Despite all the headway that science has made in understanding autism in recent years, knowing which children will one day develop autism is still almost impossible to predict. Children diagnosed with autism appear to behave normally until around two, and until then there is often no indication that anything is wrong.
But by scanning the brains of babies whose siblings have autism and then running the data from those scans through a machine learning algorithm, researchers say they may have come up with a method for accurately predicting which children will wind up diagnosed with autism at as young as six months.
For autism researchers, this feat has long been elusive. Diagnosing autism spectrum disorder before children develop symptoms could allow families to begin treatments like behavioural therapy earlier in hopes of making it more effective, as well as allowing researchers to test potential treatments, enabling them to more accurately judge whether these treatments actually work.
In a paper out Wednesday in Science Translational Medicine, researchers from the University of North Carolina at Chapel Hill and Washington University School of Medicine scanned the brains of 59 high-risk, six-month-old infants to examine how different regions of the brain connect and interact. At age two, after 11 of those infants had been diagnosed with autism, they scanned their brains again. After that, the researchers turned to artificial intelligence, using an algorithm that trained itself to identify patterns in brain connectivity that separated those six-month-olds who developed autism and those who did not. Using deep learning, they were then able to develop a model capable of predicting which six-month-olds would eventually develop autism.
Using this method, researchers were able to accurately predict nine of the 11 infants who would wind up with an autism diagnosis. And it did not incorrectly predict any of the children who were not autistic.
“Our treatments of autism today have a modest impact at best,” said Joseph Piven, a psychiatrist at UNC Chapel Hill and author of the study, told Gizmodo. “People with autism continue to have challenges throughout their life. But there’s general consensus in the field that diagnosing earlier means better results.”
Estimates suggest that about one out of every 68 children in the US has autism. The Australian Bureau of Statistics found that around one in 150 Australians had autism in 2015. Still, there are no good biomarkers to predict who is most at risk for developing it. Some rare genetic mutations are linked to autism, but most cannot easily be linked to genetic risk factors. While some findings have indicated that brain-related changes occur in children with autism before any behavioural symptoms emerge, those changes have been difficult to identify.
The study was a follow-up to one published earlier this year that looked at whether brain growth could be a biomarker for autism, since children with autism tend to have larger brains than developmentally normal children. In that study, MRI scans revealed that the volume of the brains of infants with autism grew faster between 12 and 24 months. Based on those scans, an algorithm was able to detect which children between six and 12 months would develop autism about 80 per cent of the time, though it also identified a few false positives.
By looking instead at connectivity, the new study shows a method of prediction that’s more accurate and identifies children at a younger age. In total, they found 974 functional connections that were associated with autism-related behaviours.
“It’s a data driven approach,” said Piven. “We didn’t start with a particular hypothesis.”
Piven said they hope to reproduce the study, as well as expand it to not just predict whether a child might wind up with autism, but how severe it will be and what sorts of behaviours they will exhibit. Autism is a spectrum disorder ranging from mild symptoms to ones that severely inhibit a person’s life, so this would make the tool much more useful and potentially also make treatment more impactful.
The study is only an early indicator of a good predictive measure. It will have to be reproduced before it’s ready for clinical use. And the test in its current form would unlikely be used in the general population, but rather as a measure to be taken after an infant has already been identified as high risk. One in five siblings of children with autism, for example, eventually develops autism. Developing other screening techniques for high-risk infants would make such a test more useful.
“I would look at this study as a proof of principle,” he said. “Our intention is to provide early detection intervention just like we now do for Alzheimer’s and Parkinson’s.”