Ten minutes in a brain scanner could be all it takes to diagnose autism. So says Christine Ecker at the Institute of Psychiatry, UK, who has developed software that identifies the anatomical signatures of the condition.
Ecker's team carried out MRI scans on the brains of 20 adult males with autism, 20 with attention-deficit disorder and 20 healthy controls. They used a machine-learning tool called a support vector machine (SVM) – which analyses data and identifies patterns – to identify key differences between the groups, such as in the cortical folding and curvature of the brain.
The SVM was then used to build a model to predict whether brain scans fall into the autistic or control group. When the original scans were fed into this model, it diagnosed autism with a 90 per cent success rate.
Current diagnosis tools are based on time-consuming and potentially stressful behavioural tests and interviews. Ecker now plans to test her model on children, for whom she predicts more accurate results "because the differences in anatomy between the healthy and autistic brain are more prominent in childhood".
"The authors have been very careful to use this tool on a very specific group, and are not claiming that it will work on the whole spectrum of autism," says Uta Frith at University College London. "This is just the beginning, a proof of concept. It would be necessary to try it out on a new group and see whether the same results would be obtained."
She adds that studying brain anatomy alone is not enough: "We need to understand how the differences in anatomy are somehow bound up with functional differences in mental life, in how people think and feel."
Journal reference: Journal of Neuroscience, DOI: 10.1523/jneurosci.5413-09.2010
Many more stories ...
There are many more versions of this story in the internet, for example ...
and some commentary ... see