Deepnoid Demonstrates Excellence of LLM-Based AI Medical Solutions... Leading the 56 Trillion Won Market

Joint Research with Prof. Lee Ro-un of Inha University Hospital Radiology Department
Research Paper Published in SCI-Level International Peer-Reviewed Journal
M4CXR Surpasses ChatGPT in Localization and Diagnostic Accuracy

Deepnoid, a first-generation domestic medical artificial intelligence (AI) specialist company, jointly conducted a study titled “A Comparative Study on the Chest X-ray Interpretation Performance between Deepnoid’s M4CXR and OpenAI’s ChatGPT” with Professor Lee Ro-woon’s team from the Department of Radiology at Inha University Hospital. The study was published in the SCI-level internationally recognized journal, Journal of Clinical Medicine.

Deepnoid’s M4CXR is a cloud-based medical AI solution designed as a large language model (LLM) that analyzes chest X-ray images and generates draft reports. Through the study, M4CXR was proven to be suitable as an auxiliary tool that enhances clinicians’ capabilities in actual medical settings during chest X-ray diagnosis.

This research was led by Professor Lee Ro-woon, an expert in medical AI from the Department of Radiology at Inha University Hospital. The interpretation results of M4CXR and ChatGPT were evaluated based on five indicators: accuracy, false positives, location errors, count errors, and hallucinations. M4CXR consistently outperformed ChatGPT across all evaluation categories. In terms of location accuracy (76?77.5%), M4CXR significantly surpassed ChatGPT (36?36.5%). Regarding diagnostic accuracy, M4CXR (60?62%) was also markedly superior to ChatGPT (42?45%).

The study results demonstrated that M4CXR is suitable for use in professional medical environments and showed significantly better performance compared to the general-purpose AI, ChatGPT. It is expected that when M4CXR is used in actual medical settings in the future, it will optimize diagnostic workflows and complement the expertise of medical professionals.

Professor Lee Ro-woon, the first author of the paper from the Department of Radiology at Inha University Hospital, stated, “Through this study, we were able to confirm the potential of M4CXR as a diagnostic support tool in the medical field,” and added, “The excellent performance demonstrated in location accuracy and diagnostic precision will provide great value in real clinical settings.”

He further added, “We plan to strengthen the clinical validity of M4CXR through comparative studies with actual specialists’ interpretations and, based on this, further improve diagnostic accuracy and efficiency.”

Choi Woo-sik, CEO of Deepnoid, said, “The study proves that M4CXR has excellent performance surpassing ChatGPT in the specialized medical field,” and added, “We are preparing to enter the imaging interpretation market with commercialization targeted for the first half of next year.”

He emphasized, “The global imaging interpretation market is expected to grow at an average annual rate of 23.7% until 2030, reaching approximately 56 trillion KRW in size,” and stated, “We will lead the market through M4CXR.”