Diagnosing autism is not easy. Doctors currently diagnose autism in children by observing behavior. But researchers at Standford University believe they have developed a way to use brains scans that may help identify autism in children in the future.
Using MRI scans, researchers were able to determine that autistic brains have a unique shape when compared to typically developing brains.
They found that there are significant differences in areas of the brain called the Default Mode Network, a set of brain structures associated with social communication and self-awareness.
A study published Friday in Biological Psychiatry finds that the greater the difference in brain structure, the more severe the case of autism.
Researchers applied new algorithms to analyze the brain scan data and found they are highly accurate– correctly distinguishing between autism and non-autism about 90% of the time, according to the study.
What algorithms lack is the ability to identify autism in a real-world setting, where a patient may fall anywhere along the autism spectrum, or have other conditions as well, like Attention deficit hyperactivity disorder (ADHD).
"We haven't investigated what's called positive prediction value, which is: If I'm tested positive with this kind of diagnosis, do I have the disorder? Whereas what we've done is: If I have the disorder, do I test positive?" says Vinod Menon, Ph.D., who led the research. "And they're completely different questions."
Other researchers, like Christine Ecker, Ph.D., at King's College London, are working on how to move from confirming diagnosis, as Menon describes, to being able to assist with diagnosing cases of autism, with the help of MRI's and algorithms.
Ecker says algorithms have to learn how to distinguish between autism and non-autism, and the more samples of each brain type are available, the better algorithms can become at figuring out which is which.
Brain-imaging studies typically have small sample sizes. In the study published Friday, researchers used just 24 high-functioning children with autism and 24 typically developing children.
One reason for the small sample size is cost. Brain imaging is expensive. Also, low-functioning autistic children usually cannot lie still in the scanner, which is essential, and many can't tolerate the noise, which further limits the available sample size, says Ecker.
Another difficulty: All the data used in a study has to come from the same scanner model or MRI.
“It’s just a matter of getting the data,” says Ecker, whose clinic now collects about 2 brain scans a week in hopes of conducting research with a larger sample size.
“We're going to make a prediction based on the brain images and those then will be compared to the clinician's gold standard evaluation, and hopefully it validates. That would be a main step forward to using those techniques in a clinical setting.”
The scanning technology could eventually be especially useful in diagnosing toddlers, and lead to better treatments in the future, says study author Dr. Antonio Hardan, who also treats patients with autism.
This is the first study of its kind in which an algorithm predicted autism in brain scans of children aged 8 to 18, using data to map out the Default Mode Network.
Autism spectrum disorder now affects about 1 in 110 children in the United States, according to the Centers for Disease Control and Prevention.