Attention Analysis in Robotic-Assistive Therapy for Children With Autism

Area of Science:

• Robotics in developmental neuroscience
• Computational psychiatry
• Human-robot interaction for autism therapy

Background:

• Children with Autism Spectrum Disorder (ASD) often exhibit significant attention deficits, impacting skill acquisition.
• Objective, quantitative biomarkers are needed to assess and monitor attention in ASD during therapeutic interventions.
• Existing methods for quantifying attention in ASD during human-robot interaction are limited by movement constraints.

Purpose of the Study:

• To develop a quantitative model for evaluating the attention response of children with ASD during unconstrained, robotic-assisted therapy sessions.
• To create a reliable system for therapists to objectively measure and track the attention of children with ASD.
• To establish an automated attention assessment tool for clinical use in autism therapy.

Main Methods:

• Utilized the Gaze360 model for accurate gaze extraction from video data.
• Defined angular Areas-of-Interest to pinpoint attention towards specific objects within the therapy environment.
• Integrated gaze data with defined interest areas to quantify attention periods during sessions.

Main Results:

• The developed system achieved a mean test accuracy of 79.5% in assessing attention.
• The quantitative attention index demonstrated consistency with clinical evaluations provided by therapists.
• The system successfully evaluated attention in 12 children with ASD in unconstrained settings.

Conclusions:

• The proposed quantitative system provides a meaningful and interpretable measure of attention for children with ASD.
• The high accuracy and therapist-consistent results support the potential clinical utility of this automated tool.
• This approach overcomes previous limitations, enabling attention assessment in more naturalistic therapeutic scenarios.