Integration of Social Behaviour Analysis and Facial Image Analysis for Autism Detection

This project presents a novel diagnostic system for early autism spectrum disorder (ASD) detection through the integration of two complementary biomarkers: eye movement data and facial image analysis. Unlike existing multimodal ASD frameworks that primarily combine behavioural or static imaging data, this system uniquely fuses questionnaire-based eye-tracking metrics with facial imaging features, enabling a more comprehensive and objective assessment of neurodevelopmental patterns. Eye-tracking data, derived from gaze fixation, saccadic movements, and social attention patterns, are processed using Support Vector Machines to classify visual attention anomalies. Concurrently, facial imaging data are analysed using a pre-trained Recurrent Neural Network for deep spatial-temporal feature extraction. The outputs from both modalities are integrated through a late fusion strategy, enhancing the model’s diagnostic robustness. Experimental results demonstrate that the proposed system achieves an overall accuracy of 94.8%, outperforming recent multimodal ASD detection frameworks by approximately 5-9% across key performance metrics such as precision, recall, and F1-score. This improvement highlights the advantage of combining dynamic eye-tracking data with static facial biomarkers. By offering a non-invasive, low-cost, and accessible approach, the study contributes to the advancement of AI-driven diagnostic tools for early ASD detection. The system’s enhanced accuracy and interpretability hold significant potential for aiding clinicians, educators, and caregivers in facilitating timely interventions and improving long-term developmental outcomes.