Social signal processing in depression

As social signal processing develops as a field of enquiry and application, there is emerging focus on individual differences in social signaling. My colleagues and I have been particularly interested in social signal processing in depression. Depression has among the highest lifetime prevalence and morbidity of any psychiatric disorder [1]. An understanding of interpersonal mechanisms and accurate, efficient diagnosis and assessment of course and response to treatment are high priorities of national and international efforts. Social signal processing is primed to contribute to these tasks. I review previous efforts to characterize social signal processing in depression, the need for longitudinal designs to control for trait emotion, and recent breakthroughs in multimodal measurement of social signaling in depression. In a clinical sample undergoing treatment for depression [2], my colleagues and I used active appearance models (AAM) [3, 4] and speech recognition and prosodic measures [5, 6] to evaluate change over time between symptom severity and social signaling. I report our latest findings as well as new efforts to produce assessment measures that can be implemented efficiently in both research and clinical settings. Because AAM approaches must be specifically tuned to lighting, camera, and person characteristics prior to use, their general use is limited. Constrained local models (CLM) are a promising alternative [7, 8]. CLM require no training or adaptation for use with new persons, illumination, or cameras. When coupled with registration-invariant registrations, our initial findings suggest that they can achieve accuracy approaching that of AAMs [9]. In summary, I review what is known of social signal processing in depression, recent longitudinal research using AAM and prosodic measures of depression severity and partner effects, and prospects for generic approaches using CLM that lend themselves to use in a wider range of research and clinical applications.