If scientists create an autism test for children this would make many people's lives easier:
1) Children on the higher end of the spectrum (more mild) are often not diagnosed correctly until they are well past the 18-24 months that a more severely autistic child routinely diagnosed.
2) I've seen enough doctors get it wrong even with obviously autistic kids for me to believe there is value in a general autism test for anyone.
3) Many parents will find it helpful to have medical confirmation of their child's diagnosis versus just going on the doctor's say-so.
But... there are some issues with the study.
First: it's incredibly difficult to read and understand. There's a chance I didn't learn about everything they did because I focused on the main part of the results section. (Here... you try reading it: http://www.jneurosci.org/cgi/content/full/30/32/10612
If I got anything wrong, I'm sure you guys will tell me.)
The smallest issue is this: they only tested a small sample size of adults. Would brain scans also work with children? Will these results hold in a larger study?
But the biggest issue is this: according to the study (or more accurately, according to my attempt at navigating the dense, scientific jargon laced pages), 80% of the non-autistic control group were correctly classified as non-autistic.
That may sound pretty great, but a 20% false positive rate is a nightmare.
Using Bayes' Theorem (google it if you want to know more) I found that if 1000 random children take the test, seven will be correctly diagnosed autistic. But one autistic child would not be diagnosed (assuming the 85% success rate) and 200 non-autistic children would be diagnosed autistic. This is because 20% of the time, you get that false positive.
This means if your child took the brain scan and the test came back that he was autistic, there would only be a 4% chance that it was correct.
I don't have to have spent the thousands of hours that I have with parents in my career to know that's not going to fly.