A Korean research team has developed an artificial intelligence (AI) model that can preliminarily check on a smartphone whether an infant or toddler is on the autism spectrum.
Severance Hospital said on the 10th that a research team led by Severance Hospital Professors Cheon Geun-a and Kim Hwi-young and Seoul National University Hospital Professor Kim Bung-nyun developed an AI model based on data from 1,242 infants and toddlers aged 18 to 48 months who visited nine hospitals in Korea.
Autism spectrum disorder is a developmental disability characterized by difficulties in social interaction or communication and restricted, repetitive behaviors.
Early diagnosis and treatment can significantly improve communication and interaction skills, but symptoms are hard to detect, so diagnosis and treatment are often delayed. According to a 2020 announcement from the U.S. Centers for Disease Control and Prevention (CDC), 1 in 3 children with autism spectrum disorder were not diagnosed until after age 8.
The AI developed by the team helps determine a disability based on a child's voice recorded by parents on a smartphone. It analyzes the tone and rhythm of the child's voice when interacting with others, along with vocal patterns, improving accuracy over conventional survey methods.
It presents tasks by month of age—such as prompting a response when the child's name is called, imitating a parent's actions, pretend play using a ball or toys, and asking for help—and has parents record them. The AI integrates and analyzes these voice recordings with results from an existing autism spectrum disorder screening completed by parents.
The accuracy of existing survey-based tests is around 70%, but this AI model's accuracy exceeds 94%, the research team said. The team first distinguished typically developing children from autism risk groups, and said it identified high-risk groups and children with actual autism with 85% accuracy.
The research team expected that using this at home to first check the possibility of a disability and then receiving a specialist's diagnosis would enable earlier diagnosis and treatment.
Professor Cheon Geun-a said, "There are many cases where children visit for an initial consultation only after autism spectrum disorder has become more severe," adding, "Because the AI we developed can be used at home, faster diagnosis becomes possible and that can lead to better treatment outcomes."
Supported by the National Center for Mental Health's digital therapeutics development project for developmental disabilities, the study was published in the latest issue of Nature's partner journal npj Digital Medicine.
References
npj Digital Medicine (2025), DOI: https://doi.org/10.1038/s41746-025-01914-6