Will artificial intelligence system catch lung cancer in health checkup?

-Use of 100,525 chest x-rays for the general public who received medical examination

-Significantly increase the probability of lung cancer diagnosis and expect early diagnosis

The artificial intelligence chest X-ray diagnosis system developed by Seoul National University Hospital proved its performance again.

Seoul National University Hospital radiology department professor Changmin Park team (Jonghyuk Lee and Hyeyoung Sun) announced the results of verifying the performance of the diagnosis system using chest X-ray images of 50,070 people who had undergone health checkups in 2008-2012.

Figure 1. Chest X-ray photograph taken by medical examination. There is a shadow suspected of lung cancer on the upper right lung (left), which is also suspected of lung cancer on a chest computed tomography (center). The artificial intelligence system identifies the presence and location of this lesion and determines it as lung cancer (right).

The collected data was a total of 100,576 copies, and the actual lung cancer was 98. Among them, the performance of the diagnostic system was measured after excluding 51 sheets, where it was difficult to determine whether it was lung cancer using only chest X-rays. As a result, the artificial intelligence diagnosis system showed a diagnosis accuracy of about 97% and proved excellent performance. It also showed an excellent sensitivity of about 83%. Sensitivity is a good indicator of the performance of a test method or predictive tool. In particular, it showed a sensitivity of 100% in lung cancer, which is very clearly visible.

Table 1. Summary Table of Research Results

Real lung cancer THE

Real lung cancer X


Positive test result




Negative test result








This study is meaningful in that it has verified the system’s diagnostic ability in the actual health examination situation for healthy generalists. The performance of the AI ​​diagnostic system has been verified through previous studies, but the results are unknown when applied to the general public with low disease frequency.

In this study, the incidence of lung cancer was very low, about 0.1% of the total 50,070 examinees. As a result of the experiment, the diagnostic system proved its performance with high accuracy even in real situations.

In Korea, more than 5 million people receive health check-ups per year. As a large number of chest X-ray examinations are performed, the work of the radiologist to be read becomes heavy, and there is a high risk of reading errors. In the future, it is expected that artificial intelligence will be able to relieve the heavy work in the area.

Professor Chang-min Park said, “Through this study, we have confirmed that the artificial intelligence system is useful for finding lung cancer in a large-scale health checkup.” “Research so that artificial intelligence can be of practical help to actual patients and the general public beyond the laboratory level. We will continue development.”

Figure 2 A graph comparing the diagnostic capabilities of an artificial intelligence system and radiology specialists for lung cancer diagnosis on a chest X-ray of a medical examination. The AI ​​system’s diagnostic ability graph (blue) is located on the dot (red) representing the diagnostic ability of radiologists, indicating better diagnostic ability.

This study was conducted with the support of intensive nurturing research at Seoul National University Hospital, and utilized the Lunet Insight CXR artificial intelligence jointly developed by Seoul National University Hospital and Lunit Co., Ltd. The research results were published in the online edition of’Radiology’, the leading academic journal in the field of radiology.



Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.