In a study published in Nature, researchers from Google and medical centers in the United States and Britain report that artificial intelligence (A.I.) could help doctors in diagnosing breast cancer on mammograms.
- According to the American Cancer Society, 268,600 new cases of invasive breast cancer and 41,760 deaths were expected among women in the United States last year.
- Globally, there are about 2 million new cases each year, and more than half a million deaths.
- In the United States, around 33 million screening mammograms are performed each year, and about 20 percent of breast cancers are missed during the test.
- Using computers trained to recognize patterns and interpret images, Google created an A.I. system that can read mammograms, which are X-rays of the breast used to find breast cancer.
- Their system was tested on images where the diagnosis was already known.
- In one test, the researchers used mammograms from about 25,000 women in Britain, and 3,000 in the United States, to compare the system’s performance with that of the radiologists who had originally read the X-rays.
- They found that A.I. was more accurate than the radiologists, reducing both false negatives (when a mammogram is mistakenly read as normal and cancer is missed) and false positives (where the scan is incorrectly read as abnormal but there is no cancer).
- In another test, they presented 500 mammograms to be interpreted by A.I. and six radiologists. Once again, the A.I. system outperformed the humans.
- However, there were cases where A.I. missed a cancer that all six radiologists has found, and vice-versa.
- The next step in the research would be to have radiologists test the use of this A.I. tool as part of the routine practice in reading mammograms, with the hope of enhancing human performance and reducing the occurrence of false positives and false negatives.