Automated facial expressions analysis in schizophrenia: A continuous dynamic approach

Abstract

Facial expressions play a major role in psychiatric diagnosis, monitoring and treatment adjustment. We recorded 34 schizophrenia patients and matched controls during a clinical interview, and extracted the activity level of 23 facial Action Units (AUs), using 3D structured light cameras and dedicated software. By defining dynamic and intensity AUs activation characteristic features, we found evidence for blunted affect and reduced positive emotional expressions in patients. Further, we designed learning algorithms which achieved up to 85% correct schizophrenia classification rate, and significant correlation with negative symptoms severity. Our results emphasize the clinical importance of facial dynamics, and illustrate the possible advantages of employing affective computing tools in clinical settings.