Researchers at the Stanford University School of Medicine and Lucile Packard Children’s Hospital have used MRI data and machine learning algorithms to classify children into autistic and non-autistic classes.
Their discovery reveals that the gray matter in a network of brain regions known to affect social communication and self-related thoughts has a distinct organization in people with autism. The findings will be published online Sept. 2 in Biological Psychiatry.
"The new findings give a uniquely comprehensive view of brain organization in children with autism and uncover a relationship between the severity of brain-structure differences and the severity of autism symptoms," said Vinod Menon, PhD, a professor of psychiatry and behavioral sciences and of neurology and neurological sciences, who led the research.
"We are getting closer to being able to use brain-imaging technology to help in the diagnosis and treatment of individuals with autism," said child psychiatrist Antonio Hardan, MD, who is the study’s other senior author and an associate professor of psychiatry and behavioral sciences at Stanford. Hardan treats patients with autism at Packard Children’s.
The study compared MRI data from 24 autistic children aged 8 to 18 with scan data from 24 age-matched, typically developing children. The data was collected at the University of Pittsburgh.
The analysis method, called "multivariate searchlight classification," divided the brain with a three-dimensional grid, then examined one cube of the brain at a time, and identified regions in which the pattern of gray matter volume could be used to discriminate between children with autism and typically developing children.
Instead of comparing the sizes of individual brain structures, as prior studies have done, the new analysis generated something akin to a topographical map of the entire brain. The scientists essentially mapped the autistic brain’s distinct cliffs and valleys, uncovering subtle differences in the physical organization of the gray matter.
Such analysis may be a more useful approach than previous tacks. Earlier studies, for instance, suggested that people with autism may have larger brains in toddlerhood or have a large defect in one brain structure. This study took a different approach and discovered several autism-associated differences in the Default Mode Network, a set of brain structures important for social communication and self-related thoughts. Specific structures that differed included the posterior cingulate cortex, the medial prefrontal cortex and the medial temporal lobes. These findings align well with recent theoretical and functional MRI studies of the autistic brain, which also point to differences in the Default Mode Network, Menon said.
Once Menon and his team had found where the differences in autistic brains were located, they were able to use their analysis to classify whether individual children in the study had autism. They used a subset of their data to "train" the mathematical algorithm, then ran the remaining brain scans through the algorithm to classify the children.
"We could discriminate between typically developing and autistic children with 92 percent accuracy on the basis of gray matter volume in the posterior cingulate cortex," said Lucina Uddin, PhD, the study’s first author. Uddin is an instructor in psychiatry and behavioral sciences at Stanford.
In addition, the children with the most severe communication deficits, as measured on a standard behavioral scale for diagnosing individuals with autism, had the biggest brain structure differences. Severe impairments in social behavior and repetitive behavior also showed a trend toward association with more severe brain differences.
When such integrated assessments are possible, the researchers hope they will allow clinicians to build detailed profiles of each patient. "We hope we’ll eventually be able to tell parents, ‘Your child will probably respond to this treatment, or your child is unlikely to respond to that treatment,’" Hardan said. "In my mind, that’s the future."