The Silicon Review ^Asia

Emergency departments usually depend on CT scans to identify life-threatening conditions like acute intracranial hemorrhage (ICH), but sometimes because of the low-resolution images, even the most experienced radiologists are known to miss subtle signs of the condition. Recently, researchers from the University of California, San Francisco, and the University of California, Berkeley, have come up with a deep learning AI that can detect and provide an expert-level reading of ICH. The AI can perform on par with a standard radiologist, and it can find tiny brain bleeds. To identify the abnormalities, the researchers made use of end-to-end, single-stage, fully conventional deep learning networks.

The abnormalities were manually highlighted at the pixel level,and the algorithm used was tested on a data set of over 4000 CT exams. The researchers compared the results of the algorithm’s performance with four experienced radiologists who were certified by the American Board of Radiology, and they found that the algorithm was able to identify ICH successfully. The test contained 200 head CT exams, and the algorithm was able to achieve 100 percent sensitivity. Furthermore, the abnormalities ignored by the experts were identified by the algorithm. The tests concluded with researchers giving a nod to use the algorithm in an ER by a radiologist to detect ICH.