Novel AI Method Identifies Early ASD Indicators in Children

Researchers at ETRI have developed an artificial intelligence (AI) method for early screening of autism spectrum disorder (ASD). This technology can help identify early signs of ASD in children more quickly, potentially improving access to screening and intervention.

The Electronics and Telecommunications Research Institute (ETRI) created a “social interaction recognition AI” by analyzing “social interaction-inducing content” for young children. The AI screens for ASD using video footage recorded within six minutes of the child watching this content.

ASD can be detected by observing specific behavioral and developmental patterns. These include repetitive actions and limited social communication. Early detection and proper intervention are important, as they can improve developmental outcomes.

Despite this, diagnosis often takes two to six years after symptoms appear. This delay is due to a shortage of specialists, limited awareness, and constraints in time and resources. Early screening is critical, since primary symptoms can appear as early as 12 to 24 months of age, or even earlier.

Using data from 3,531 infants and toddlers aged 42 months or younger, ETRI collaborated with Prof. Yoo Hee-Jung’s team from the Department of Psychiatry at Seoul National University Bundang Hospital. They studied the effectiveness of the AI screening index for ASD. Based on this data, they developed a scenario for observing children that supports AI-based analysis.

The researchers created the world’s first “social interaction-inducing content” for this purpose. It prompts various social behaviors such as showing interest in objects, responding to names, imitating actions, pointing, and making eye contact.

They also developed a “Social Interaction Recognition AI” system that uses cameras to capture how children react to the content. The AI can analyze individual traits, predict emotional states, track gaze, respond to name recognition, and identify pointing, repetitive, and imitative behaviors.

In 2020, a living lab for ASD screening was established at the Korea Institute of Robotics and Technology Convergence (KIRO) Seoul Center. Over the past five years, researchers have gathered data and conducted observational studies on infants and toddlers, helping to advance the technology.

This is the first multidisciplinary AI tool for ASD screening. It addresses limitations of current methods and offers a more objective, quantitative approach.

The technology helps lower barriers to mental health services. It allows children to be screened while interacting with content at preschools, daycare centers, developmental facilities, or at home.

It is expected to support early screening of infants and children by increasing awareness of autism spectrum disorders. It also encourages early intervention and preventive testing.

We hope that this will help shorten the time between symptom detection and diagnosis, along with changing societal perceptions of autism. In addition, it is important for our research to solve hard problems, but we also hope that it will also contribute more to solving important problems like autism.

Dr. Yoo Jang-Hee, Principal Researcher, Social Robotics Research Section, Electronics and Telecommunications Research Institute

More than 50 domestic and international applications have been filed based on this technology. It has also led to 18 publications in international journals. In 2024, it was selected as one of the “100 National R&D Excellence Achievements” (SCIE).

1. Autism Spectrum Disorder (ASD): A neurodevelopmental disorder characterized by limited and repetitive behaviors and interests, along with difficulties in social communication and interaction. The term “spectrum” refers to the variation in symptom patterns and severity among individuals.
2. Early screening and intervention: The process of identifying developmental disorders as early as possible. When necessary, it involves providing medical or psychological support to manage or reduce symptoms.
3. Gaze tracking and response to name: Gaze tracking refers to observing where a child looks when shown specific objects. Response to name refers to how strongly a child reacts when their name is spoken.
4. Stereotyped and imitative conduct: Stereotypical behavior involves repeating certain words or actions. Imitative conduct refers to copying another person's actions or movements.
5. Living Lab: A research setting that allows technology to be tested and developed in environments that mirror real-world conditions. In this study, data was gathered and the technology was tested as babies and caregivers interacted with the content.

This study was a collaboration among multiple research teams. Prof. Yoo Hee-Jung’s team from Seoul National University Bundang Hospital, Prof. Kim Hong-Guk’s team from the Gwangju Institute of Science and Technology (GIST), and Dr. Kim Min-Kyu’s team from the Korea Institute of Robotics and Technology Convergence (KIRO) all contributed. The project was part of the Institute of Information and Communications Technology Planning and Evaluation’s (IITP) “SW Computing Industry Core Technology Development Project,” supported by the Ministry of Science and ICT.