LAUSR

Key Points Question Can joint attention be quantified for detecting autism spectrum disorder (ASD) and assessing ASD symptom severities? Findings In this diagnostic study of 45 children with ASD and 50 with typical development, a deep learning system trained on videos acquired using a joint attention–eliciting protocol for classifying ASD vs typical development and predicting ASD symptom severity showed high predictive performance. This new artificial intelligence–assisted approach based predictions on participants’ behavioral responses triggered by social cues. Meaning These findings suggest that this method may allow scalable, digitalized measurement of joint attention, enabling deep learning–based analysis and modeling to facilitate development of automated detection and symptom severity assessment tools for individuals with social deficiencies.