Machine learning could help reduce false positives from mammography screening

May 8 2017

Machine learning (or coding) could help reduce false positives from mammography screening, according to an article study published online in the May 4, 2017 issue of the Journal of the American Medical Association (JAMA) Oncology. The national coding competition known as the DREAM Challenge, launched during the inauguration of Vice President Biden's Cancer Moonshot Challenge, may help mitigate this harm associated with routine screening. "Women want to know that they can rely on mammography screening to deliver accurate information regarding their breast health status," says Dr. Lee, a sub-specialty-trained breast imager at the SCCA, Associate Professor of Radiology at the University of Washington School of Medicine, and Faculty Investigator at the Hutchinson Institute for Cancer Outcomes Research. "False positive findings detected at screening mammography can cause women anxiety and lead to unnecessary biopsies, and we are looking to change that through this national competition."

Breast cancer is the most frequently diagnosed solid cancer and second leading cause of cancer death among U.S. women, and mammography is known to catch breast cancer earlier and help save lives. Despite this benefit, routine mammography is associated with a high-risk of false positive results and may lead to over-diagnosis of clinically insignificant lesions. The article points out that of every 1,000 women screened, only 5 will have breast cancer. Yet, 100 out of every 1,000 women screened will be recalled for further testing. The challenge aims to mitigate this potential harm.

False-positive mammography results may lead to unnecessary anxiety, biopsies, surgery, and treatment. Expert radiologists, with additional sub-specialization in breast imaging, can provide very accurate interpretations, but still need to call back about 10% of women for additional testing. With recent technological advances in image analysis and the growth of digital health data, newer machine learning techniques that combine statistical models with imaging variables hold the potential to better predict patient outcomes. The Digital Mammography DREAM Challenge launched as part of the national Cancer Moonshot Initiative, aims to accelerate big data analytics to improve breast cancer screening outcomes. A collaboration between SCCA, Sage Bionetworks, and the open coder community, it is one of several Coding4Cancer challenges. The $1.2 million prize competition, which concludes at the end of the summer, gives coders access to more than 640,000 de-identified digital mammography images from more than 86,000 women, including their demographic characteristics, medical history, and clinical data. Over 1,100 teams worldwide will compete with the goal of developing a computer algorithm that can accurately detect cancer on mammograms while decreasing the overall false-positive rate.