I recently saw an article about AI diagnosis of chest X-rays. Has anyone actually used it in clinical practice? I've heard the accuracy is quite high, but I'm also worried about the possibility of misdiagnosis.
I agree, I was curious too lol. It's a bit iffy to just rely on the article.
AI diagnostic accuracy itself is high, but its clinical application remains controversial. In particular, there are studies suggesting that it sometimes misses abnormal findings.
We piloted it at our hospital, and it picked up things like tuberculosis pretty well. But for lung nodules, its sensitivity was lower than that of a radiologist. Still, it seems decent as a supplementary tool.
But even if the accuracy is high, if the training data is mostly Westerners, there can be errors when applying it to Koreans. I've heard that such cases have actually been observed. So I think we still need to be cautious about it.