Early Detection of Neurodevelopmental Disorders: Quantifying Autism Behavioral Markers with Computer Vision and Artificial Intelligence

Autism spectrum disorder (ASD) is traditionally diagnosed through clinical observation and standardized tests, processes that are time-consuming and require expert intervention. Early detection is crucial for effective intervention, yet current methods often delay diagnosis, particularly in resource-limited settings. This study presents a scalable, non-invasive method for early autism detection using standard webcam technology to measure biomarkers associated with ASD. The system captures and analyzes eye-tracking data, head movements, and behavioral responses during a controlled 4 -minute video presentation. Data is processed through a series of carefully selected machine learning models, with the most advanced model—utilizing a Convolutional Neural Network (CNN) with ResNet layers—achieving an 91% accuracy rate. Our pilot study, involving 147 children, resulted in the creation of a proprietary dataset, supporting the robustness of this method.