Mammography is viewed as the gold standard in early detection of breast cancer, but interpretation of the screening data can be challenging and result in high rates of false positives and false negatives. An artificial intelligence (AI) system has been developed by researchers from the U.S. and U.K. for this diagnostic purpose and demonstrated to outperform expert radiologists in accurately interpreting mammograms from screening programs.

The AI algorithm was trained using mammograms for almost 29,000 patients in the U.S. and U.K. and used to identify the presence of breast cancer in mammograms of women who were known to have had either biopsy-proven breast cancer or normal follow-up imaging results at least 365 days later.

When applied to the data for U.S. patients, software performed significantly better than human experts, producing 5.7% fewer false positive diagnoses and 9.4% fewer false negatives. The program also outperformed radiologists interpreting mammograms for U.K. patients, generating 1.2% fewer false positives and 2.7% fewer false negatives.

The AI approach was then pitted against the double reading process used by the U.K. National Health Service (NHS) in which scans are interpreted by two separate radiologists. The assessment compared the AI’s decision with that of the first reader, and scans were submitted to a second reviewer if there was a disagreement between the first reader and the AI. The system was shown to perform comparably to its human counterparts and to reduce the workload of the second reviewer by as much as 88%.

The system developed by researchers from Google Health, DeepMind, Imperial College London, the NHS and Northwestern University offers scope for improving the accuracy and efficiency of breast cancer screening.