Digital Behavioral Phenotyping Detects Atypical Pattern of Facial Expression in Toddlers with Autism

Area of Science:

• Neuroscience
• Developmental Psychology
• Computer Science

Background:

• Current autism spectrum disorder (ASD) screening relies on subjective caregiver reports, necessitating more objective and scalable methods.
• Behavioral observation for ASD is accurate but resource-intensive, creating a need for accessible diagnostic tools.
• Early detection of ASD is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To evaluate a tablet-based behavioral assessment using computer vision analysis (CVA) for detecting facial expression patterns in toddlers.
• To determine if CVA-detected facial expression patterns can differentiate toddlers with and without ASD.
• To explore the feasibility of using technology for objective ASD risk behavior assessment.

Main Methods:

• A tablet-based assessment was administered to 104 toddlers (22 with ASD), recording facial expressions while they watched movies.
• Computer vision analysis (CVA) automatically tracked facial landmarks to quantify head position and expressions (Positive, Neutral, All Other).
• Statistical analysis identified specific facial expression patterns that distinguished between toddlers with and without ASD.

Main Results:

• CVA successfully identified facial expression patterns that differentiated toddlers with and without ASD (AUCs 0.62–0.73).
• Toddlers with ASD exhibited more Neutral facial expressions.
• Toddlers without ASD showed more 'All Other' expressions, often linked to engagement (e.g., raised eyebrows, open mouth).

Conclusions:

• Computational analysis of facial movements via a tablet assessment can detect differences in affective expression, a core feature of ASD.
• This technology-based approach shows potential for the early, objective detection of ASD symptoms.
• Tablet-based behavioral assessments offer a scalable and feasible tool for characterizing ASD risk behaviors.