Innovative AI Model Promises Reliable and Transparent Autism Diagnosis Support

Scientists at the University of Plymouth have developed a cutting-edge deep learning model designed to enhance autism assessment processes. This advanced AI system analyzes resting-state functional MRI (fMRI) data—a non-invasive imaging technique that reflects brain activity through blood-oxygen levels—and reliably predicts Autism Spectrum Disorder (ASD) with accuracy rates reaching up to 98% during cross-validation. Notably, the model not only provides high-precision results but also offers clear, explainable insights by highlighting key brain regions responsible for its predictions, making its decision process transparent for clinicians.

The significance of this innovation stems from the current challenges in autism diagnosis, which traditionally depends on behavioral evaluations conducted in clinical settings and can involve long waiting periods—sometimes taking many months or even years. Early diagnosis is crucial as it allows for timely intervention, improving developmental outcomes and quality of life for autistic individuals.

This study, published in eClinicalMedicine, was led by undergraduate researcher Suryansh Vidya under the supervision of Dr. Amir Aly and supported by teams from the School of Engineering, Computing and Mathematics, and the School of Psychology at the University of Plymouth. Dr. Aly emphasized that the AI tool aims to assist, not replace, clinicians by providing supplementary, explainable insights that could help prioritize assessments and personalize support strategies.

The research utilized data from the Autism Brain Imaging Data Exchange (ABIDE), encompassing 884 participants aged 7 to 64 across 17 different sites. The team compared various explainability techniques and found gradient-based methods to be most effective. Additionally, the study’s Map outputs demonstrated consistent brain regions influencing predictions across preprocessing approaches.

Moving forward, Ph.D. researcher Kush Gupta is expanding this work by integrating different data types and machine learning techniques to develop a robust, generalizable model usable worldwide. Senior author Professor Rohit Shankar highlighted that while AI holds great potential in early autism detection and diagnosis, further validation and research are necessary to bring these prototypes into clinical practice.

Dr. Aly remarked on the potential impact, stating, "AI can support clinicians with accurate, transparent insights, including probability scores, to improve assessment efficiency and decision-making." Overall, this innovative approach represents a significant step toward faster, more accurate autism diagnosis through advanced AI techniques.

For more information, see: Identification of critical brain regions for autism diagnosis from fMRI data using explainable AI.