A highly accurate AI model can estimate lung function using only chest X-rays

If there is one medical examination that everyone in the world has undergone, it is a chest x-ray. Doctors can use X-rays to see if someone has tuberculosis, lung cancer, or other diseases, but they cannot use them to see if the lungs are functioning properly.

Until now, that is.

From findings published in The Lancet Digital Healtha research group led by Associate Professor Daiju Ueda and Professor Yukio Miki at the Osaka Metropolitan University Graduate School of Medicine has developed an artificial intelligence model that can estimate lung function from chest X-rays with high accuracy.

Conventionally, lung function is measured using a spirometer, which requires the cooperation of the patient, who is given specific instructions on how to inhale and exhale into the instrument. Accurate evaluation of measurements is difficult if the patient has difficulty following instructions, which may occur in infants or persons with dementia, or if the person is predisposed.

Professor Ueda and the research group trained, validated and tested the AI ​​model using more than 140,000 chest x-rays over a period of almost twenty years. They compared the actual spirometric data with the AI ​​model’s estimates to fine-tune its performance. The results showed a remarkably high level of agreement, with a Pearson’s correlation coefficient (r) of more than 0.90, indicating that the method is sufficiently promising for practical use.

The AI ​​model developed in this study has the potential to expand lung function assessment options for patients who have difficulty performing spirometry.

“Very significant is the fact that our method, just by using the static information of chest X-rays, suggests the possibility of accurately estimating lung function, which is normally evaluated through tests in which patients exert physical energy have to exercise,” explains Professor Ueda.

“This AI model was built thanks to the cooperation of many people, from doctors, researchers and technicians to patients in different institutions. If it can help reduce the burden on patients while lowering medical costs, that would be great.”