Researchers at Seoul National University Hospital have confirmed that artificial intelligence (AI)-based diagnoses system can diagnose abnormalities in chest x-rays with higher accuracy.
|Professor Park Chang-min|
The team, led by Professors Park Chang-min and Hwang Eui-jin at the hospital, tested the AI diagnostic system's ability to read chest X-rays on 1,135 patients who visited the emergency room from January to March 2017.
The reading sensitivity was 82 to 89 percent, higher than the on-call doctor’s reading. The reading sensitivity also improved when the on-call doctor made a diagnosis based on the results of the AI diagnosis system.
Until now, on-call doctors have had to observe the chest X-rays for themselves. However, the reading sensitivity of the on-call radiologist was 66 percent, and it took a median of 88 minutes for the doctors to read the images.
Images with abnormal findings, which required further examination or treatment, took longer with a median of 114 minutes for the physicians to confirm the results. Therefore, patients had to wait for one to two hours to get the image reading result.
With the use of an AI diagnostic system in an emergency room, however, the team expects that it will improve the delay of treatment by reducing the reading errors and time required.
"The significance of this study is that it demonstrated the possibility of using the AI diagnosis system in the actual medical field," the team said. "The performance of AI diagnostic aids using has been reported several times in other studies."
However, as most studies confirm the performance of the device only with experimental data, it was not sure if the devices could be used in actual clinical practice, the team added.
Professor Park said, "The study will be a milestone in demonstrating that AI can be fully utilized in actual patient care. In the future, we will continue to study ways to maximize the utilization of AI in developing more advanced AI systems and verify their performance."
The journal Radiology published the result of the study.