Donate to Science & Enterprise

S&E on Mastodon

S&E on LinkedIn

S&E on Flipboard

Please share Science & Enterprise

Diagnostic Software Developed to Analyze Video for Autism

Baby portrait

(44444 U.A.E./Flickr)

22 May 2014. Computer scientists and medical researchers at Duke University in North Carolina developed software that uses computer vision to analyze video of an infant’s behavior for signs of autism spectrum disorder. The team from the lab of Duke computer engineering professor Guillermo Sapiro — with colleagues from Duke, University of Minnesota, and University of Campinas in Brazil — published an advance manuscript of their research online last week in the journal Autism Research and Treatment.

The researchers are seeking a simple early-warning and screening system to spot signs of autism in infants. Current methods for analyzing infant behavior for autism and related disorders require frame-by-frame analysis of video by trained experts, which can be a burden on clinics and not suitable for large-sample research.

The approach devised by Duke computer engineering graduate student Jordan Hashemi simplifies the process by harnessing computer algorithms to track, measure, and analyze tests of the child’s behavior. The tests are derived from the Autism Observation Scale for Infants, a standard diagnostic index designed to detect signs of autism among high-risk infants, i.e. younger siblings of children with autism spectrum disorder.

The tests gauge the reactions of infants to toys shaken to the left and right of their field of vision, tracking delayed reactions of children to stimuli in their field of vision, and checking for eye contact, a sign of engagement with a play partner. The tests currently call for clinicians to not only track reactions, but also measure the time needed for children to react to the test stimuli.

The software devised by Hashemi and colleagues tracks and measures indicators of development in video of the tests, such as changes in the infant’s gaze or motor behaviors that provide signs of autism. Algorithms in the software assess the risk of possible autism based on calculations of those test measurements.

In their research paper, the team tested the ability of the software to identify signs of autism, and compared the results with assessments made by autism experts, as well as medical clinicians and students interested in the subject, but not considered experts in the field. The results show the software does as good a job as autism experts at identifying indicators of autism. The findings also show the software does a better job at finding signs of autism than non-experts.

Hashemi notes the aim of the software is to extend the diagnostic abilities of autism experts, not replace them with computers. “We want to give people tools they don’t currently have,” says Hashemi in a university statement, “because research has shown that early intervention can greatly impact the severity of the symptoms common in autism spectrum disorders.”

The Duke team aims to develop a tablet app that can record and analyze an infant’s behavior without the need for a parent or caregiver to administer diagnostic tests. “We’re currently working with autism experts at Duke Medicine to determine what sorts of easy tests could be used on just a computer or tablet screen to spot any potential concerns,” says Sapiro. “The goal is to mimic the same sorts of social interactions that the tests with the toys and balls measure, but without the toys and balls.”

Read more:

*     *     *

Comments are closed.