MRIs were certainly performed on autistic patients in order to determine that certain physical variations in brain structure are associated with autism. But if your goal is to diagnose autism, there are much cheaper (and more accurate) ways to do so, and if your goal is to treat it, the MRI results don't really inform your treatment options, so the consensus is probably that an MRI would be nothing more than an extremely expensive way to get a vague confirmation of the diagnosis.
(Of course, if the author's hypothesis is correct, that may be about to change.)