Development of a Lightweight Model Optimized for Practical Use in Medical Settings

On April 1, Lunit announced that it has successfully completed the first phase of clinical validation in the Ministry of Science and ICT's national strategic project, the 'AI Specialization Foundation Model,' and has now entered the second phase, which focuses on nationwide-scale validation and further model advancement.


Lunit is leading the consortium for the project titled "Development of a Multiscale Medical Science Foundation Model for Full-Cycle Medical Innovation from Molecule to Population," as the principal organization, bringing together 23 participating industry, academic, research, and medical institutions.

Lunit Passes First Phase of Government-Led 'Medical Science Foundation Model' Project View original image

In this first phase of the project, the Lunit consortium independently developed a medical science foundation model with 16 billion parameters from scratch, achieving a lightweight model optimized for both practical use in medical settings and cost efficiency.


For the first-phase training, an extensive, specialized dataset was intensively utilized, covering all areas of the medical field, including academic papers, clinical records, pharmaceutical information, and medical guidelines. Through this, the consortium established a Retrieval-Augmented Generation (RAG) function that automatically searches for and presents relevant evidence, as well as an agent system that connects various specialized models to solve complex problems step by step. The model achieved its target performance in all five major evaluation metrics, including medical literature comprehension and evidence-based responses. The company explained that it also produced superior results compared to other large language models (LLMs).


The clinical validation test conducted at National Health Insurance Service Ilsan Hospital was also completed successfully. The initial model for the Clinical Decision Support System (CDSS) was applied to adverse drug reaction (ADR) analysis, emergency room patient triage, and initial diagnosis support processes. Lunit's foundation model demonstrated potential to reduce pharmacists' review workload during ADR analysis and enhance efficiency in pharmaceutical safety management.


In the second phase of the project, the scope of validation will be expanded to nine medical institutions across the country, including Yongin Severance Hospital, Korea University Medical Center, Konyang University Hospital, and Yangsan Pusan National University Hospital, as well as two pharmaceutical companies.



Beomseok Seo, CEO of Lunit, stated, "Having confirmed the feasibility of field application for the lightweight model in the first phase, we will contribute to tangible changes in medical practice through further model advancement and expanded validation in the second phase."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing