1. Can you please briefly summarize the paper? 

We developed a two-stage, multimodal AI framework to stratify the risk of Autism Spectrum Disorder (ASD) in toddlers aged 18-48 months. This study represents the first research outcome from a landmark large-scale national project in Korea, commissioned by the Ministry of Health and Welfare's National Center for Mental Health. It was conducted through a Seoul National University-Yonsei University consortium, using data from nearly 1,300 toddlers.

Using a mobile app, we collected both voice and video data during standardised parent-child play tasks, along with parent-reported screening questionnaires (M-CHAT-R/F, SCQ-L, SRS-2). For this study, we focused on voice data to enable fast screening, while planning to integrate video analysis in future work.
Key results:
    • Stage 1: AUROC = 0.942, F1 = 0.85
    • Stage 2: AUROC = 0.914, F1 = 0.90
    • Model predictions showed a strong correlation with ADOS-2 assessments (Pearson r = 0.83, p < 0.001).
By integrating validated screening questionnaires and naturalistic interaction data, this framework demonstrates a scalable, accurate, and clinically relevant approach to early ASD detection.

2. Can you please tell us the main difficulties you had in the laboratory work and how you overcame them?

Collecting reliable multimodal data from nearly 1,300 toddlers across nine hospitals over the past three years was a challenging endeavor. This study is the first outcome of a large-scale national project commissioned by the National Center for Mental Health, under the Ministry of Health and Welfare, based on data jointly collected by the Seoul National University?Yonsei University consortium.

I would like to take this opportunity to express my deep gratitude to the principal investigators, Professors Bung-Nyun Kim and Keun-Ah Cheon, and to all the clinical professors and teams for their tremendous dedication. I am especially thankful to Prof. Keun-Ah Cheon for her guidance on the clinical interpretations and implications, and to Prof. Hwiyoung Kim for his clear and robust leadership in shaping the AI analysis directions.

Furthermore, this extensive dataset would not have been possible without the families and caregivers who willingly participated in the complex process of recording home videos and completing multiple assessments. We are sincerely grateful.

Currently, we have initiated Phase 2 of this project to collect longitudinal follow-up data on these toddlers. We anticipate that this long-term data will allow us to derive even more meaningful results and deeper insights into developmental trajectories.

3. Please introduce your laboratory, university or organization to bio-researchers in Korea.

Our research was conducted at the tAI Lab (Transformative AI Lab), Digital Healthcare Center, Yonsei University College of Medicine. The lab focuses on AI-driven solutions for early detection, diagnosis, and personalized treatment.
Key research areas:
    - Multimodal AI models for ASD, ADHD, and neurodevelopmental disorders
    - Integration of surveys, voice, video, and sensor data
    - Digital health platforms for clinical decision support
    - LLM-based knowledge extraction for medical applications


4. Please tell us your experiences and your thoughts related to research activities abroad.

I gained valuable international research experience at King’s College London and the University of Waterloo, which shaped my approach to interdisciplinary AI-driven healthcare and highlighted the importance of global collaboration.
In addition, my prior experience working in global strategy consulting roles at companies such as Accenture and other international firms has provided me with a broader perspective on industry-driven innovation and the need for cross-border collaboration.
These combined experiences have enabled me to establish a solid foundation for translating AI and digital health research into practical, real-world applications with a global impact.

5. Can you provide some advice for younger scientists who have plans to study abroad?

I believe it is essential to remain open-minded and adaptable.
I encourage embracing interdisciplinary learning and building global networks early.
Experiencing diverse research environments creates new opportunities and significantly enhances the ability to collaborate across borders.

6. Future plan?

We aim to deploy this multimodal AI framework as a digital health platform by:
    • Expanding multi-center longitudinal studies
    • Adding video analysis for better prediction
    • Collaborating internationally to set AI screening standards

7. Do you have anything else that you would like to tell Korean scientists and students?

Korea has tremendous potential to lead AI-driven digital healthcare innovation, leveraging its rich clinical data and advanced digital infrastructure.
By strengthening global collaboration and embracing interdisciplinary research, Korean researchers can play a greater role on the global stage and accelerate meaningful healthcare innovations.