A study, published in the January 2020 issue of the scientific journal Nature, found that by using A.I. technology, there was a reduction in false positives and false negatives when it came to diagnosing forms of breast cancer. Besides skin cancers, breast cancer is the most common cancer in American women, with a 13% chance that a woman will develop breast cancer sometime in her life. And while breast cancer is the second leading cause of cancer death in women (lung cancer kills more women each year), the chance that a woman will die from breast cancer is around 2.6%.
In the US, clinicians fail to catch about a fifth of all breast cancer cases, and half of the women receiving annual mammograms in a given decade will be incorrectly told they have breast cancer.
Digital mammography, or X-ray imaging of the breast, is the most common method to screen for breast cancer and intends to identify breast cancer at earlier stages of the disease when treatment can be more successful. However, identifying and diagnosing breast cancer still faces many challenges and can often result in both false positives and false negatives. False positives can lead to patient anxiety, unnecessary follow-up, and invasive diagnostic procedures. So in collaboration with partners at DeepMind, Northwestern University, Cancer Research UK Imperial Centre, and Royal Surrey County Hospital, Google set out to examine if AI could assist radiologists to spot the signs of breast cancer more accurately.
What’s remarkable about the published outcomes is how Google’s AI system didn’t have access to patient histories and prior mammograms, like doctors would ordinarily use. The model was trained from de-identified mammograms of 15,000 women in the US and 76,000 women in the UK. Evaluation of the assessment showed that Google’s AI model was able to decrease false positives by 5.7% for women in the U.S. and 1.2% for those in the U.K. False negatives were reduced 9.4% in the U.S. and by 2.7% in the U.K.
Reading mammograms is the perfect problem for AI and machine learning. AI excels when it has to do the same task over and over again and has to find the one thing that might appear one time out of 10,000. But I honestly did not expect it to work this much better [than doctors]. I was surprised. – Dr. Mozziyar Etemadi, Research Assistant Professor of Anesthesiology and Biomedical Engineering at Northwestern University/ Co-Author of published study
The optimal use of the AI system within clinical workflows remains to be determined. Beyond improving reader performance, this robust assessment of the AI system advances a new way for clinical trials to improve the accuracy and efficiency of breast cancer screening. With continued research, future applications could likely improve the accuracy and efficacy of screening programs, as well as decrease wait times and anxiety for patients. The study has some limitations, with most of the tests being performed on the same type of imaging equipment.
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. This sets the stage for future applications where the model could potentially support radiologists performing breast cancer screenings. – Daniel Tse, M.D., Product Manager, Google Health