Health

Google Health is Using AI To Improve Breast Cancer Mammograms

In collaboration with researchers in the US and the UK, the technology helps radiologists to decrease false negatives and false positives when screening for breast cancer.

07.01.2020 | by Reve Fisher
Photo by marijana1 from Pixabay
Photo by marijana1 from Pixabay

Breast cancer is the second most common cancer diagnosis in the world and the first among women, according to the World Cancer Fund. Mammograms play an essential role in early breast cancer detection and minimising deaths from the disease. However, false positives and false negatives are still a problem, leading to stress for patients, delays in treatment and an increased workload for radiologists and other health professionals. 

In collaboration with top clinical researchers in the US and UK, Google Health has set out to determine if artificial intelligence could help alleviate this issue.

Over the past two years, Google Health researchers have worked with DeepMind, a UK-based artificial intelligence startup purchased by Google in 2014; Cancer Research UK Imperial Centre; Northwestern University in Evanston, Illinois, USA; and Royal Surrey County Hospital. The team developed an AI-based model designed to detect signs of breast cancer in mammogram scans. The model was trained using a database of over 76,000 women in the UK and 15,000 in the US and evaluated on more than 25,000 UK women and 3,000 US women.

The results, which have been published in Nature, demonstrate that the AI model may be more capable of detecting breast cancer than a trained human expert—and with less information. The human experts had access to patient histories and previous mammograms, whereas the AI model only had the current anonymised mammogram with no extra information about the patients or their prior health exams.

In the evaluation, the AI-based system reduced false positives by 5.7 percent in the US and 1.2 percent in the UK. It also reduced false negatives by 9.4 percent in the US and 2.7 percent in the UK.

As the US and the UK have very different healthcare systems, the researchers performed a separate experiment to compare results between countries. Using a database of over 25,000 UK women to train the model and 3,000 US women to evaluate it, the system found a 3.5 percent reduction in false positives and an 8.1 percent reduction in false negatives.

“This is another step along the way of trying to answer some of the questions that will be critical for us to actually deploying this in the real world,” Dominic King, director and UK lead of Google Health, said in a statement. “This is another step closer to trying to deploy this type of technology safely and effectively.”

Join the discussion

Leave a Reply

Related Shakers

Phil Bevan

International…

Geoff McDonald

Minds@Work

Related Shakers