An image-based analytics framework for early autism detection using eye movements

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition whose early detection is crucial for improving social and cognitive outcomes. Current diagnostic tools are often costly, subjective, and inaccessible to many clinics. This work presents GazeScan, an image-based analytics framework that identifies ASD from eye-tracking behavior using only standard video input. The system non-invasively performs gaze estimation via a 16-point geometric calibration and transforms gaze trajectories into grayscale scanpath images. These images are classified using a lightweight convolutional neural network. GazeScan was evaluated on the Eye-Tracking Scan Path (ETSP) dataset with five-fold cross-validation, achieving 97.01% accuracy and an AUC of 0.98. The model’s compact architecture enables real-time inference and mobile deployment without specialized hardware. The results obtained highlight the potential of accessible, AI-enabled digital screening tools to support early ASD detection and broader behavioral healthcare delivery. • Identify autism risk using visual scanpath images derived from eye movement patterns. • Leverage deep learning on grayscale scanpaths for efficient behavioral screening. • Visualize user gaze behavior to support early-stage autism diagnosis. • Analyze facial landmarks to generate reliable gaze-based health assessments. • Support mobile diagnostics with lightweight visual analytics from standard video.