Photo: Buyukguzel Recep/Getty Images
Seoul National University Hospital has collaborated with Korean internet giant Naver Corporation to develop a Korean-style medical large language model.
The specialised LLM, called KMed.ai, builds on SNUH's existing LLM, which was released early this year, and is now incorporated with Naver's locally developed AI technology.
The model is not only trained on Korean medical laws and treatment guidelines, but also reflects the medical language and decision-making processes of clinicians who trained it.
WHY IT MATTERS
Following enhancements, the upgraded model, which initially scored 86.2% on the Korean Medical Licensing Examination (KMLE) over the past three years, has achieved an average score of 96.4% in the competitive test.
SNUH sees the newly unveiled Korean medical LLM to serve as a foundation for a future medical artificial general intelligence that "can continuously understand, infer, and judge various medical tasks."
THE LARGER TREND
In March, SNUH introduced what could be the first Korean medical LLM in South Korea. It was developed to address the limitations of popular LLMs, which are not optimised in the Korean medical context.
SNUH disclosed that it is also developing a medical agent platform to further improve efficiencies in clinical and administrative processes. It is also working on EMR-based document creation and diagnosis assistance functionalities, which will be gradually implemented across its workflows soon.
SNUH and Naver also recently collaborated on a transformer-based deep learning model that calculates a person's biological age and predicts their health risks using their health checkup data.
Late last year, PhynX Lab, a SK Group-backed startup, introduced South Korea's first LLM-based search platform for healthcare. The model, called Cheiron, assists pharmacists and pharmaceutical industry researchers in retrieving relevant information. The startup raised $4 million in seed funding in August.
Another Korean hospital, Korea University Anam Hospital, is also expected to pilot an LLM this year.
