Like many neurological and cognitive disorders, diagnosing autism spectrum disorder can be tricky business. As the Centers for Disease Control and Prevention puts it, there's no blood test for it. Fortunately, it might not be too long before reading words describing social interactions and a quick brain scan could diagnosis autism with remarkable accuracy, researchers report today in the journal PLoS One.
Some kind of autism affects about one in 68 children in the United States, according to CDC statistics, though exactly what causes autism remains a bit of a mystery. Genetic, neurological, and other explanations abound, and there's no particularly clear consensus on which are correct. Still, there's an intriguing common thread. Somehow, people with autism have an abnormal sense of self.
Some kind of autism affects about one in 68 children in the United States, according to CDC statistics, though exactly what causes autism remains a bit of a mystery.
"This is a very long-standing idea, that the representation of self is altered," says Marcel Just, the new study's lead author. It goes back to early studies of autism in the 1940s, when researchers noticed that autistic children "referred to themselves as 'you.'" Most kids learn that while others use second or third person to describe them, they're supposed to use first person. "In autism, that somehow doesn't work."
The fact that autism is connected to mental representations of the self fit in nicely with another of Just's research interests, algorithms for decoding what words a person is thinking about based on their brain activity. Since autism involves atypical representations of the self, Just says, it made sense to try out these methods as a way of diagnosing the disorder. First, they used functional magnetic resonance imaging to track brain activity in 17 high-functioning autistic people and 17 healthy people as they read eight words referring to social interactions—compliment, insult, adore, hate, hug, kick, encourage, and humiliate—while thinking about that interaction from their own perspective or another person's. Using that data, the team extracted a series of fine-grained activation patterns correlated with reading the social-interaction words at both the individual and then group level. Sorting those patterns, or "features," with a standard classifier computer algorithm, Just and his colleagues could identify which were associated with autism—that is, which showed up more or less often in people with autism.
While there's something to learn from which features worked best, the more interesting thing may be how well the researchers could tell the difference between people with and without autism. As a test of their method, the team used data from 33 participants to identify the most useful features and then used the presence or absence of those features in the remaining person to predict whether he or she had autism. They got the answer right 33 out of 34 times.
That suggests that fMRI-based autism diagnosis—and perhaps diagnosis of myriad psychiatric diseases as well—could be just around the corner, Just says. When combined with additional results on differences between white matter in autistic brains and others, the findings may also help sort out the disorder's underlying causes, he says.