Emerging research suggests that a new software program called “SimSensei” may be able to use facial cues and body language to detect depression using Microsoft's Xbox Kinect. Additional pilot research from Australia suggests that computers can use non-verbal signals to distinguish between people with and without depression diagnosis at 90% accuracy.
This is currently an active area of research in several academic centers around the world, and computers have long been effective in recognizing human emotion. This work has significant implications for behavioral health as we begin to think about how to more effectively leverage technology solutions to provide better care.
Better than a coin flip
The clinical challenge of psychiatric diagnosis is an especially important problem to solve. We know that our most highly trained diagnosticians, psychiatrists, get the diagnosis correct about 50% of the time in community settings. Of course, many consumers in community mental health settings are never even evaluated by a psychiatrist, so that is perhaps an optimistic estimate of standard practice. The diagnosis, of course, is the lynchpin upon which the treatment plan is developed, and an incorrect diagnosis is likely to lead to poorly matched treatment, delayed recovery, and wasted resources. Improved diagnosis is perhaps the most important thing we could do to improve outcomes and reduce unnecessary costs.
So, the prospect of a computerized system achieving 90% accuracy would appear to offer great value to behavioral healthcare. Consider for a moment the current access challenges common to behavioral healthcare. It is not uncommon for people with serious psychiatric symptoms to wait weeks to receive an initial assessment, much less talk to a psychiatrist. As of February 2013, Microsoft has sold 76 million Xbox 360 systems and more than 24 million Kinects worldwide. Could this be an avenue for universal mental health screening in your living room?
While early results are promising, there are reasons to be skeptical. Although the initial pilot studies were able to sort people into depressed and non-depressed groups with 90% accuracy (under ideal conditions), that impressive feat does not reflect the immense challenges faced by diagnosticians in the real world. In clinical environments, patients most frequently present with complex comorbid conditions, which are a challenge to sort out even under ideal circumstances. Effective differential diagnosis is an elite skill, and it is hard to imagine how such a system would even diagnose common diagnostic pairs, say co-occurring bipolar disorder and substance dependence.
Reinventing depression care