Breast cancer screening for women is absolutely essential, One in eight women in the U.S. will have breast cancer in their lifetime, according to the American Cancer Society. What’s more, the breast cancer incidence rates have increased by 0.3% annually in recent years. The fact is that it is difficult to interpret mammography images accurately, with frequent false positives and false negatives. We’ve written about developments using artificial intelligence (AI) to detect breast cancer at the Karolinska Institute in Stockholm, by a research team from New York’s Rory Myers College of Nursing, and a group from MIT, Havard Medical School, and Massachusetts General Hospital.

Google Health recently published a paper in Nature that reveal the initial findings of a two-year study. The company worked with partners in the U.S. and U.K. to use deep learning to improve breast cancer detection. In this first study, the experts trained the deep learning model using mammograms from more than 91,000 women in the U.S. and U.K. After training the system, the group ran subsequent testing with images from 28,000 American and British women that were not part of the initial training set. The researchers reported their AI model was more accurate than expert radiologists, while producing fewer false positives and false negatives. The researchers also noted that the AI model had only the x-ray images to work from while the human experts were able to access earlier x-rays and medical histories. This indicates that the neural network mode didn’t need as much data to perform at higher accuracy rates than physicians.

The next steps for Google Health and its international partners include further testing, additional research, and regulatory approvals. This study is yet another in the growing collection of clinical trials that demonstrate the potential benefits that could come from the use of AI in medical image screening and diagnosis.