Breast cancer is one of the most common cancers to affect women. In both the US and UK, it is the most common form of cancer, with around 1 in 8 women developing the disease at some point in their life. When breast cancer is caught early, it is much easier to treat, so regular checks and accurate diagnosis are key. Now, an artificial intelligence (AI) program has proved to be even better than human doctors at recognizing breast cancer from scans.
In the UK, all women over the age of 50 are advised to get a routine breast check called a mammogram, which involves X-raying the breasts every three years. This can pick up on signs of cancer long before a noticeable lump appears.
The results of a mammogram are examined by two independent doctors, but every now and then, a diagnosis is missed, or someone who does not have cancer is wrongly diagnosed with the disease. To see if this small amount of error could be improved by technology, researchers at Google Health trained an AI model to spot breast cancers from mammograms of women in the US and UK.
Despite knowing nothing about the medical histories of the patients, the AI was just as good at identifying cancer as real-life physicians, and gave a reduced number of false positives and false negatives. For American women, the AI reduced the number of falsely identified cancers by 5.7 percent, while for UK patients, the drop was 1.2 percent. Meanwhile, the proportion of cases that went unnoticed declined by 9.4 percent and 2.7 percent for US and UK women respectively. The findings are reported in Nature.
The researchers point out that instead of simply replacing the opinions of medical professionals, the AI could help to reduce the workload of already busy doctors. Where two experts normally examine the results of a mammogram, a single doctor could and then compare their conclusion to that of the AI. If the two disagree, a second doctor could be brought in to examine the scan. The team believes this would reduce the workload of the second doctor by as much as 88 percent.
"Find me a country where you can find a nurse or doctor that isn't busy," UK Google Health lead Dominic King told AFP. "There's the opportunity for this technology to support the existing excellent service of the (human) reviewers."
This isn’t the first time AI has been shown to beat humans when it comes to medical diagnoses. It’s better than doctors at detecting skin cancer, noticing eye disease, and can even spot signs of Alzheimer’s disease years before recognizable symptoms appear.
As for the new AI system, the team hopes future studies will build on their findings and determine whether it should be used clinically. “This robust assessment of the AI system paves the way for clinical trials to improve the accuracy and efficiency of breast cancer screening,” they write.