Seoul National University Hospital Develops Korea's First Korean Medical Giant Language Model (LLM)...Accuracy 86.2%
Mar 21, 2025
Seoul National University Hospital recently announced that it has developed the first Korean Medical Large Language Model (LLM) in Korea.
The model was developed using large-scale medical data such as electronic medical records (EMR), medical image storage and transmission systems (PACS), digital pathology systems, and dielectric data at Seoul National University Hospital, and made important technological advances to process medical information specific to Korea's medical system, increase treatment efficiency, and enhance patient safety.
Through this model, Seoul National University Hospital plans to lead competitive technologies in the global medical field and lay the foundation for providing better medical services to patients.
Research and development of giant language models are actively underway around the world, and various medical-specific models such as OpenAI's ChatGPT, Google's PaLM-MED2, and Microsoft's Med-LLaVA have emerged. However, existing medical LLM models are mainly optimized for Western medical knowledge, and there was a limit to not understanding Korean medical texts or Korean medical laws and medical guidelines.
In line with this global trend, Seoul National University Hospital has met the needs of domestic medical staff who mix Korean and English, and has reduced the need to develop LLM to understand medical knowledge at the level of specialists and linguistic characteristics such as abbreviations and abbreviations.
To solve this problem, Seoul National University Hospital began developing a Korean-style medical giant language model (LLM) based on Korean-style medical knowledge in March last year. Using 38 million clinical texts such as hospital admission first-time, outpatient records, surgery, prescription, and nursing records, 'Korean medical text corpus' was established, and it was disclosed so that it could be safely used in the hospital after pseudonymization and de-identification of personal information. This text corpus was utilized as basic data for model learning, which played an important role in developing information processing capabilities suitable for the Korean healthcare system.
Starting this year, Seoul National University Hospital further developed this model, integrating Korean medical laws, abstracts of Korean papers, and guidelines for medical treatment, and standardizing medical terminology dictionaries and terms. This establishes and discloses a dataset of each clinic-specific instruction training that mimics real-world care processes, and develops a knowledge graph-based search augmentation generation (RAG) and multidisciplinary multi-agent framework.
Through this process, a 'Korean Medical Giant Language Model (LLM)' that can be used practically in hospitals has been completed in one year. Seoul National University Hospital will verify the performance and safety of this model and use it for research and work assistance in the hospital.
As a result of conducting experiments on data from the last three years of the Korean National Doctoral Examination, the model achieved an accuracy of 86.2%, surpassing the actual average accuracy (79.7%) for the first time among open-source models. This is evaluated as an example of proving that Korean-style medical LLM is a practical and feasible technology in the medical field. Furthermore, with superior translation performance capable of processing approximately 50,000 words of vast text at once, the model is expected to expand performance in various medical fields in the future and further improve the accuracy and efficiency of medical data processing.
Seoul National University Hospital plans to improve the performance of LLM in the future and develop it so that it can be used in actual treatment sites.
To this end, we extend it to multimodal AI, which combines medical imaging and biosignal data, to automatically summarize medical records such as outpatient, admission, and discharge.AI', CLAIM to streamline the work of primary and insurance claims.AI', a researcher-customized curation of the latest paper 'RESEARCH.Promote the AI' project. These projects are expected to reduce burnout for medical staff and significantly improve work efficiency. As a result, medical staff will be able to spend more time on direct communication and treatment with patients away from administrative tasks, further strengthening the patient-centered care environment.
Professor Lee Hyung-chul (vice-chairman of Healthcare AI Research Institute), who led the development of LLM, said "The Korean-style medical giant language model was developed based on the medical knowledge of excellent medical staff at Seoul National University Hospital and made important technological progress to increase treatment efficiency and patient satisfaction."LLM technology will be an important tool to help doctors with their work, which will further improve the quality of healthcare."
