Digital Phenotyping to Quantify Signs of Autism in Young Children

Early ASD detection is vital but current diagnostics are slow, subjective, and resource-intensive. Since 2012, the Geneva Autism Cohort has tracked 500+ young children using eye-tracking, EEG, MRI, and standardized tests to find predictive markers and assess early interventions. This doctoral project adds digital phenotyping and deep learning to develop a non-invasive tool that captures behavioral data and detects early ASD signs for routine pediatric use. The plan emphasizes unsupervised behavior classification—adapting the LISBET model to child–adult interactions—plus pattern mining and contrastive learning, with privacy preserved by analyzing pose estimates rather than raw videos. The work will benchmark accuracy/speed against traditional methods and address ethics, particularly data security and confidentiality.