Healthcare (Commonwealth Union) – Canadian neuroscientists have stated that when evaluating autism—a neurodevelopmental condition that impacts approximately 80 million people globally—medical professionals tend to focus too heavily on a child’s social difficulties while overlooking their interests and spontaneous interactions with objects.
They further indicated that to enhance diagnostic accuracy, healthcare authorities should harness the analytical capabilities of artificial intelligence alongside clinical expertise to establish more precise criteria.
This is the argument put forth in a newly published study in the journal Cell.
Laurent Mottron, a clinician-researcher in psychiatry at the University of Montréal (UdeM) and co-senior author of the study indicated that a data-driven revision of autism criteria, rooted in clinical certainty, would supplement traditional expert panel assessments and human judgment, which is not infallible.
The co-first author Emmet Rabot, an UdeM clinical associate professor of psychiatry says “This project marks the successful outcome of a fruitful partnership between McGill University and UdeM. We hope our results will make a meaningful contribution to advancing diagnosis and support for the autistic community.”
The research was conducted by Danilo Bzdok, Jack Stanley Siva Reddy, and Eugene Belilovsky, all scientists at Mila – Quebec Artificial Intelligence Institute, which is affiliated with both UdeM and McGill University. Additionally, Stanley and Bzdok are connected to The Neuro – Montreal Neurological Institute-Hospital, which is also affiliated with McGill.
 Researchers of the study pointed out that since no definitive biological markers for autism have been identified in a person’s genes, blood, or brain, diagnosis continues to rely heavily on clinical evaluations performed by doctors and their assessment teams. However prior research has revealed that the cells of those with autism and those without autism reveal certain differences.
The primary method for diagnosing autism involves assessing how a child meets the criteria outlined in authoritative manuals like the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). Additionally, standardized diagnostic tools are aligned with the DSM framework.
These criteria are categorized into two main areas: differences in social communication and interaction, and patterns of restricted or repetitive behaviors, actions, or activities.
Researchers of the study indicated that ultimately, though, it is the clinicians—drawing on years of expertise—who determine whether a child receives a diagnosis, and the extent to which they align with DSM-5 criteria can vary significantly.
To systematically examine which criteria clinicians most frequently identified in individuals diagnosed with autism, researchers from McGill and UdeM analyzed over 4,200 observational clinical reports from a French-speaking cohort of children in Montreal suspected of having autism, using an AI-powered approach.
They customized and implemented large language models (LLMs) to predict diagnostic outcomes based solely on these reports. Notably, the team developed a method to pinpoint key sentences within the reports that were most indicative of a positive diagnosis.
This enabled a direct comparison with the widely recognized U.S. diagnostic criteria—yielding unexpected findings.
They discovered that socialization-related criteria—such as emotional reciprocity, nonverbal communication, and relationship-building—were not strongly indicative of an autism diagnosis. In other words, these traits were not significantly more common in children diagnosed with autism than in those who were not.
On the other hand, repetitive behaviors, intense special interests, and sensory-related behaviors showed a strong association with autism.
Based on these findings, the researchers suggested that the medical community should reassess the existing diagnostic criteria for autism. The current standards appear insufficient and may contribute to the widespread over-diagnosis of autism observed globally.
The authors argue that less emphasis should be placed on a child’s social difficulties, a factor that has been heavily weighted in diagnoses for decades. While social challenges are common among autistic children, they note that other distinctive traits—ones that are easier to recognize—also play a crucial role.
They suggest that greater attention should be given to repetitive behaviors, sensory-related traits, and intense special interests, as these may be more strongly linked to autism than previously believed.
