AI Helps Accurately Predict Autism Risk in Infants

The study involved 59 infants at high risk for autism, meaning each had an older sibling diagnosed with the disorder. The AI ​​predicted with 100% accuracy that 48 of the children would not develop autism.

Remarkably, of the 11 children identified with the disorder at age 2, the system correctly predicted nine cases.

"The results are incredibly accurate," Robert Emerson, lead author of the study and a former postdoctoral fellow in cognitive neuroscience at the University of North Carolina (UNC), told Live Science.

According to Emerson, about 20 percent of infants with an autistic sibling will develop the disorder, compared with just 1.5 percent in the general population.

The researchers believe the discovery could pave the way for advanced diagnostic tools that can identify autism before symptoms appear, which could provide opportunities for early intervention that could improve or even prevent the disease from progressing.

Regarding this study, Dr. Joseph Piven, Professor of Psychiatry at UNC School of Medicine said: "Our goal is to reach children early, before autism manifests, so that we can intervene more effectively."

This study was published in the American international scientific journal Science Translational Medicine.

The development of autism

Autism spectrum disorder is a brain disorder characterized by difficulties with social communication and repetitive behaviors, affecting about 1 in 68 children in the United States. Typically, behavioral symptoms begin to appear by age 2.

However, Emerson and colleagues have demonstrated that they can detect biomarkers of the disorder before symptoms appear.

In their study, the team used magnetic resonance imaging (MRI) to image the brains of sleeping infants, recording neural activity from 230 different brain regions. They paid particular attention to the degree of synchrony between these regions, also known as functional connectivity.

In total, the team measured 26,335 functional connections important for cognition, memory, and behavior. Using the MRI data, they sought to identify infants at high risk for autism.

When children turn 2, they return for behavioral assessments, including social interaction, communication, motor development, and tendency to perform repetitive actions. This data helps determine which children have autism.

With all the data collected, the researchers began training a machine learning program, which they then used to predict autism risk.

Their goal was to test whether the algorithm could accurately identify which infants would develop autism based solely on functional connectivity data from 6 months of age.

Machine learning is a form of AI that gets smarter as it processes more data. In this study, the system learned to recognize differences in functional brain connectivity in 6-month-old babies and correlated them with behavioral assessments at age 2.

To test the accuracy, the team did not use data from all 59 infants at the outset. Instead, they trained the model with data from 58 infants and used data from one remaining infant to test the predictions. This process was repeated for each infant.

“Each child is predicted individually based on the learning patterns of the other children in the group,” Emerson said.

The end result was that the machine learning program accurately predicted autism in children with 82% accuracy. This is a major step forward in the early detection of autism spectrum disorders, opening up opportunities for timely intervention to improve children's lives.

Research success thanks to the dedication of parents

Dr Piven said the team had published a previous study with impressive prediction rates, but that study required two MRI scans, one at 6 months and one at 1 year. Being able to detect autism risk with just one scan earlier is a big step forward.

What particularly moved the team was the dedicated participation of parents in this study and many others over the years. Despite extensive preparation, they were still surprised by the willingness of families to participate.

“They are an amazing group of people,” Piven said. “Not only do they have older children with autism, but they also bring their younger children, often multiple times and from far away, to one of our four clinical sites across the United States.”

“They are really dedicated,” Emerson stressed.

The research team hopes that these contributions will pave the way for more effective interventions that will give children at risk of autism a better chance of developing.