Google Health and DeepMind claim to have created an AI tool capable of outperforming human radiologists at spotting breast cancer from mammograms.
The AI was trained on two datasets of mammogram scans from the US and the UK, then pitted against six radiologists.
While the research is still at an early stage, DeepMind suggested it could be used as a "second reader" to give radiologists an automated second opinion.
Google Health and DeepMind have created an AI tool capable of spotting breast cancer with as much accuracy as a human radiologist, according to a new paper published in Nature on Wednesday.
To train the model the researchers used two data sets of breast scans from the UK and the US. The UK dataset included scans from 25,856 women, while the US set contained mammograms from 3,097 women.
Applying the AI resulted in a reduction of false negatives of 9.4% for the US dataset and 2.7% for the UK. There was a slightly smaller reduction for false positives, 5.7% for the US dataset and 1.2% for the UK one.
The new model was then pitted against six human radiologists, and outperformed them all on average by 11.5%.
"These results highlight the significant role that AI could play in the future of cancer care. Embracing technology like this may help improve the way we diagnose cancer in the years to come," Cancer Research UK chief executive Michelle Mitchell told Business Insider in a statement.
Mitchell said screening was a process that helps diagnose breast cancer at an early stage, when treatment is more likely to be successful. But it also has issues such as false positives. AI, she said, could help improve the screening process and take pressure of the UK's national health service.
"While further clinical studies are needed to see how and if this technology could work in practice, the initial results are promising," she said.
Breast cancer kills roughly 11,400 people in the UK every year according to Breast Cancer Research UK.
DeepMind has suggested that the AI tool could be used as an automated "second reader" to assist radiologists. In the UK it is already standard procedure for two radiologists to view a scan. DeepMind claims to have run a simulation which indicated using the AI could alleviate the work of second readers by 88%.
This is far from DeepMind's first foray into healthcare. The company has previously developed AI models for identifying eye diseases and detecting neck cancer.
In 2017 its healthcare research came under scrutiny after the UK's data watchdog ruled it had illegally entered into a patient data sharing agreement with the NHS. For its breast cancer research DeepMind stressed that the datasets collected were anonymised and subject to strict privacy controls.