Google Health has developed a system that can identify breast cancer more accurately than radiologists, in the latest sign that artificial intelligence could improve early detection of disease in images.
In a paper published in the scientific journal Nature, experts from Google Health, Alphabet’s DeepMind unit, and UK and US universities showed the AI model reduced both false positives, in which patients are wrongly told they have cancer, as well as false negatives, where the disease is present but not diagnosed.
Screening mammograms is known to be imperfect, failing to detect about one in five breast cancers, according to the American Cancer Society. More than half of all women are given a false positive every 10 years, causing anxiety and leading to unnecessary treatment, which was estimated in a 2015 study in the journal Health Affairs to cost the US more than $4bn a year.
Dominic King, the UK lead for Google Health, said the results were “really exciting” and showed how AI could be used to help screen for cancers in earlier stages, when the disease is harder to detect accurately. The algorithm was trained and tested on de-identified images from almost 120,000 mammograms in the US and the UK.
DeepMind recently transferred control of its health division to parent company Google. Mustafa Suleyman, who oversaw the health team, is leaving DeepMind for a new job examining the opportunities and impacts of applied artificial intelligence at Google.
Dr King, who was previously a breast cancer surgeon, said Google embarked on the study partly because senior radiologists did not think the UK’s cancer-screening services were sustainable. In 2018, the Royal College of Radiologists estimated that the country would need more than 1,000 additional full-time diagnostic radiologists to meet demand.
He hopes the AI could one day obtain regulatory approval to be a support tool to clinicians — but not replace them. “It could be a second opinion, giving a nudge or a recommendation here and there to spend more time looking at this scan, or to flag examples where cases get missed,” he said.
Large technology companies are becoming increasingly interested in using their artificial intelligence expertise in healthcare, in particular deploying computer vision algorithms to spot patterns in images that require examining visual signs, in fields such as pathology, ophthalmology and dermatology.
Google has already created the Lymph Node Assistant, which is 99 per cent accurate at detecting late-stage breast cancer cells that have spread around the body. DeepMind is preparing for the commercial launch of a device that can diagnose complex eye diseases as accurately as specialists.
Big Tech also faces competition from start-ups such as Kheiron Medical, which is training a similar algorithm to identify signs of breast cancer. Based on training data from hospitals in Hungary, it beat the average performance of a radiologist.
The Google Health paper was authored in collaboration with the Cancer Research UK Imperial Centre, Northwestern University in Illinois, and the Royal Surrey County Hospital. It found that the AI model reduced the number of false positives by 5.7 per cent in the US and 1.2 per cent in the UK, where each scan is checked by two radiologists. It reduced false negatives by 9.4 per cent in the US, and 2.7 per cent in the UK.
Professor Ara Darzi, an author on the paper and director of the cancer centre at Imperial, said it was “very encouraging”.
“There will of course be a number of challenges to address before AI could be implemented in mammography screening programmes around the world, but the potential for improving healthcare and helping patients is enormous,” he said.
