The artificial intelligence system was better at analyzing mammograms than radiologists. None of the six specialists participating in the experiment won with it, the researchers said in the journal Nature.
Mammography is a test that detects early, potentially serious changes in the breast. A quick diagnosis means a much better chance of recovery. However, as researchers from Google point out, false positive results quite often appear in this method and the actual disease is skipped.
Scientists from the IT group in the magazine “Nature” described artificial intelligence, which managed to make significantly fewer mistakes than specialists.
With the help of a computer program, the researchers analyzed mammographic images of two large groups of patients – from the USA and Great Britain. Compared to the analysis carried out by doctors, the number of false positive results decreased by 5.7 and 1.2%, respectively, and negative – by 9.4 and 2.7.
In another experiment, involving six radiologists, the AI system did better than each of them.
In turn, in a study simulating a typical diagnostic process, with analysis performed by two doctors, the computer reduced the amount of work of the second radiologist by as much as 88 percent.
The creators of the program are counting on starting clinical trials with the belief that the system will improve the accuracy of mammography in real conditions.