AI-Driven Pupillometric Biomarker Analysis for Early Detection of Genetic Disorders in Children

Genetic disorders in children, such as Autism Spectrum Disorder (ASD), Fragile X Syndrome, and Rett Syndrome, often go undetected during early developmental stages due to overlapping clinical features and limited access to specialized diagnostics. Early identification is crucial for timely intervention, yet current diagnostic workflows remain resource-intensive and inaccessible in many settings. This study proposes an artificial intelligence (AI)-driven approach for multi-disorder screening using pupillometry—a non-invasive method for measuring pupil responses to light stimuli. The primary objective is to develop a deep learning model capable of distinguishing between multiple pediatric genetic conditions based on pupillometric features. A synthetic dataset was constructed to simulate pupil light reflex (PLR) data for four classes: ASD, Fragile X Syndrome, Rett Syndrome, and neurotypical controls. Features such as latency, constriction velocity, and recovery time were extracted and standardized before model training. A fully connected deep neural network (DNN) was implemented and benchmarked against conventional classifiers, including Random Forest, Support Vector Machine (SVM), and Logistic Regression. The proposed DNN achieved an overall accuracy of 89%, with a macro-averaged F1-score of 0.87, outperforming the best baseline (Random Forest, 82% accuracy, 0.79 F1-score).