Researchers are actively working to expand an AI model that incorporates multimodal data, aiming to create a more robust and generalizable tool for autism diagnosis worldwide.

The study utilized data from the ABIDE cohort, involving 884 participants aged 7 to 64 across 17 sites, and found that gradient-based explainability methods produced the most consistent brain maps.

This AI model analyzes resting-state fMRI data, a non-invasive technique that measures brain activity through blood-oxygenation changes, to identify key brain regions involved in autism spectrum disorder.

By estimating the probability of ASD, the AI tool can help clinicians prioritize assessments and customize interventions, potentially enabling earlier detection and better support.

The research was led by Dr. Amir Aly in collaboration with the University of Plymouth’s engineering, psychology, and medical research groups, including the CIDER group.

While early prototypes of this AI show promise, experts stress that further validation and research are essential before it can be used clinically, emphasizing that AI should support, not replace, clinicians.

Current autism diagnosis heavily depends on behavioral assessments that can take months or even years, highlighting the urgent need for more efficient and accessible diagnostic tools.

Scientists have developed a deep-learning AI model capable of supporting autism assessment with up to 98% accuracy in classifying ASD versus neurotypical individuals, providing both predictions and explainable brain maps.