Multimodal assessment of depression from behavioral signals

Reliable, valid, and efficient assessment of depression is critical to identify individuals in need of treatment and to gauge treatment response. Current methods of assessment are limited to subjective measures of patient self-report and clinical interview. They fail to take into account observable measures of behavior that could better inform detection of the occurrence and severity of depression. Recent advances in computer vision, signal processing, and machine learning have potential to meet the need for improved depression screening, diagnosis, and ascertainment of severity (i.e., assessment). This chapter reviews these advances. We describe multimodal measures of behavior and physiology, how these measures can be processed to extract features sensitive to depression, and how classification or prediction may be used to provide automatic assessment of depression occurrence and severity.