Artificial intelligence could refine autism diagnosis: Montreal Neuro

Criteria associated with repetitive movements, markedly restricted interests, and perception-based behaviors are more strongly linked to an autism diagnosis than socialization criteria, including emotional reciprocity, nonverbal communication, and relationship building, according to research conducted at the Montreal Neurological Institute-Hospital (Neuro).

Obtaining an autism diagnosis can sometimes take years, the authors point out, which further delays interventions to improve the situation.

This increase in diagnoses could therefore be reduced by focusing on certain repetitive behaviors and specific interests, since people who present with a social disorder may have been diagnosed as autistic.

“We’ve made a lot of efforts in genetics and (medical) imaging to better understand autism, but it’s yielded much less progress than we expected,” said the study’s co-lead author, Danilo Bzdok, a neuroscientist at The Neuro and Mila – Quebec Institute for Artificial Intelligence.

“So we’re turning to other research directions, (…) so we used big data in text form, more specifically the notes of doctors who see patients who are potentially autistic. In the field of autism, to my knowledge, this has never been done before.”

The researchers used artificial intelligence to analyze more than 4,200 clinical observation reports from a cohort of French-speaking children in Montreal. Based solely on this data, they adapted and implemented extensive language modeling methods to predict such a diagnostic decision.

In particular, they found a way to identify the most relevant key phrases in a report for making a positive diagnosis, thus allowing for a direct comparison with the diagnostic criteria.

The study’s authors believe, in light of these results, that the medical world may need to reconsider and revise the established criteria for diagnosing autism.

“Our goal is not to replace doctors or to classify autistic people more efficiently,” emphasized Bzdok. “We wanted to deconstruct the process doctors use to arrive at their diagnosis.”

For several years, autism diagnoses have been on the rise in developed countries, and this study suggests that the importance placed on socialization in autism assessment may play a role.

The Diagnostic and Statistical Manual of Mental Disorders, the DSM-5, also places considerable emphasis on socialization in the diagnosis of autism, Bzdok points out, while motor behaviors such as repetitive movements, restricted interests, and perception-based behaviors are given less weight.

“We find exactly the opposite in our results,” he said. “We contradict forty years of practice and research thinking.”

This increase in diagnoses could therefore be reduced by focusing on certain repetitive behaviors and specific interests, since people who actually present a social disorder may have been diagnosed as autistic.

This would optimize the effectiveness and efficiency of diagnosis, as the assessment of social factors is quite time-consuming, tedious, and imprecise compared to more obvious behavioral traits, the study authors emphasize.

The findings of this study were published in the medical journal Cell.

–This report by La Presse Canadienne was translated by CityNews