Clinical Psychopharmacology and Neuroscience Papers in Press available online.

A Deep Learning Driven Simulation Analysis of the Emotional Profiles of Depression Based on Facial Expression Dynamics
Taekgyu Lee 1, Seunghwan Baek 2, Jongseo Lee 1, Eun Su Chung 1, Kyongsik Yun 3,4, Tae-Suk Kim 5, Jihoon Oh 5,*
1College of Medicine, The Catholic University of Korea, Seoul, Korea, 2UniAI Corporation , 3Computation and Neural Systems, California Institute of Technology, Pasadena, CA 91125, USA, 4Bio-Inspired Technologies and Systems, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA, 5Department of Psychiatry, Seoul St. Mary’s Hospital, The Catholic University of Korea, College of Medicine, Seoul, Korea
Diagnosis and assessment of depression rely on scoring systems based on questionnaires, either self-reported by patients or administered by clinicians, and observation of patient facial expressions during the interviews plays a crucial role in making impressions in clinical settings. Deep learning driven approaches can assist clinicians in the course of diagnosis of depression by recognizing subtle facial expressions and emotions in depression patients.
17 simulated patients who acted as depressed patients participated in this study. A trained psychiatrist structurally interviewed each participant with moderate depression in accordance with a prepared scenario and without depressive features. Interviews were video-recorded, and a facial emotion recognition algorithm was used to classify emotions of each frame.
Among seven emotions (anger, disgust, fear, happiness, neutral, sadness, and surprise), sadness was expressed in a higher proportion on average in the depression-simulated group compared to the normal group. Neutral and fear were expressed in higher proportions on average in the normal group compared to the normal group. The overall distribution of emotions between the two groups was significantly different (p < 0.001). Variance in emotion was significantly less in the depression-simulated group (p < 0.05).
This study suggests a novel and practical approach to understand the emotional expression of depression patients based on deep learning techniques. Further research would allow us to obtain more perspectives on the emotional profiles of clinical patients, potentially providing helpful insights in making diagnosis of depression patients.
Accepted Manuscript [Submitted on 2023-01-25, Accepted on 2023-04-10]