AI Predicts Autism Based on Infant Brain Scans
Brain scans, analyzed using a type of artificial intelligence, can reveal whether 6-month-old babies are likely to develop autism, a new study shows.
The study examined 59 infants who were at high risk of developing autism; that is, each had an older sibling with autism. The artificial intelligence predicted with 100 percent accuracy that 48 infants would not develop autism. In addition, of the 11 infants who did develop the disorder by the time they were 2 years old, the system correctly predicted nine of the cases.
"It was extremely accurate," Robert Emerson, the lead author on the study and a former cognitive neuroscience postdoctoral fellow at the University of North Carolina (UNC), told Live Science. [5 Things That May Cause Autism]
Studies show that 20 percent of babies who have older siblings with autism will develop the disorder; among babies in the general population, 1.5 percent develop autism, Emerson told Live Science.
The results of the new research could lead to new diagnostic tools that identify autism before symptoms occur, giving clinicians the opportunity to intervene early, the researchers said.
"The idea is that we can be more effective if we can get to these kids before they develop autism, perhaps ameliorating or preventing it," Dr. Joseph Piven, a professor of psychiatry at the UNC School of Medicine and director of the Carolina Institute for Developmental Disabilities, told Live Science. The researchers published their results today (June 7) in the journal Science Translational Medicine.
The development of autism
Autism spectrum disorder, a brain-based disorder characterized by a wide range of social-communication challenges and repetitive behaviors, affects about one out of every 68 children in the United States. Behavioral symptoms typically begin to appear in children at around age 2.
Emerson and his colleagues demonstrated that they could identify biomarkers for the disorder before the symptoms occur.
As part of their study, the researchers used MRI scanners to image the brains of the infants while they slept. During the scans, the researchers recorded the neural activity of 230 different regions in the brain, looking particularly at whether or not pairs of these region — referred to as functional connections — were synchronized with each other, and if so, to what extent.
In total, the researchers measured 26,335 functional connections crucial for cognition, memory and behavior.
When the children reached age 2, they came back in for a behavior assessment. The researchers looked at the children's social interactions, communication, motor development and tendency to perform repetitive actions, and determined whether each child had autism. [11 Facts Every Parent Should Know About Their Baby's Brain]
With all of the data in hand, the researchers set out to first train their machine learning program, and then use it to run predictions. They wanted to see how accurately it could predict which infants had developed autism, using only the functional connections data from when the children were 6 months old. In other words, although the researchers knew which babies had developed into toddlers with autism and which had not, the machine learning program did not.
Machine learning is a kind of artificial intelligence system that gets smarter based on the data it processes. In this case, the program was learning to spot differences between the functional connections imaged in the MRI data collected at 6 months old that correlate with cognition, memory and behavior and the details from the behavioral assessments collected at 24 months.
As the program did this, it separated the children into two groups — those with autism and those without the condition. Once it was trained, it could make predictions. [7 Baby Myths Debunked]
But during the training process, the researchers didn't use the data from all 59 kids. Instead, they fed data in from 58 of the 59 infants to train the model, and then to get the prediction, they inputed the data from the one infant they had left out. They repeated this for all 59 children.
"Each child was predicted separately based on a model from the other children in the group," Emerson said.
In the end, the machine learning program was correct in 82 percent of the cases in which the children did develop autism.
An amazing group of families
Piven said the team published a study earlier in the year that also showed an impressive prediction rate, but that study required two MRI scans, one at 6 months of age and one at 1 year. Moving the needle to an earlier age is a big advance, Piven said.
The research team was amazed by the parents who participated in this and other studies over the years, given all that was required, he said.
"This is an extraordinary group of people," Piven said. "Not only do they have an older child with autism, but they bring their babies, often multiple times and from very far away, to one of our four clinical sites around the country."
"They're very committed," Emerson said.
The researchers said they hope their contributions will lead to more effective interventions for children on the brink of developing autism.
Originally published on Live Science.
Live Science newsletter
Stay up to date on the latest science news by signing up for our Essentials newsletter.
By Sascha Pare
By Ben Turner
By Sascha Pare
By Harry Baker
By Ben Turner