McKinney, S.M., Sieniek, M., Godbole, V. et al.

International evaluation of an AI system for breast cancer screening.

Nature 577, 89–94 (2020). https://doi.org/10.1038/s41586-019-1799-6

Mammography for breast cancer detection was originally implemented in the 1960’s for screening and detection of breast cancer. In 2011, the Food and Drug Administration (FDA) approved 3D Breast Tomosynthesis which provides higher resolution images of the breast to improve breast cancer detection. Recently published in Nature, Artificial Intelligence (AI) may be the next phase of technology to further improve our ability to detect breast cancers. In the US and UK, 42 million mammography exams are performed per year. Frequently, patients are called back for additional imaging due to concerns on screening mammograms which can cause patient anxiety. Although mammography currently remains gold standard, cancer can sometimes be missed on screening mammograms which can lead to presentation of cancer with more progressed disease.

This study compares radiologist’s performance in reading screening mammograms in the US and UK against AI systems. AI is defined as the development of computer systems with ability to perform tasks normally performed by humans through large amounts of data processing and algorithms that allow software to learn and recognize patterns. This AI system was trained on a large dataset of breast mammogram images from the UK. After a training period, this study by McKinney and colleagues examined the AI system’s performance in reading patients’ mammograms versus the readings of six independent radiologists.

In the US, AI mammogram reading technology exceeded the average performance of radiologists by a significant margin. Of note not all the radiologists in this study had fellowship training in breast imaging. The UK uses a slightly different protocol whereby all breast images are read by two independent radiologists. In comparing the results of AI to the UK protocol, the AI system performed equally as well as the two-radiologist reading system. However, The AI system did not perform better than the single radiologist reading in the US.

Currently, the best use of AI for our patients has not been determined. From this study we understand that the AI system may be able to reduce the number of additional images and biopsies that women are currently receiving. In the US, the AI system may be able to detect cancers earlier and more reliably than currently seen by the average radiologist. This improved detection could lead to improved breast care for our patients. AI technology is an exciting modality that may be capable of detecting cancers earlier and improving the health of our community. It is important for women to have discussions with their doctor regarding the latest technology available to help with breast cancer surveillance.