A new artificial intelligence system of Google Health -subsidiary of Google in the field of “smart” health- can study X-rays of lungs and draw conclusions as reliable as those of experienced radiologists. Google’s deep learning system, according to its creators, can be a valuable assistant to physicians in the future.
The researchers -who published the study in the journal «Radiology» of the Radiological Society of North America- trained their system by feeding it with more than 860,000 X-rays. Lung X-rays are among the most commonly performed tests worldwide, but have some limitations.
”We found that the subjective factor in the interpretation of this X-ray, which limits its effectiveness, is greatly involved,”
Google Health’s chief researcher Sravia Ceti said.
”Lung x-rays are often interpreted in a subjective way, something problematic from the perspective of deep learning. We have managed to produce a system that makes a more reliable evaluation”,
said his colleague, Daniel Che.
The tests of the system by the Committee of Radiologists showed that it manages at least as well as the doctors in detecting various problems (diagnosis of fractures, nodules, tumors, pneumothorax, etc.).