Psychomotor cues for depression screening

Depression is a cognitive impairment, which according to the World Health Organisation is the leading cause of disability worldwide. One key trait of depression is psychomotor retardation, which adversely affects both emotional and physical behaviour of an individual. In this paper we perform experiments on the Audio Visual Emotion recognition Challenge 2016 - Depression Classification sub-Challenge (AVEC 2016 - DCC) dataset, with the objective to develop methods for automated screening of depression. We discuss two frameworks for the task of automated depression screening. The first is based on computation of facial movement features, followed by regression to yield a higher level feature which is correlated to depression severity of an individual. The second framework focuses on classification between depressed and non-depressed individuals on the basis of their speech spectra. We show that the regression framework yields a correlation value of 0.55 with depression severity scores, which is better than individual features, thereby demonstrating its efficacy. Meanwhile, the classification framework provides a mean E1 score of 0.906 on the development subset, which is comparable to the state-of-art on the AVEC 2016 - DCC development set i.e. 0.910.