A team of University of North Carolina researchers has developed a way to potentially predict autism in children as young as 6 months old using artificial intelligence.
Using decades-old research that found a correlation between infant brain volume to autism diagnosis, researchers developed a high-tech method for diagnosing autism spectrum disorder in young children.
The algorithm compares scans of babies’ brains and can find the physical clues that are key in predicting whether a child at high risk of autism would be diagnosed with the disorder within 24 months.
The algorithm reveals increases in the infant’s brain surface area when the baby is less than a year old, which often precedes an overgrowth in brain volume that previous research linked to autism.
Never miss a local story.
The UNC team tested 106 infants with an older sibling who had already been diagnosed with autism and 42 infants with no family history of autism. Using the infants’ brain scans at 6, 12 and 24 months, the algorithm correctly predicted the infants’ eventual diagnosis among the higher risk group with 81 percent accuracy.
In comparison, behavioral questionnaires which are used to determine early autism diagnoses around 12 months are accurate just 50 percent of the time.
The algorithm found that brain overgrowth was a telling factor in most diagnoses according to the compared scans. The greater the overgrowth, the more severe the child’s autism tended to be.
But Hazlett said the results of the UNC team’s work needs to be replicated before the public could see a computer-detected biomarker for autism. That will take time since it’s difficult and expensive to get brain scans of young children. And those tests won’t necessarily be appropriate for all children.
“It’s not something I can imagine being clinically useful for every baby being born,” Hazlett said. But if genetic testing or other markers show a risk for autism, the brain scans and algorithm could help identify the brain changes that put them at a greater risk.
The algorithm was created in partnership with UNC computer scientists and the College of Charleston.
Abbie Bennett: 919-836-5768; @AbbieRBennett