Figure 1. Summary of the pattern recognition analyses.

(A) Feature Extraction: the beta images were transformed into an input vector. (B) New pattern recognition framework: we first trained the classifier using data from all but one healthy control subject (by a leave-one-out procedure) to discriminate between brain patterns of activation to emotional (100% sad or 100% happy) from neutral faces. We then tested the classifier using data from the healthy control left out and a gender and age matched depressed patient. Finally, we compared the predictive probabilities between healthy controls and depressed patients.