Machine learning models using mobile game play accurately classify children with autism.

Area of Science:

• Digital Health
• Neurodevelopmental Disorders
• Human-Computer Interaction

Background:

• Telemedicine adoption is increasing due to COVID-19, presenting an opportunity to address healthcare access inequities.
• Smartphone-based interventions offer a scalable approach to healthcare delivery, particularly for developmental disorders.
• Autism spectrum disorder (ASD) diagnosis can be challenging, especially in communities with limited access to specialists.

Purpose of the Study:

• To evaluate the efficacy of a gamified smartphone application, GuessWhat?, in distinguishing children with ASD from neurotypical (NT) children.
• To assess the feasibility of using naturalistic gameplay data for ASD screening.
• To explore the potential of digital tools in improving ASD identification in underserved populations.

Main Methods:

• Development of GuessWhat?, a charades-style gamified therapeutic intervention delivered via smartphone.
• Collection of "in-the-wild" gameplay data from children with ASD and NT children.
• Training a random forest classifier on gameplay data to differentiate between ASD and NT groups.

Main Results:

• The random forest classifier achieved an Area Under the Receiver Operating Characteristic Curve (AU-ROC) of 0.745.
• The classifier demonstrated a recall of 0.769 in distinguishing between children with ASD and NT children.
• Feasibility of using naturalistic gameplay data for ASD detection was established.

Conclusions:

• The GuessWhat? application shows potential as a tool for facilitating ASD screening, particularly in hard-to-reach communities.
• Digital interventions like GuessWhat? can leverage increased telemedicine familiarity to improve healthcare access.
• Future research should focus on expanding the training dataset and analyzing demographic variations in predictive accuracy.