MIT Algorithm Reads Your Face to Detect Pain

first_img In one of my favorite Scrubs scenes, J.D. and Elliot refer to an “archaic” chart—illustrated expressions in varying degrees of pain—to gauge a patient’s discomfort.Mr. Peele, whose face looks like he just bit into a lemon, is a seven. Todd, hanging four stories from his banana hammock, is a 10.Sacred Heart Hospital’s rating scale, however, isn’t as antiquated as the fictional doctors believe. According to Medscape, single-dimensional scales, like the one held up by Dr. Reed, are still used to assess acute pain.More complex issues like chronic pain require multi-dimensional scales, which measure intensity, nature, and location of suffering, as well as the impact on a patient’s activity or mood.But these tools often require patients to have good verbal skills and sustained concentration—not always possible when your body is throbbing, convulsing, or cramping.Researchers at the Massachusetts Institute of Technology are working on a modern solution that takes automatic pain estimation to a new level.In a paper published last month, MIT scientists describe a two-stage hierarchical learning algorithm called DeepFaceLIFT (Learning Important Features).Trained on videos of people with shoulder pain grimacing through range-of-motion exercises, the neural network has learned to detect subtleties in facial micro-expressions.When paired with self-reported pain scores, the algorithm can better estimate the user’s degree of discomfort. Age, sex, and skin complexion personalize the technique for more accurate results.MIT researchers Dianbo Liu, Fengjiao Peng, Andrew Shea, Ognjen Rudovic, and Rosalind Picard may not sound like heroes to you. But the team has developed a system that can help doctors and nurses distinguish between patients in need and those trying to scam prescription painkillers.For now, however, physicians must remain reliant on the face pain chart, while sadistic scientists continue schooling their algorithm in what it means to feel pain.“We plan to derive more advanced statistics that could potentially capture additional information and improve estimates of subjective pain,” according to MIT’s study. “Finally, we hope that these findings will advance applications of pain estimation in clinical settings.”Let us know what you like about Geek by taking our survey. Stay on target MIT’s AI Knitting System Designs, Creates Woven GarmentsMIT, IBM Train AI to Create and Edit Fake Images last_img read more