Artificial intelligence is achieving a new milestone with each passing day. Adding to this streak, Google's AI now often outperforms the experts, detecting breast cancer in the human body.
Screening for breast cancer is usually done by scanning mammograms (or X-ray images of the breasts), but they have limitations. They can give a false positive or false negative result.
A false negative mammogram looks normal even when breast cancer is present. Likewise, a false-positive mammogram shows breast cancer even if it isn't present.
Google's AI can detect breast cancer, but...
Google AI has been trained to read and analyze mammograms for the presence of cancer cells. For that, the researchers used about 76,000 anonymous mammograms from women in the UK and 15,000 mammograms from women in the US.
They used a separate anonymous mammogram database of 25,000 women from the United Kingdom and 3,000 women from the United States to assess AI.
The results returned by the AI were compared to actual medical reports to verify accuracy. It managed to reduce the number of false negatives by 9,4% for American women and 2,7% for UK women.
On the other hand, it reduced false positives by 5,7% in the US and 1,2% in the UK.
The findings have been published in the journal Nature by the team which includes researchers from DeepMind, Royal Surrey County Hospital, Northwestern University and Cancer Research UK Imperial Centre.
still not perfect
While Google's AI managed to beat the human experts in many cases, there were times when the experts detected the presence of signs of cancer in cases that the AI missed.
However, here, Google points out that the AI had access to less information than human experts, such as patient histories and previous mammograms. Still, he managed to do quite well.
So it means that AI cannot replace humans in this field at the moment. However, the goal is that the technology should complement radiologists during screening programs and increase the accuracy and efficiency of results.
As development continues, one thing the researchers are trying to figure out is how they can generalize the AI results.
For starters, they ran a separate test where the AI was trained on data from UK women and evaluated the US women's data set. It reduced false negatives by 8.1% and false positives by 3.5%.
Via The Verge