An artificial intelligence model developed by Google and other entities was able to identify breast cancer in screening mammograms with greater accuracy, fewer false positives, and fewer false negatives than human experts.
In a recent blog post, Shravya Shetty, MS, technical lead at Google Health, and Daniel Tse, MD, product manager at Google Health, noted that digital mammography, the most common way to screen for breast cancer, is often challenging for providers.
“Reading these X-ray images is a difficult task, even for experts, and can often result in both false positives and false negatives. In turn, these inaccuracies can lead to delays in detection and treatment, unnecessary stress for patients and a higher workload for radiologists who are already in short supply,” the authors wrote.
A more efficient method of detecting breast cancer is necessary, as the disease affects a significant number of women around the world. About one in eight women in the US will develop breast cancer in their lifetime, and more than 55,000 women in the UK are diagnosed with the disease each year, Shetty and Tse said.
Over the last two years, researchers at Google have partnered with teams at DeepMind, Cancer Research UK Imperial Centre, Northwestern University and Royal Surrey County Hospital to develop an artificial intelligence model that can improve breast cancer detection.
Researchers evaluated the model using a set of de-identified mammograms from more than 3,000 women in the US and more than 25,000 women in the UK. The AI model produced a 5.7 percent reduction of false positives in the US, and a 1.2 reduction in the UK. The model also produced a 9.4 percent reduction in false negatives in the US and a 2.7 percent reduction in the UK.
Researchers also set out to see if the model could generalize to other healthcare systems. The team trained the model only on the data from women in the UK, and then evaluated it on the dataset from women in the US. The model achieved a 3.5 percent reduction in false positives and an 8.1 percent reduction in false negatives. This demonstrates the model’s potential to generalize to new clinical settings while still outperforming experts.
Shetty and Tse also pointed out that the model received less information than human experts did when making its decisions.
“The human experts (in line with routine practice) had access to patient histories and prior mammograms, while the model only processed the most recent anonymized mammogram with no extra information,” researchers said.
“Despite working from these X-ray images alone, the model surpassed individual experts in accurately identifying breast cancer.”
This study builds on Google’s past efforts to test the use of AI in diagnosing and detecting breast cancer. In October 2018, the company developed a deep learning tool could identify metastasized breast cancer with 99 percent accuracy, reducing the time it takes clinicians to review pathology slides.
The team expects that with further research and development, their model could improve cancer care and detection.
“Looking forward to future applications, there are some promising signs that the model could potentially increase the accuracy and efficiency of screening programs, as well as reduce wait times and stress for patients,” the authors concluded.
“But getting there will require continued research, prospective clinical studies and regulatory approval to understand and prove how software systems inspired by this research could improve patient care. We’re looking forward to working with our partners in the coming years to translate our machine learning research into tools that benefit clinicians and patients.”