It is established that early detection of any cancer is helpful towards eliminating it efficiently. However, sometimes the cancer is harder to find, and it takes radiologists longer to detect, which can impact the time-sensitive treatment that must be given to a patient.?
However, now Google¡¯s AI system can detect breast cancer through mammograms faster than radiologists, helping in eliminating errors and speeding up the healing process.
According to the American Cancer Society, on average, radiologists are known to miss around 20 percent of breast cancer in mammograms with half of all the women who undergo screenings (over a 10-year period) have a false-positive result.?
For this study, researchers from the US¡¯s Northwestern Medicine in Chicago, UK¡¯s Imperial College London and Britain's National Health Service have taken help of Alphabet¡¯s DeepMind AI that recently fused with Google Health in September.? The AI is capable of accurately detecting early signs of breast cancer by looking at mammograms.
UK¡¯s teams helped in training the AI to identify thousands of mammograms. They tested it against a set of 25,856 mammograms from the UK and 3097 from the US. The AI system was able to detect cancer as good as an expert radiologist while dropping the ratio of false positives by 5.7percent in the US-based set and 1.2 percent in the British-based set.
The AI also brought the number of false negatives by 9.4 percent for the US-based set and 2.7 percent by the UK-based set.
The massive difference in the numbers is due to the different methods in which mammograms are read. In the US, only one radiologist reads the results with tests happening every one to two years. In the UK however, the tests are every three years and is read by two radiologists. When both disagree -- a third one is consulted.?
They even tested the AI against six expert radiologists and discovered that the AI offered far better accuracy in detecting breast cancers.
While computer-aided detection isn¡¯t new, earlier they weren¡¯t as accurate as expert radiologists as tehy they were trained to look for things humans could. The AI system, on the other hand, learns to spot cancers depending on actual results of thousands of mammograms, creating an unparalleled dataset.?
The system, however, is far from perfect as they¡¯ve used the same type of imaging equipment and the US group had a lot of confirmed breast-cancer patients in their data set. Moreover, the test is still far from getting regulatory approval.?