AI-Powered Autism Detection in Children from Images Beyond Human Perception

Early diagnosis of autism spectrum disorder (ASD) is time-consuming, and standard measurements are not sufficient; hence, new methods based on artificial intelligence (AI) were developed to be employed as diagnostic tools by examining autistic children's facial images. Existing methods rely on subjective behavioral assessments, which are time-consuming, prone to error, and ineffective at detecting minor symptoms of ASD. The paper introduces a computer-like emotion recognition model that automatically detects face features including eye staring, asymmetry, and micro-expressions all of which are commonly overlooked by human observers using Deep Learning (DL) via Convolutional Neural Networks (CNNs). The results show that the proposed system outperformed the previous system in recall (94%), precision (91 %), and F1-score (92%) and accuracy (92%). It also significantly reduces false positives and negatives. The algorithm significantly improved its accuracy over a wide range of demographics and performed better with larger datasets, making it a scalable and efficient new approach for ASD detection and early intervention.