[Skip to Content]
Sign In
Individual Sign In
Create an Account
Institutional Sign In
OpenAthens Shibboleth
Purchase Options:
[Skip to Content Landing]
Views 2,160
Citations 0
Biotech Innovations
February 11, 2020

Artificial Intelligence Improves Breast Cancer Screening in Study

JAMA. 2020;323(6):499. doi:10.1001/jama.2020.0370

An artificial intelligence (AI) system for breast cancer screening outperformed radiologists in a recent study in Nature. The technique spotted more cancers and raised fewer false alarms.

Researchers at Google Health and collaborators developed a deep-learning AI model for identifying breast cancer using screening mammograms from 2 large UK and US data sets. The test sets, which were not used to train or tune the system, included scans from 25 856 women at 2 screening centers in England and 3097 women at a US academic medical center.

Small, irregular mass with associated microcalcifications that US radiologists missed but AI correctly identified.

Northwestern University

The researchers evaluated the AI system’s cancer predictions and clinical radiologists’ original decisions based on biopsy-confirmed breast cancer outcomes. In the US data set, the system produced 5.7% fewer false-positives and 9.4% fewer false-negatives than radiologists. The system also performed better on average than 6 US board-certified radiologists in a separate comparison involving 500 randomly selected mammograms from the US test set. Notably, most of the cancers identified only by the AI system were invasive.

Two or more readers—the counterpart of US radiologists—interpret mammograms in the United Kingdom. In the UK data set, the AI system outperformed the first reader for specificity, resulting in fewer false-positives. It was noninferior for sensitivity but showed a tendency toward fewer false-negatives, the authors said in an email. The system performed on par with the second reader and with consensus judgments.

“While this is exciting, early-stage research, validation in future trials is needed to better understand how models like these can be effectively integrated into clinical practice,” coauthor Mozziyar Etemadi, MD, PhD, of Northwestern Medicine in Chicago, said in a statement.