AI to Detect Breast Cancer at an Early Stage Outperforming Radiologists
According to the new paper published in Nature on Wednesday, the new AI tool created by Google Health and Deepmind is capable of spotting breast cancer as efficient as a human radiologist.
Researchers trained the model using the two data sets of breast scans from the US and UK. The US data set included data sets of 25,856 women, while the UK data set had the mammograms of 3,097 women.
The new AI tool reduced the false negatives in the UK by 2.7% and the US by 9.4%, while false positives were reduced by 1.2% in the UK data set and 5.7% in the US data set.
When pitted against six human radiologists, the new model outperformed all of them on an average by 11.5%.
The results indicate the significant role AI could play in cancer care in the future. The growth of technology would improve cancer diagnosis in the near future.
Screening helps to diagnose breast cancer at an early stage when the treatment is more likely to be successful. But there is a possibility of false positives in screening. AI could improve the screening process and take the pressure of the UK’s national health service, said Cancer Research UK chief executive Michelle Mitchell.
Although further clinical studies are needed to see if the tool can work in practice, the initial results are promising.
According to Breast Cancer Research UK, 11,400 people are killed every year due to breast cancer.
The AI tool to detect breast cancer could be used as a second opinion to help radiologists, Deepmind suggested. It is a standard procedure in Uk to analyze the scan by two radiologists. Deepmind claims to have run a simulation that confirms that the work of second readers can be reduced by 88% using the AI tool.
Deepmind had previously developed AI tools for detecting neck cancer and eye diseases. In 2017 Deepmind’s healthcare research was under observation after UK’s data watchdog accused them of entering into an agreement with NHS to share patients data. This time Deepmind stressed that the data collected are subjected to strict privacy controls and were anonymized.