Social robot holds promise in autism screening

A social robot named HUMANE holds promise in screening for autism by detecting children's eye gaze during storytelling sessions, researchers from the Chinese University of Hong Kong (CUHK) have reported.

The prevalence of childhood autism in Hong Kong has increased markedly from 3,378 cases in 2012 to 12,367 cases in 2022. “However, that number may be an underestimate due to [reliance on human-based assessment] and substantially long waiting time for assessment,” wrote the researchers. The waiting time for assessment was 6─12 months in 35.1 percent of the cases, 12─18 months in 27.1 percent of the cases, and >18 months in 13.1 percent of the cases. Late diagnosis of autism is associated with emotional, behavioural and social difficulties, as well as more severe problems as affected children enter adolescence. [www.legco.gov.hk/research-publications/english/2022issh36-special-educational-needs-20221230-e.pdf; Autism Res 2024;doi:10.1002/aur.3087; J Child Psychol Psychiatry 2022;63:1405-1414]

Social robots have been used widely in autism therapy in recent decades, but research on their potential use in autism screening has been limited. HUMANE is a social robot designed specifically for autism screening by detecting gaze aversion – a reliable indicator of autism. To investigate its reliability, the CUHK researchers conducted a cross-sectional case-control study involving 199 children between the ages of 3 years and 8 years (mean age, 5.52 years; male, 74 percent). Among these children, 87 had a confirmed diagnosis of autism, 55 were suspected to have autism, and 57 showed no signs of autism.

In each session, HUMANE told a 3-minute story and monitored a child’s eye gaze simultaneously. In the event of discontinuation of eye contact for a predetermined interval of 5 seconds, HUMANE would pause the story narration and prompt the child with statements (eg, “Child, please look at me!”). When the child re-established eye contact, HUMANE provided positive feedback with praises (eg, “You have done a good job! Thanks for looking at me!”), and continued with storytelling. The mean numbers of prompts and praises were 5.38 and 3.48, respectively.

The Cohen's Kappa coefficient was used to evaluate agreement between raters. “The average Cohen’s value between HUMANE and human raters was 0.90 [standard deviation, 0.08; range, 0.83─0.91], which is considered to be almost perfect agreement,” pointed out the researchers.

The number of prompts and the duration of inattention were two effective indicators for identifying autism. The sensitivity and specificity of these indicators were ≥0.88, and the diagnostic odds ratios were >190, suggesting high accuracy. [Autism Res 2024;doi:10.1002/aur.3087]

“Our findings suggest a potential future for autism screening using social robots. However, lack of eye gaze itself is not unique to autism and could be related to attention-deficit/hyperactivity disorder or conduct disorders,” noted the researchers.

According to the researchers, there is a need to develop social robots that can interact with children and identify other signs of autism, such as stereotypical speech patterns, difficulties in expressing and comprehending emotions, and repetitive behaviours.