Google’s artificial intelligence proved more accurate in identifying breast cancer from mammograms than doctors, according to a study published in Nature yesterday.

The initial results of Google’s AI for breast cancer screening, found that the system was able to reduce the number of both false positives and false negatives in breast cancer detection. The study was born out of a collaboration between DeepMind (an Alphabet-owned company), Cancer Research UK Imperial Centre, Northwestern University and Royal Surrey Country Hospital. 

“These findings show that our AI model spotted breast cancer in de-identified screening mammograms (where identifiable information has been removed) with greater accuracy, fewer false positives, and fewer false negatives than experts,” Shravya Shetty, technical lead at Google health, and Dr. Daniel Tse, product manager at Google health, wrote in a blog. “This sets the stage for future applications where the model could potentially support radiologists performing breast cancer screenings.”

TOPLINE DATA 

Researchers found that the AI was able to reduce the percentage of false negative mammograms by 5.7% in the US and 1.2% in the UK. The study also showed a reduction in false negatives by 9.4% in the US and 2.7% in the UK. 

In the UK the practice is for two doctors to read each mammogram, so here the UK results were compared to the initial mammogram reader. 

“Consensus double reading has been shown to improve performance compared to single reading, and represents the current standard of care in the UK and many European countries,” authors wrote. “Our system did not outperform this benchmark, but was statistically non-inferior to the second reader and consensus opinion.”

Meanwhile in the US, the readings are performed by one radiologist every other year. The US results were compared to the Breast Imaging Report and Data System or BI-RAD score “assigned to each case in the original screening context as a proxy for human cancer prediction.”

Authors of the investigation also conducted an independent "comparison with a reader study" of six US-board certified radiologists, and found that AI “exceeded the average performance of radiologists by a significant margin.” 

HOW IT WAS DONE

Researchers trained the AI system on de-identified mammograms from over 76,000 women in the UK and 15,000 women in the US. 

During the study the system was evaluated on a different set of de-identified data that came from 25,000 women in the UK and over 3,000 women in the US. 

THE LARGER TREND 

Breast cancer is the most common type of cancer in women, with a rate of one in eight women in the US developing the condition at some point in her lifetime, according to the National Breast Cancer Foundation. Early detection and treatment is linked to better survival rates. 

In both the US and UK there is a shortage of trained radiologists — making the push for new tools to lighten doctor’s workload in higher demand. However, stakeholders warn the tools must be properly test in the future. 

Google, meanwhile, has been developing AI applications across a range of conditions. The past few years have seen investigations of its tech involving the detection of diabetic retinopathy, heart conditions and more. 

IN CONCLUSION 

“McKinney and colleagues’ results suggest that AI might someday have a role in aiding the early detection of breast cancer, but the authors rightly note that clinical trials will be needed to further assess the utility of this tool in medical practice,” Etta D. Pisanoof the American College of Radiology, wrote in the accompanying commentary to the study. “The real world is more complicated and potentially more diverse than the type of controlled research environment reported in this study.”