Researchers in the US developed a new artificial intelligence algorithm, which manages better than experienced radiologists in detecting small cerebral bleeding. This development is expected to help physicians in the future to diagnose and better treat patients with head injuries, strokes and aneurysms.
The new system can relieve radiologists from a heavy workload, as artificial intelligence technology will first distinguish imaging tests with abnormal findings and then doctors will examine them more carefully.
The physicians and computer engineers of the Universities of California-San Francisco and California-Berkeley, headed by Assistant Professor of Radiology Esther Yach, published this in the journal of the US National Academy of Sciences (PNAS).
The algorithm -a fully co-revolutionary neural network trained in 4,400 pre-existing tomographs- takes only one second to judge if there is any bleeding in the entire head. It can even find invisible abnormalities that escape the attention of a radiologist, classifying them according to the type of bleeding.
“We wanted something practical and clinically useful, with a level of accuracy close to perfect. The requirements for such an application are high due to the potential consequences of an escaping anomaly. People will not tolerate anything less than a human level of performance or accuracy.”
Dr. Yach said.
”A 95% accuracy rate on an X-ray or even 99% is not okay. On the other hand, if the computer displays many false positive results, it will confuse and slow down the radiologist, possibly leading to more errors”,
The ability of artificial intelligence to detect even invisible bleeds is a major advance.
“The bleeding can be tiny, though significant. This makes the radiologist’s job so difficult. If a patient has an aneurysm and starts bleeding, but you send him home, he can die.”
radiology professor Pratik Mukherjee said.