Seoul National University Hospital (SNUH) has created what is believed to be the first medical large language model (LLM) in South Korea.
Project Overview
- The development of the medical LLM commenced in March of last year.
- The research team collected 38 million clinical texts, including records from hospital admissions, outpatient visits, surgeries, prescriptions, and nursing documentation.
- All data was de-identified and anonymized to establish a foundational corpus for model training.
Knowledge Base Integration
- By early 2025, the team developed specific knowledge bases for various departments.
- These bases were integrated into the LLM using retrieval-augmented generation (RAG).
- The knowledge bases include local medical laws, abstracts of research papers, treatment guidelines in Korean, and medical terminology standards.
Performance Testing
- The LLM was evaluated using questions from the Korean Medical Licensing Examination over the past three years, achieving a score of 86.2%, which is higher than the average score of 79.7%.
- Additionally, the model demonstrated strong translation capabilities, processing texts of up to 50,000 words simultaneously.
Future Plans
The SNUH research team will spend the next year verifying the LLM’s performance and safety before it is utilized in clinical settings and research. Plans are also in place to broaden its application across various medical fields and enhance its data processing capabilities.
Importance of the Development
SNUH noted that existing medical LLMs from major tech companies, such as Google’s Med-PaLM 2 and Microsoft’s LLaVA-Med, struggle with Korean medical texts and guidelines. These models, optimized for Western contexts, are often unreliable for Korean healthcare professionals. This prompted the development of a tailored Korean medical LLM to better serve local clinicians.
Broader Context
- Last year, a startup supported by SK Group introduced the first LLM-based healthcare search platform in South Korea, named Cheiron, which aids pharmacists and researchers in retrieving relevant information.
- Korea University Anam Hospital is also working on its own LLM, expected to be piloted this year.
- Asan Medical Center has implemented an LLM to create a voice-based clinical scribe used across its 16 departments.
Ongoing AI Initiatives
SNUH is actively pursuing additional AI projects aimed at reducing administrative burdens on clinicians. These initiatives include:
- A multimodal AI for automatic patient summary generation.
- AI solutions to enhance insurance claim processing.
- AI that curates the latest research papers for medical professionals.
Official Statement
Dr. Kim Young-tae, president and CEO of SNUH, stated, “Through the development of this Korean-style medical LLM, we have opened a new chapter in medical innovation by maximizing the work efficiency of medical staff and providing faster and more accurate medical services to patients.”