A University of Minnesota researcher and a team of colleagues have developed a computer vision system that automatically analyzes toddlers' movements to detect signs of autism early.
Autism is estimated to affect more than two million people in the US and tens of millions worldwide, but diagnosing it is notoriously difficult, particularly in young children. Recently, however, University of Minnesota researcher Jordan Hashemi and a team of colleagues developed a computer vision system that automatically analyzes toddlers’ movements to detect signs of autism early. The new system starts with video footage of the child in question in an ordinary setting. Focusing in particular on four types of behaviors that can indicate autism — including maintaining asymmetrical body positions and delayed tracking of objects within the child’s field of vision — it assesses the position of the child’s head as well as the position of the arms, torso and legs, building a 2D stick skeleton of the child along the way. Having already shown encouraging results, the technology holds considerable promise for automating at least part of what is otherwise a labor-intensive and time-consuming diagnosis process. Early intervention can greatly improve the outlook for autistic people, Hashemi et al. note, and this new system could be a key step in that direction. Medically minded entrepreneurs: one to get involved in?