Kakao Brain Unveils Labeler Project Exclusively for Medical Image preliminary report

• A Labeler project capable of diagnosing specific diseases from image readings written in natural language.
• Capable of identifying a total of 13 diseases with high incidence rates and importance, including fractures, pleural lesions, and pneumothorax.
• Recorded an accuracy of 90.39%, higher than other models, when measuring the accuracy of 10 diagnosed diseases.

SEOUL, South Korea, Feb. 27, 2024 /PRNewswire/ -- On the 16th, Kakao Brain released a medical image labeler project on GitHub, capable of diagnosing specific diseases from chest X-ray image draft readings.

The Labeler Project is a research and development initiative that extracts specific disease names from readings written in unstructured natural language, such as bullet format. It features the ability to identify and diagnose a total of 13 diseases with high incidence rates and importance, such as fractures, pleural lesions, and pneumothorax, based on chest X-rays. For example, when a user inputs a reading written in natural language, the reading is analyzed, and the user is notified of positive/negative results for 13 diseases.

Aiming to contribute to the improvement of medical diagnoses, Kakao Brain initiated research on a labeler project that extracts specific disease names more accurately and efficiently than models previously released by other companies. This project has also been internally used to validate and study the performance of Kakao Brain's chest X-ray draft reading technology.

Kakao Brain tested the accuracy of 10 diagnosed diseases, such as fractures, pneumothorax, and pulmonary edema. Kakao Brain's Labeler project recorded an accuracy rate of 90.39%, significantly higher than other companies' models (about 76%).

Kakao Brain unveiled the Labeler project on GitHub and published a paper titled 'CheX-GPT: Harnessing Large Language Models for Enhanced Chest X-ray Report Labeling' on arXiv, an open-access archive.

Meanwhile, Kakao Brain is striving to share AI technology know-how and create new value in line with Kakao and the community's core value of 'a better world created by technology.' The unveiling of the Labeler project is also aimed at promoting the AI open-source ecosystem, exemplified by the open-source project 'Honeybee' (tentative name) released in January.

Kim Il-du, CEO of Kakao Brain, stated, "We plan to release a test set we created so that many researchers can use Kakao Brain's Labeler project as a benchmark for testing." He also added, "We plan to further improve the performance of the labeler project by utilizing our language model and incorporating additional chest X-ray data."

Kakao Brain is a Korea-based global AI technology company. Founded in 2017, Kakao Brain aims to develop AI technology that will change people's lifestyle and embraces the challenge of asking the 'Unthinkable Question' to spark innovation, allowing everyone to live better lives. Kakao Brain has developed numerous innovative AI technologies and services designed to enhance people's quality of life through continuous technological innovation. These include the large-scale language model KoGPT and the image generation model Karlo. Additionally, Kakao Brain contributes to the development of the AI technology community by releasing 'Coyo,' a dataset consisting of approximately 740 million images and text. More information about Kakao Brain can be found on their official website: https://KakaoBrain.com/.