Autism spectrum disorder (ASD) can be a very difficult condition to diagnose, particularly in young children. Previously, we've seen technology intended to detect it by "reading" kids' faces. A new system, however, also gets children to read the faces of others.
Currently, doctors determine if children have ASD based on interviews with them, or via questionnaires that the young patients complete. Unfortunately, both of these approaches can be quite tiresome for restless toddlers, potentially resulting in inaccurate diagnoses.
With that in mind, scientists at Canada's University of Waterloo have developed what is said to be a less irksome system, in which the direction of children's gaze is monitored as they look at pictures of faces. According to the researchers, ASD kids move their eyes between the facial features in a distinct manner, which the new technology can detect.
In a test of the system, 17 ASD children and 23 "typical development" children (all of them aged four to five years old) were shown a total of 44 photographs of faces on a 17-inch screen. As the test subjects viewed those faces, an infrared eye-tracking system was used to note how their eyes moved between seven key areas of interest (AOIs) on the images – those areas were under the right eye, the right eye itself, under the left eye, the left eye itself, along with the nose, mouth and other parts of the screen.
Among other things, it was found that the ASD children moved their gaze between AOIs more quickly than the other group, plus they spent more time looking at the mouth and less time viewing the eyes. It is now hoped that once the system has been refined, it could be combined with traditional diagnostic procedures.
"Many people have autism, and we need early diagnosis especially in children," says master's student and team member Mehrshad Sadria. "The current approaches to determining if someone has autism are not really child-friendly. Our method allows for the diagnosis to be made more easily and with less possibility of mistakes."
A paper on the research was recently published in the journal Computers in Biology and Medicine.
Source: University of Waterloo