Almost one-quarter of a million new cases of breast cancer were diagnosed in 2015, the last year for which the CDC has data. In that year 41,523 women died of breast cancer in the U.S. The CDC states that for every 100,000 women, 125 new breast cancer cases were reported in 2015 and 20 women died of cancer. We’ve written about a variety of tech used to detect breast cancer such as research at Ben-Gurion University of the Negev detecting breast cancer through breath and urine. Also, we’ve looked at New York’s Rory Myers College of Nursing study of the use of AI to detect breast cancer survivor’s lymphedema.
A research group from MIT, Harvard Medical School, and Massachusetts General Hospital (MGH) recently reported on the use of an automated artificial intelligence deep-learning model to assess dense breast tissue in mammograms. The group published the results of the study in Radiology. Tissue density is a known risk factor for breast cancer. According to the researchers, more than 40% of women in the U.S. have dense breast tissue. The density can make mammogram screening harder, and now 30 U.S. states require that radiologists inform women whose mammograms show they have dense breast tissue. The Boston-based group trained a deep-learning model with 41,000 high-quality digital mammograms with expert assessments. They then tested the system with 8,600 additional images. The bottom line was 94% agreement on a binary density determination and 90% using a four-quadrant rating. The researchers also reported that the AI assessments took less than one second per image. As noted in the MIT CSAIL news report, “MGH is a top breast imaging center with high inter-radiologist agreement, and this high-quality dataset enabled us to develop a strong model.”
This is yet another large-scale study of machine learning diagnostics that we’ve covered. We also wrote about AI detecting melanoma with equal to better accuracy than a large group of expert dermatologists. The healthcare world isn’t close to turning medical diagnosis over entirely to algorithms, but using AI for triage screening or second opinions may soon become a common, acceptable practice.