Magnetic resonance imaging (MRI) studies of brain connectivity in 6-month-old infants at high risk for autism successfully identified 9 of 11 babies who went on to be diagnosed with the disorder at 24 months of age, according to findings published online today in Science Translational Medicine. Moreover, the imaging data correctly categorized all 48 of the infants who did not receive a subsequent autism spectrum disorder (ASD) diagnosis.
The findings are proof of concept that predisposing brain patterns are already present long before the defining behaviors of ASD emerge at age 2 years and may be amenable to early treatment, according to the authors. Usually, treatment begins long after the atypical brain features have been consolidated. However, if behavioral and, perhaps, pharmacologic interventions could begin in infancy, they could have a much greater effect than do the modestly effective interventions now initiated after 2 years of age.
That said, there is no existing method for diagnosing ASD before the onset of behavioral symptoms. Therefore, researchers, led by Robert W. Emerson, PhD, a cognitive neuroscientist at the University of North Carolina, Chapel Hill, and John Pruett Jr, MD, PhD, a child psychiatrist at Washington University School of Medicine in St. Louis, Missouri, applied a 15-minute scanning technique called functional connectivity MRI (fcMRI) to 59 sleeping infants at high familial risk because of the presence of an older sibling with ASD. (Having an affected sibling is known to raise an infant’s risk for autism to about 20% compared with the roughly 1.5% risk seen among those with no affected siblings.)
Functional connectivity as measured in the current study assesses how the different regions of the brain work in a synchronized manner during tasks and at rest. “It basically measures how regions of the brain are in sync with each other or not in sync with each other,” Dr Emerson told Medscape Medical News.
As part of a larger study started 10 years ago, the researchers collected complex data on 26,335 pairs of functional connections between 230 different brain regions. After scanning, the authors used a self-training computer technique, called machine learning, to read the fcMRI data and develop algorithms that identify brain patterns that accurately predict subsequent ASD diagnosis.
Functional connections were selected as correlating with at least one of the ASD-related behaviors the study assessed at 24 months, which included social behavior, language, motor development, and repetitive behavior.
“We think what we see from the resting state is a signature of how different regions of the brain work together to do things from moving an arm to looking at a picture or interacting with someone socially,” said Dr Pruett during an interview with Medscape Medical News. The extremely complex patterns that emerge indicate how brain regions support these behaviors typically or atypically.
Overall, the fcMRI machine-learning technique for classifying infants who would develop ASD was 96.6% accurate (95% confidence interval [CI], 87.3% – 99.4%; P < .001), with a positive predictive value of 100% (95% CI, 62.9% – 100%), and a sensitivity of 81.8% (95% CI, 47.8% – 96.8%).
Moreover, there were no false positives. All 48 of the infants not diagnosed with ASD were correctly classified, for a specificity of 100% (95% CI, 90.8% – 100%) and a negative predictive value of 96% (95% CI, 85.1% – 99.3%).
The authors caution that the findings need to be confirmed in larger groups. “We are hoping to expand our studies and replicate our findings,” study coauthor Joseph Piven, MD, a psychiatrist and director of the university’s Carolina Institute for Developmental Disabilities, told Medscape Medical News. He notes that the European Autism Interventions study is already conducting brain scans on at-risk infants to clarify the biology of ASD and eventually develop pharmacologic treatments.
According to Dr Piven, the fcMRI/machine learning technique used in the current study would not be practical for routinely screening infants. “Alternatively, a less expensive method could be identified to indicate increased risk. Such a test could potentially come from DNA collected in saliva and performed cheaply and routinely to screen babies for many diseases,” he said. Neuroimaging could then be used as a second level of assessment to confirm a very high risk for autism.
This study was funded by the National Institute of Child Health and Human Development and the National Institute of Mental Health. One coauthor has reported financial ties with Western Psychological Services and Biospective Inc, and coauthor Dr McKinstry is a paid consultant for Siemens Healthcare.
Sci Transl Med. Published online June 7, 2017. Abstract