Health and Science

Google's DeepMind A.I. beats doctors in breast cancer screening trial

Key Points
  • Anonymous scans of 29,000 women were used in the trial.
  • The biggest improvements over human scanning was found in the U.S. portion of the study.
  • Google-owned DeepMind has already used AI to read eye scans and spot neck cancer.
A woman undergoes a mammogram exam.
Media for Medical | Getty Images

Artificial Intelligence (AI) powered by Google's DeepMind algorithm may be more accurate at spotting breast cancer than real life doctors.

The findings, published in on Wednesday, come after researchers from Imperial College London and Google Health "trained" a computer to spot abnormalities on X-ray images of nearly 29,000 women.

Separate studies used imagery from U.K. and U.S. women and concluded that in both countries the computer reduced instances where a cancer was either incorrectly identified or incorrectly missed.

In the United States, the improvement was more noticeable — offering a reported reduction of 5.7% in false positives, where a mammogram is wrongly diagnosed as abnormal.

There was also a reduction of 9.4% in false negatives, where a cancer is missed.

"In an independent study of six radiologists, the AI system outperformed all of the human readers," claimed the report.

Unlike the U.S., the National Health Service (NHS) in Britain typically employs two radiologists to screen images. Where there is disagreement, a third will make a judgment.

But even when two human doctors worked together on the same images, the AI computer was found to be still at least as effective at finding the correct diagnosis. It is expected the system could improve further if it is pre-loaded with complimentary patient history that is typically available to doctors.

"Our team is really proud of these research findings, which suggest that we are on our way to developing a tool that can help clinicians spot breast cancer with greater accuracy," Dominic King from Google Health told the BBC.

In the U.K., it is hoped that the system could alleviate some of the backlog arising from the shortfall of radiologists. In April last year, The Royal College of Radiologists claimed that three quarters of radiology bosses said they didn't have enough consultants to deliver safe patient care.