An artificial intelligence (AI) program designed by Google and collaborators has proved to be better than radiologists at detecting breast cancer in mammograms, according to researchers in the United States and Britain.
Image Credit: Shutterstock.com/Syda Productions
The algorithm, which was developed by an international team, including researchers from Google Health and Imperial College London, outperformed six radiologists by spotting cancers that the radiologists missed, while ignoring features that they falsely flagged as potential signs of cancer.
The technology was designed and trained in collaboration with DeepMind, Cancer Research UK Imperial Centre, Northwestern University, and Royal Surrey County Hospital.
The study, recently published in the journal Nature, is the latest to demonstrate that AI could improve the accuracy of breast cancer screening and if it proves effective in clinical trials, it could be used to ease the burden on services such as the NHS where radiologists are in short supply.
This is a great demonstration of how these technologies can enable and augment the human expert.
Dominic King, UK lead, Google Health
King continued, “The AI system is saying ‘I think there may be an issue here, do you want to check?’”
Breast cancer statistics
About one in eight women are diagnosed with breast cancer globally. In England alone, screening detects more than 18,000 cases per year. However, tumors are still missed, with some women receiving false negatives, while others are incorrectly suspected of having cancer due to false positives.
The American Cancer Society says radiologists fail to detect cancer in mammograms in around one-fifth of cases and that more than 50% of women who receive screening over 10 years receive a false positive.
What did the study involve?
For the study, the researchers trained the algorithm to spot breast cancers in mammograms from more than 76,000 women in the UK and about 15,000 women in the US.
They then asked it to assess 25,856 new mammograms from women in the UK and 3,097 from women in the US, who either had biopsy-confirmed breast cancer or no signs of cancer during follow up at least one year later.
The results showed that the AI system could detect cancer with a similar degree of accuracy to the radiologists, but that it also reduced the number of errors, with false positives reduced by 5.7% in the US group and by 1.2% in the UK group.
It also reduced the number of false negatives, by 9.4% in the US group, and by 2.7% in the UK group.
Differences in screening between US and UK
These differences in error reduction reflect differences between the United States and the UK in how the mammograms are read.
In the US, women tend to be screened every one to two years and one radiologist only examines the results. In Britain, women are screened every three years and the results are examined by two radiologists, with a third radiologist consulted if the first two disagree.
The development could be used to improve screening
The findings suggest that AI could improve the accuracy of screening in the US and maintain a similar level of quality in the UK, with the technology being used to assist or replace a second radiologist.
"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,” says King.
King continued, "Further testing, clinical validation and regulatory approvals are required before this could start making a difference for patients, but we're committed to working with our partners towards this goal."
In the UK, breast cancer screening has been placed under particular strain due to a shortage of at least 1,100 radiologists, according to The Royal College of Radiologists. In the case of breast radiology, 8% of hospitals are unfilled, largely because older radiologists are retiring more quickly than new radiologists are joining.
Like the rest of the health service, breast imaging, and UK radiology more widely, is understaffed and desperate for help… AI programs will not solve the human staffing crisis, as radiologists and imaging teams do far more than just look at scans, but they will undoubtedly help by acting as a second pair of eyes and a safety net.”
Caroline Rubin, Vice-president for clinical radiology, The Royal College of Radiologists
Rubin says the next step is for the technology to be used in clinical trials, evaluated in practice and used on patients screened in real-time.
Cancer Research UK’s chief executive, Michelle Mitchell, says that although breast cancer screening helps diagnose cancer in the early stages when treatment is likely to be more effective, it can also fail to detect cancers, or it may flag up cancers that would never have gone on to cause harm.
“This is still early stage research, but it shows how AI could improve breast cancer screening and ease pressure off the NHS,” says Mitchell. “And while further clinical studies are needed to see how and if this technology could work in practice, the initial results are promising," she concludes.
McKinney, et al. (2019). International evaluation of an AI system for breast cancer screening. Nature. https://www.nature.com/articles/s41586-019-1799-6