An artificial intelligence program designed by Google Health and assessed in partnership with academic cancer researchers reduced both false positive and false negative results in assessments of mammograms for signs of breast cancer compared with radiologists, CNN Business reports.
Publishing their findings in the journal Nature, a research team led by Scott Mayer McKinney, MD, of Google Health fed mammograms of tens of thousands of women from the United States and the United Kingdom into a program designed to detect breast cancer.
Also participating in the research was the Cancer Research U.K. Imperial Centre, Northwestern University and the Royal Surrey County Hospital.
Compared with assessments conducted by radiologists, the computer programs reduced the rate of false positives (detecting cancer that is not actually present) by 5.7% for the U.S. women and by 1.2% for the U.K. women. The program reduced false negatives (failing to detect cancer that is present) by 9.4% for U.S. women and 2.7% for U.K. women.
This greater level of accuracy on the part of the program held even though it had access to less information than the doctors, including patient histories and previous mammogram results.
According to the American Cancer Society, mammograms fail to identify about 20% of breast cancers.
The researchers also conducted a simulation in which the AI program participated in the U.K. system of having two readers for each mammogram. The program had a comparable level of accuracy in this context and reduced the workload of the second reader by 88%. This finding suggests that the program could help address the critical shortage of radiologists in the United Kingdom.
According to a 2018 report from the Royal College of Radiologists, the United Kingdom needs an additional 2,000 radiologists by 2023 to close the current deficit.
“This robust assessment of the AI system paves the way for clinical trials to improve the accuracy and efficiency of breast cancer screening,” the study authors concluded.
To read the CNN Business article, click here.
To read the study abstract, click here.