineteen medication-free patients with depression and 19 healthy volunteers had been recruited for a functional magnetic resonance imaging (fMRI) study involving serial scans. The fMRI paradigm entailed incidental affective processing of sad facial stimuli with modulation of the intensity of the emotional expression (low, medium, and high intensity). The fMRI data were analyzed at each level of affective intensity with a support vector machine pattern classification method.

Results: The pattern of brain activity during sad facial processing correctly classified up to 84% of patients (sensitivity) and 89% of control subjects (specificity), corresponding to an accuracy of 86% (p < .0001). Classification of patients' clinical response at baseline, prior to the initiation of treatment, showed a trend toward significance.

Conclusions: Significant classification of patients in an acute depressive episode was achieved with whole brain pattern analysis of fMRI data. The prediction of treatment response showed a trend toward significance due to the reduced power of the subsample. Such methods may provide the first steps toward developing neurobiological markers in psychiatry.