Researchers at Imperial College London ‘trained’ a computer, using the DeepMind algorithm, to spot signs of cancer on nearly 29 000 X-ray images of women. The rate of accuracy at spotting cancer was compared to the success of doctors.
The experiment was also conducted using photos of women from the US; in both cases, it was concluded that the AI was more effective than the medical professionals at correctly identifying cases of breast cancer.
The study found noticeable improvements, particularly in the US, namely a reduction of 5.7% in false positives (exam incorrectly diagnosed as abnormal) and 9.4% in false negatives, where a cancer hasn’t been identified but should have been. These margins are smaller in the UK study, though are still noticeable (1.2% and 2.7%).
When implemented into hospitals, this discovery has the potential to provide an immediate second opinion, immensely aid doctors when conducting mammogram exams and reduce the rate of both false positives and missed cancers.
Featured Image: National Cancer Institute on Unsplash