Seoul National University Hospital President Kim Young-tae said, `This development of a Korean-style medical giant language model has 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"We will continue to introduce the latest technologies to provide the best medical services to patients in the future."
The model was developed using large-scale medical data such as electronic medical records (EMR), medical image storage and transmission systems (PACS), digital pathology systems, and dielectric data at Seoul National University Hospital, and made important technological advances to process medical information specific to Korea's medical system, increase treatment efficiency, and enhance patient safety.
Through this model, Seoul National University Hospital plans to lead competitive technologies in the global medical field and lay the foundation for providing better medical services to patients.
Research and development of giant language models are actively underway around the world, and various medical-specific models such as OpenAI's ChatGPT, Google's PaLM-MED2, and Microsoft's Med-LLaVA have emerged. However, existing medical LLM models are mainly optimized for Western medical knowledge, and there was a limit to not understanding Korean medical texts or Korean medical laws and medical guidelines.
In line with this global trend, Seoul National University Hospital has met the needs of domestic medical staff who mix Korean and English, and has reduced the need to develop LLM to understand medical knowledge at the level of specialists and linguistic characteristics such as abbreviations and abbreviations.
To solve this problem, Seoul National University Hospital began developing a Korean-style medical giant language model (LLM) based on Korean-style medical knowledge in March last year. Using 38 million clinical texts such as hospital admission first-time, outpatient records, surgery, prescription, and nursing records, 'Korean medical text corpus' was established, and it was disclosed so that it could be safely used in the hospital after pseudonymization and de-identification of personal information. This text corpus was utilized as basic data for model learning, which played an important role in developing information processing capabilities suitable for the Korean healthcare system.
Starting this year, Seoul National University Hospital further developed this model, integrating Korean medical laws, abstracts of Korean papers, and guidelines for medical treatment, and standardizing medical terminology dictionaries and terms. This establishes and discloses a dataset of each clinic-specific instruction training that mimics real-world care processes, and develops a knowledge graph-based search augmentation generation (RAG) and multidisciplinary multi-agent framework.
Through this process, a 'Korean Medical Giant Language Model (LLM)' that can be used practically in hospitals has been completed in one year. Seoul National University Hospital will verify the performance and safety of this model and use it for research and work assistance in the hospital.
As a result of conducting experiments on data from the last three years of the Korean National Doctoral Examination, the model achieved an accuracy of 86.2%, surpassing the actual average accuracy (79.7%) for the first time among open-source models. This is evaluated as an example of proving that Korean-style medical LLM is a practical and feasible technology in the medical field. Furthermore, with superior translation performance capable of processing approximately 50,000 words of vast text at once, the model is expected to expand performance in various medical fields in the future and further improve the accuracy and efficiency of medical data processing.
Seoul National University Hospital plans to improve the performance of LLM in the future and develop it so that it can be used in actual treatment sites.
To this end, we extend it to multimodal AI, which combines medical imaging and biosignal data, to automatically summarize medical records such as outpatient, admission, and discharge.AI', CLAIM to streamline the work of primary and insurance claims.AI', a researcher-customized curation of the latest paper 'RESEARCH.Promote the AI' project. These projects are expected to reduce burnout for medical staff and significantly improve work efficiency. As a result, medical staff will be able to spend more time on direct communication and treatment with patients away from administrative tasks, further strengthening the patient-centered care environment.
Professor Lee Hyung-chul (vice-chairman of Healthcare AI Research Institute), who led the development of LLM, said "The Korean-style medical giant language model was developed based on the medical knowledge of excellent medical staff at Seoul National University Hospital and made important technological progress to increase treatment efficiency and patient satisfaction."LLM technology will be an important tool to help doctors with their work, which will further improve the quality of healthcare."
Seoul National University Hospital President Kim Young-tae said, `This development of a Korean-style medical giant language model has 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"We will continue to introduce the latest technologies to provide the best medical services to patients in the future."
|
This article was translated by Naver AI translator.
