This AI tool is great at spotting abnormalities in chest X-rays
We previously reported about an AI that detected lung diseases by going through X-ray images of patients. Well, now we have now another exciting AI program that can differentiate between a normal and abnormal chest X-ray.
Identifying an abnormal chest X-ray is important as it lets a doctor know if a patient is suffering or likely to suffer from lung cancer, tuberculosis, emphysema, pneumonia, heart failure, and various other chest-related health issues.
The researchers claim this new AI tool could speed up the process of diagnosing chest and lung problems. Plus, its accuracy is at par with clinical board-certified radiologists (doctors who use imaging techniques like x-rays to diagnose and treat their patients).
A report from the Association of American Medical Colleges (AAMC) reveals that hospitals in the US may have up to 41,000 fewer radiologists on staff than necessary by 2033
“There is an exponentially growing demand for medical imaging, especially cross-sectional such as CT and MRI. Meanwhile, there is a global shortage of trained radiologists,” said Dr. Louis Lind Plesner, one of the study authors, in the press release.
The new AI can assist health professionals, reduce their workload, and at the same time, it could help hospitals prevent the problems that could arise due to the shortage of radiologists.
Testing the AI on real patients
The claims made by Dr. Plesner and his team are backed by the findings from an experiment involving 1,529 patients. This large pool incorporated outpatients (people not in the hospital but were taking treatment), in-hospital patients, and emergency patients.
During the study, the researchers made a commercially available AI program to go through the chest x-rays of all these patients. The AI assorted some x-rays as “high-confidence normal” (normal) and the rest as “not high-confidence normal” (abnormal).
What’s more interesting is that its performance was compared against three medical board-certified chest radiologists.
The study authors noticed that the AI program was highly sensitive to abnormal x-rays. It was able to spot such x-rays with a sensitivity of 99.1 percent. Dr. Plesner highlighted, “The most surprising finding was just how sensitive this AI tool was for all kinds of chest disease. In fact, we could not find a single chest X-ray in our database where the algorithm made a major mistake.”
The only negative finding from the AI was that it identified the chest x-ray of a 44-year-old patient with pneumonia as normal.
He further added, “The AI tool had a sensitivity overall better than the clinical board-certified radiologists.” Moreover, the tool also turned out to be great at detecting normal x-ray images in the case of outpatients.
The rate at which it spotted normal x-rays in outpatients stood at 11.6 percent. According to human radiologists, there were 429 normal chest x-rays overall. The AI program also successfully classified 120 (of the same 429) such x-rays as normal.
Based on these findings, the researchers suggest that their AI tool could have been employed to automatically check nearly eight percent of all the x-rays they examined during the study. This is important because “even a small percentage of automatization can lead to saved time for radiologists, which they can prioritize on more complex matters,” said Dr. Plesner.
They believe the AI tool is amazing at spotting abnormal x-rays and has the potential to speed up the chest diagnosis process at hospitals. However, further studies are required to test this program on a much large scale and improve its performance.
The study is published in the journal Radiology.
The new book “Climate Change and Human Behavior” bridges the gap by explaining how a warming planet increases aggression and violence.