A groundbreaking machine learning model has been developed that can predict autism in young children using minimal information. This advancement is crucial for enabling early autism detection, which is essential for providing appropriate support.
A groundbreaking machine learning model has been developed to predict autism in young children using minimal information. This finding comes from a recent study conducted by Karolinska Institutet and published in JAMA Network Open. The model promotes early autism detection, which is vital for delivering the necessary support.
“With an accuracy of nearly 80% for children under the age of two, we anticipate that this will be an essential tool for healthcare,” states Kristiina Tammimies, Associate Professor at KIND, part of the Department of Women’s and Children’s Health at Karolinska Institutet and the study’s last author.
The research team utilized a vast US database (SPARK) containing data on roughly 30,000 individuals, both with and without autism spectrum disorders.
By examining 28 different parameters, the researchers created four unique machine-learning models to detect patterns in the data. The selected parameters were accessible information about children, which can be gathered without comprehensive assessments or medical tests before they reach 24 months. The top-performing model is named ‘AutMedAI’.
Among approximately 12,000 individuals, the AutMedAI model successfully identified about 80% of children with autism. Certain factors, including the age of first smile, the age at which a child first speaks a short sentence, and issues with eating, strongly indicated the presence of autism when used in specific combinations.
“The study’s findings are important as they demonstrate the potential to identify individuals likely to have autism using relatively simple and easily obtainable information,” asserts Shyam Rajagopalan, the study’s first author and an affiliated researcher at Karolinska Institutet, who is also an assistant professor at the Institute of Bioinformatics and Applied Technology in India.
According to the researchers, early diagnosis is essential for implementing effective interventions that support optimal development in children with autism.
“This could significantly enhance the process of early diagnosis and interventions, ultimately improving the quality of life for numerous individuals and their families,” emphasizes Shyam Rajagopalan.
The AI model demonstrated strong performance in identifying children with considerable challenges in social communication, cognitive capabilities, and more general developmental delays.
The research team is now looking to make further enhancements and validate the model in clinical environments. They are also working on incorporating genetic information into the model, which may enhance the specificity and accuracy of predictions.
“For the model to be dependable enough for clinical use, meticulous validation and rigorous development are necessary. I want to highlight that our aim is for the model to serve as a valuable resource for healthcare professionals and is not meant to replace a formal clinical autism assessment,” says Kristiina Tammimies.
The study received funding from the Swedish Foundation for Strategic Research, Hjärnfonden, and Stratneuro.
