Canadian researchers propose a revolutionary approach using artificial intelligence to refine autism diagnosis criteria, potentially transforming the current diagnostic process and providing better support for the autistic community.
In a significant breakthrough, Canadian neuroscientists are advocating for a revolutionary approach to diagnosing autism, suggesting that the integration of artificial intelligence (AI) with clinical expertise could vastly improve diagnostic accuracy. Their study, published in the journal Cell, comes as autism diagnoses are more prevalent than ever, affecting approximately 80 million people worldwide.
For decades, diagnosing autism has relied heavily on observing a child’s social communication and interaction deficits. However, researchers from Université de Montréal (UdeM) and McGill University assert that this focus may be misplaced and incomplete.
“A data-driven revision of autism criteria of the kind we’re proposing, grounded on clinical certainty, would complement what has historically been done by expert panels and the judgment of humans, who can be wrong,” co-senior author Laurent Mottron, a clinician-researcher in psychiatry at UdeM, said in news release.
The study was a collaborative endeavor involving scientists Danilo Bzdok, Jack Stanley, Siva Reddy and Eugene Belilovsky from Mila – Quebec Artificial Intelligence Institute, which is affiliated with UdeM and McGill. Bzdok and Stanley are also associated with The Neuro – Montreal Neurological Institute-Hospital, which is affiliated with McGill.
Through analyzing over 4,200 observational clinical reports from children suspected of autism, the team employed large language modelling (LLM) technologies to predict diagnostic outcomes based on specific report contents.
The results were a surprise. The team discovered that traditional socialization criteria, such as emotional reciprocity, nonverbal communication and relationship development, were not as specific to autism as previously believed. Instead, autism diagnosis was more accurately linked to repetitive behaviors, highly specific interests and perception-based behaviors.
“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,” added co-first author Emmet Rabot, a clinical associate professor of psychiatry at UdeM.
The implications of these findings are vast. Current diagnostic practices are heavily influenced by the DSM-5, the fifth edition of the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders. This manual is considered the gold standard but relies significantly on clinician judgment, which can vary widely.
The researchers argue that medical practitioners should reconsider the established criteria and place greater emphasis on behaviors previously deemed secondary. With no definitive biomarkers for autism, refining the diagnostic criteria through AI could potentially reduce the diagnosis time, providing timely interventions that could greatly improve the quality of life for autistic individuals.
“In the future, large language model technologies may prove instrumental in reconsidering what we call autism today,” added co-senior author Bzdok.
This study represents a critical step towards enhancing the accuracy and efficacy of autism diagnoses, potentially heralding a new era where AI and human expertise work hand-in-hand to better serve the autistic community.
Source: University of Montreal