When experts look back over early home movies of babies later diagnosed with autism, they can see the early behavioral markers of the disorder. The average age of diagnosis, however, is 5 years old. The availability and cost of trained clinicians limits not just early awareness, but also research into autism on a large scale. At Duke University, researchers from different disciplines are using computer vision algorithms to make early diagnosis more likely, and even intend to create an app for parents to use at home.
The video at the bottom of the page demonstrates the clinician's usual role in a session with an infant thought to be at risk for autism: offering toys, engaging the infant with a smile, rolling a ball. However, rather than the clinician having to simultaneously and mentally count off the baby’s reaction time to stimuli, or later review a video, a computer algorithm tracks numerous data points in the interaction: each of the baby’s eyes, the toys, and the incline of the baby’s head.
NEW ATLAS NEEDS YOUR SUPPORT
Upgrade to a Plus subscription today, and read the site without ads.
It's just US$19 a year.UPGRADE NOW
When the algorithm scores the child as either pass, delay, or stuck, it does so with the same success rate as the trained observer, and better than students in training or medical professionals who aren’t experts in autism.
Additionally, the algorithm provides a consistency that allows researchers to later compare children who were seen by different clinicians, and gives them a tool to increase the size of their data set when testing new treatments or therapies.
The research is part of an initiative at Duke aimed at putting computer processing tools into fields that need them to manage large data sets. The joint research team is interested in next creating an app that parents could use at home as a precursor before seeing a professional. On the surface, the app would seem like any other mobile game for children, but would use the algorithms developed earlier to assess the child using the device’s camera.
Below is the video, with the computer tracking overlaid on top of the footage of the clinician, child, and parent. While the footage is downsampled and blurry, this was done for anonymity. The algorithms processed the original footage, taken with a GoPro camera.
The research was originally published in Autism Research and Treatment.
Source: Duke University