An international joint research team, including senior researcher Han Gyeong-rim of the Brain Science Institute at the Korea Institute of Science and Technology (KIST), said on the 11th that it developed a method that uses functional magnetic resonance imaging (fMRI) and artificial intelligence (AI) to analyze the brain characteristics of patients with autism spectrum disorder and changes with treatment.
Autism spectrum disorder is a neurodevelopmental disorder characterized by difficulties in social interaction, restricted interests, and repetitive behaviors. Diagnosis and treatment progress assessments now rely heavily on symptom observation, leading to calls for a method to objectively identify patient-specific changes in brain function.
By analyzing fMRI data, the researchers found that autistic children tend to have generally lower functional connectivity between brain regions than typically developing children. In particular, decreased connectivity was pronounced around the inferior parietal lobule, which is involved in integrating sensory information, and the thalamus, which relays sensory signals.
The researchers then developed a Generative Manifold Auditing (GMA) framework to evaluate how well AI reflects actual changes in brain networks. After training AI on the brain connectivity patterns of the typically developing group, they blocked the connections of specific brain regions and analyzed the changes to see whether the model could distinguish simple data loss from real network structure changes.
Using the large-scale autism neuroimaging dataset ABIDE and clinical data from autistic children at Seoul National University Hospital for validation, a generative AI model, the denoising autoencoder, was found to distinguish structural changes in brain networks better than comparison models.
The researchers also confirmed that even if clinical symptoms improve, the patterns of change in brain functional networks can vary by individual. This suggests the potential to analyze patient-specific brain changes that are hard to capture through symptom assessments alone, using AI and neuroimaging.
Senior researcher Han Gyeong-rim said, "We confirmed the potential to objectively analyze brain network changes that differ by patient," and added, "We plan to expand the research to develop biomarkers that can be used for early diagnosis and predicting treatment progress."
This study involved the research team of professor Soyeon Caren Han at the University of Melbourne in Australia and the research team of professor Kim Bung-nyun in the Department of Child and Adolescent Psychiatry at Seoul National University Hospital. The results will be presented in the "AI for Science" session at ACM SIGKDD 2026, a data mining and AI conference, to be held Aug. 9–13 at the International Convention Center Jeju.