Computer Vision Analysis for Quantification of Autism Risk Behaviors

Area of Science:

• Neurodevelopmental Disorders
• Autism Spectrum Disorder Research
• Behavioral Analysis

Background:

• Observational behavior analysis is crucial for identifying neurodevelopmental disorder risk markers.
• Early behavioral markers for autism spectrum disorder (ASD) appear by 12 months, with diagnosis possible at 18 months.
• Current observational methods rely on trained specialists, making them costly, time-consuming, and difficult to scale.

Purpose of the Study:

• To introduce a novel, automated system for behavioral analysis in young children.
• To validate a mobile application designed to elicit and record behavioral responses for ASD risk assessment.
• To demonstrate the system's utility in quantifying behaviors for enhanced risk marker research.

Main Methods:

• Development of a closed-loop mobile application presenting movie stimuli to elicit child responses.
• Utilizing mobile device cameras for recording behavioral and social responses.
• Employing computer vision algorithms for automated analysis of recorded data.

Main Results:

• The system successfully measured engagement, name-call responses, and emotional responses in toddlers.
• Validation demonstrated the system's ability to differentiate responses between toddlers with and without ASD.
• Examples showed the framework's potential for fine-grained behavioral quantification in risk marker studies.

Conclusions:

• Objective, automated methods can significantly aid behavioral analysis in developmental research.
• The developed system is suitable for large-scale, longitudinal studies requiring objective behavioral assessment.
• This technology offers a promising avenue for early detection and research into autism spectrum disorder.