This autism prediction breakthrough could help with early treatment

Right now there's no way to tell if a baby has autism. We know there are genetic components to autism but there is no sure fire test. It's not a product of a specific gene, or an extra copy of a chromosome that is easy to see in a genetic screening. By two-years-old, doctors are able to see signs of the disorder, but still most children aren't diagnosed until after they're four.

A new study funded by the NIH conducted by researchers at The University of North Carolina at Chapel Hill and Washington University School of Medicine in St. Louis has made steps toward some kind of early detection. The researchers, including co-author Dr. John Pruett,  developed a technique where they were able to look at fcMRI scans of six-month old brains and quite accurately predict which of them would go on to be diagnosed with autism. In this case, out of the 59 children studied, the researchers were able to identify nine out of the 11 who were later identified as having autism.

The children involved in this study all had older siblings with autism, putting them at a far greater risk of being diagnosed themselves. 

The key to the analysis was a machine learning algorithm, which was able to look at all the data and develop a model for the connections in the brain associated with normal brains as opposed to the brains of people with autism. There are many thousands of connections in the brains that were scanned, and finding any pattern in all of it is beyond the abilities of a human, so a computer program that was able to learn from the data was needed to figure out what was going on.