Skip to Main Content

Children with autism tend to be diagnosed around age 4, after a child begins to socialize and speak. But the earlier a child is diagnosed, the better. Early-intervention speech and behavioral therapy programs have shown promise at reducing symptoms. Now, new research shows such a diagnosis could be predicted as early as one year old — based on scans of infants’ brains.

Still, the study’s findings need to be repeated with a larger sample size before they could be used in a clinical setting, the researchers noted.

A team of researchers in the United States and Canada enrolled 106 infants in the study, some of whose siblings had been diagnosed with autism. They performed MRI scans when the children were six months, one year, and two years old.


They then noted which children had been diagnosed with autism based on criteria in the Diagnostic and Statistical Manual of Mental Disorders, and designed a machine-learning algorithm to discern the differences in their brains.

Ultimately, this algorithm was pretty good at predicting from the six- and twelve-month brain scans of the same group of children if the child would be diagnosed with autism. The key differences were in how a child’s brain grew in the first year of life. Enlarged brain volume has previously been observed in children with autism. 


If the algorithm predicted that the child would be diagnosed with autism, it was correct just over 80 percent of the time. The research was published in Nature on Wednesday.

Diagnosing autism very early in a child’s life might mean better interventions and outcomes. “It’s been a continual goal of Autism Speaks and the autism community to drive the age of diagnosis to be as early as possible,” said Mathew Pletcher, Autism Speaks’ interim chief science officer. “Early diagnosis in autism does make a difference.” (Autism Speaks partially funded the study, along with the NIH and the Simons Foundation.)

On average, children aren’t diagnosed with autism until they are four years old — once their brain has begun to expand, and once they begin behaving differently than neurotypical children — though some are diagnosed as early as their second birthday, Pletcher noted.

“Our findings are pre-symptomatic, certainly pre-consolidation of the diagnosis,” said Dr. Joseph Piven, who leads the eight-center Infant Brain Imaging Study Network, which did the research. “That’s a giant step in the field.”

Screening scaled up

Researchers have known for a while that children with autism tend to have bigger brains, but they hadn’t figured out when the brain got bigger or how it changes in early childhood. The brains of the children with autism included in the study were actually more convoluted than those of their neurotypical peers — that is, their brains had more and deeper folds. That change happened before the child’s first birthday.

“There are very, very few studies done in children this young,” said University of California at Davis researcher David Amaral, who wasn’t part of the study. This is the first time he’s seen a machine-learning algorithm used to analyze infants’ brain scans. (Other studies have used similar techniques to study scans taken of adults’ or older children’s brains.)

However, Amaral points out, “When you are able to look at large numbers of kids, what you find is that the big brains are typical of 15 percent of boys with autism.” That indicates that the algorithm might not be able to predict every diagnosis. Piven noted that, in his opinion, the change in the brain size over time — not necessarily the absolute size — was the predictive factor.

Piven and others agree that “it would not be practical” to routinely scan infants with such a technology if the screening tool moves to the clinic. But children in groups with higher rates of autism — like those who have affected siblings or who carry genes linked to autism — might benefit the most from it.

As an NIH-funded Autism Center of Excellence, the researchers’ data and tools are open-source and will eventually be submitted to the NIH’s National Database for Autism Research. But replicating the study may be financially challenging — Pletcher called it “the biggest obstacle”; Piven agreed. “It’s a very expensive operation,” he said.

For research like this to become clinically relevant, scientists will likely need to arrange larger collaborations with larger numbers of subjects. Amaral cited two examples: The MSSNG whole genome sequencing collaboration, which is aiming for 10,000 participants, and the SPARK project, which has over 46,000 registered participants according to data available on the project’s website.

The group will be publishing several more papers in the future, Piven said. “Our data set is just starting to come to fruition. Because in a developmental disorder that you’re studying longitudinally, you have to wait for kids to grow up,” he said.

“We’re just getting started.”

Comments are closed.