Identification of gesture patterns

Currently, autism is one of the most frequent developmental disorders, and its prevalence is increasing. However, the detection of autism is a complex and time-consuming task that requires a multidisciplinary team. Nowadays, we do not know the biomarkers associated with autism and its detection is based on the self-report and the observation of behaviors. As a consequence, the detection of autism is highly subjective.

In recent years, research has been conducted to uncover the computational bio-behavioral markers associated with autism. A computational bio-behavioral marker is a human feature that can be identified by computational tools. In particular, it has been found that children with autism use a different amount of force than neurotypical children when playing games with a tablet. However, a tablet using a rigid display can hardly uncover the amount of force its users use when manipulating its interface, making elastic surfaces a more appropriate technology to evaluate aspects related to force.

In this work, we hypothesize that autism can be detected by analyzing the gestures children do when interacting with an elastic surface. We conducted a deployment study where 46 neurotypical children and 26 children with autism used an elastic surface called BendableSound. We conducted 3 experiments analyzing our data by participant, gesture, and activity. We used machine learning techniques to evaluate if we can discriminate between children with autism and neurotypicals. Our results indicate that it is possible to use force as a control variable to distinguish children with autism from neurotypical when interacting with an elastic surface with a precision of 97.2% and a recall of 94.6%. An analysis of the gestural movements patterns of children with autism shows that they tend to do smaller, narrower, deeper and slower gestures than neurotypicals. This implies that children with autism practiced fewer gestures and used less force than neurotypicals.