A novel study for MDD detection through task-elicited facial cues

Depression is a common mental disorder worldwide. Individuals with major depressive disorder (MDD) are at increased risk for suicide. Current clinical practice for assessing this psychosomatic state are mainly based on self-report and expert evaluation, which risking a range of subjective biases. We investigate a number of task-elicited facial features from Chinese subjects for MDD detection. Moreover, we collect the data from Kinect, which make us achieve a good detection result with low time and space consumption. Experiments are performed on an age, gender and education level matched clinical dataset of 36 MDD patients and 36 healthy controls (HCs). We can get three points from the experimental results: 1) We have presented a simple and objective means for MDD detection, and the average classification accuracies (female: 71.5%, male: 66.7%) are all much higher than chance level. The best classification accuracies (female: 86.8%, male: 79.4%) are achieved during video watching task. 2) Neutral emotion stimulus is a better choice for data collection than positive and negative valences. 3) Eyebrows and mouth have more contributions than other parts of a face in neutral emotion valence. These findings suggest that detecting MDD from facial indicators is feasible, and we provide effective emotion stimulus and facial features.