Seoul Nationwide College Hospital has developed what might be the primary medical massive language mannequin in South Korea.
The challenge to develop the medical LLM started final 12 months in March. The SNUH analysis workforce began by gathering 38 million scientific texts, together with hospital admissions, outpatients, surgical, prescription, and nursing information, and de-identified and anonymised them to create a base corpus for mannequin studying.
Then, initially of 2025, it developed department-specific data bases that had been later built-in into the LLM via retrieval-augmented era (RAG). These data bases include native medical legal guidelines, paper abstracts, and remedy tips in Korean, in addition to medical terminology requirements and abbreviation dictionary.
The LLM was examined on questions from the Korean Medical Licensing Examination of the previous three years and scored 86.2%, which surpassed the common takers’ rating of 79.7%. SNUH additionally mentioned the mannequin has proven excessive translation efficiency, processing 50,000-word texts concurrently.
The SNUH analysis workforce will confirm the LLM’s efficiency and security for a 12 months earlier than it may be utilized to help scientific work and analysis. They additionally plan to develop its utility in numerous medical fields and additional improve the mannequin’s medical knowledge processing functionality.
WHY IT MATTERS
In keeping with SNUH, present medical LLMs developed by tech giants, together with Google’s Med-PaLM 2 and Microsoft’s LLaVA-Med, are restricted to comprehending Korean medical texts, legal guidelines, and remedy tips. As these fashions are optimised within the Western context, these are discovered unreliable for Korean medical professionals. Therefore, the hospital researchers started growing a Korean medical LLM to fulfill the wants of native clinicians.
THE LARGER TREND
Late final 12 months, a startup backed by conglomerate SK Group claimed to introduce the primary LLM-based search platform particular to healthcare in South Korea. PhynX Lab’s Cheiron assists pharmacists and pharmaceutical trade researchers in retrieving related data by selecting up the person’s intent via pure language processing. It will probably additionally self-reverify data and rapidly generate high-quality solutions.
Korea College Anam Hospital can be reportedly growing its personal LLM, which is anticipated to be piloted this 12 months.
Asan Medical Heart has additionally utilised an LLM to create its voice-based scientific scribe, which is now used throughout its 16 departments.
In the meantime, SNUH is presently engaged on different AI initiatives to free clinicians from administrative work. This features a multimodal AI that may routinely generate affected person summaries, AI for enhancing insurance coverage declare processing, and AI that curates the newest papers for researchers.
ON THE RECORD
“By means of the event of this Korean-style medical LLM, now we have opened a brand new chapter in medical innovation by maximising the work effectivity of medical employees and offering quicker and extra correct medical companies to sufferers,” mentioned SNUH president and CEO Dr Kim Younger-tae in an announcement.
