HKUST Develops AI Super Agent for Autism Screening and Support

Dubbed ASD Super Agent, the platform integrates early risk screening, home-based intervention guidance, and continuous developmental monitoring to support parents and caregivers.

Researchers at the Hong Kong University of Science and Technology (HKUST) have developed an artificial intelligence (AI) platform that uses everyday home videos to assess autism risk in children, recommend personalized interventions and track developmental progress.

Dubbed ASD Super Agent, the platform integrates early risk screening, home-based intervention guidance, and continuous developmental monitoring to support parents and caregivers.

Parents or caregivers can upload short videos of their children's daily activities captured using smart devices, along with standardized questionnaires and basic developmental information. The AI model analyzes these inputs to generate an autism risk assessment and a report that parents can use when seeking professional evaluation.

In preliminary testing, the screening system achieved an area under the receiver operating characteristic curve (AUC) of 0.92, indicating its ability to distinguish between children with and without autism risk.

"In practical terms, the model had approximately a 92% chance of assigning a higher score to a randomly selected autism-labelled video than to a randomly selected comparison video," study lead Professor Kani Chen of HKUST Departments of Mathematics and Industrial Engineering and Decision Analytics explained to MobiHealth News.

To protect privacy, the system prioritizes extracting posture-related features on users' devices, limiting the information processed for analysis, according to HKUST.

The platform uses a "20+1" framework to guide interventions. It incorporates more than 20 established approaches, including Applied Behaviour Analysis, Naturalistic Developmental Behavioural Interventions and TEACCH Structured Teaching. The additional module allows therapists, special education providers and other organizations to contribute their own intervention methods and expertise.

"It is not intended to combine all methods indiscriminately," Prof Chen clarified. "The emphasis is on making support both personalised and usable: what the parent or the teacher can try, what response to observe and what to discuss at the next follow-up."

"Where approaches differ, recommendations should be coherent and appropriate to the child’s goals, rather than simply presenting conflicting instructions," he added.

For developmental monitoring, the platform builds an evolving profile of each child using text records, video assessments and developmental information submitted by families and professionals. It can also break long-term developmental goals into daily activities and track progress over time.

ASD Super Agent currently supports voice interactions in Cantonese, Putonghua and English. The researchers have also developed a multimodal autism database containing more than 1.1 million video, posture and behavioural records to support AI model development and further research.

Addressing Barriers to Autism Support

The development of ASD Super Agent drew on community research in which HKUST researchers identified challenges faced by families of children with autism, including lengthy assessment waiting times, high treatment costs and fragmented information about intervention services.

The researchers identified these barriers during visits to special education institutions, child development centers and psychiatric hospitals across Hong Kong, the Greater Bay Area and Beijing. They also consulted nearly 500 parents of children with autism about their experiences accessing assessment and support.

The university said the platform could help families identify potential developmental concerns earlier and access intervention guidance while awaiting professional assessment.

"We envisage both direct access for parents and use through professional service providers," Prof Chen said.

"Direct access can make everyday information and support easier to obtain, while provider-supported use can connect home observations and practice records with professional follow-up."

Further validation is needed to establish the platform's clinical effectiveness and suitability for wider use. The team plans to refine the user experience, improve system performance and expand clinical validation studies.

AI-Based Autism Screening Developments

Late last year, the Chinese University of Hong Kong's Faculty of Medicine unveiled an AI-powered stool test to assess early autism risk in children. Developed with the university's spin-off MicroSigX, the test is currently being piloted for free in the city.

Stay tuned for more such updates on Digital Health News