AI finds cancers that doctors missed

When it comes to medical diagnoses, it’s always good to get a second opinion. Getting a third from an artificially intelligent healthcare tool might not be a bad idea, either.

What happened: An AI tool called Mia caught multiple small cancers missed by doctors in a test of 10,000 women’s mammograms performed by Britain’s National Health Service (NHS).

  • Mia caught all the scans with cancers that radiologists also found, in addition to 11 that they missed.

Why it matters: It’s promising evidence that more advanced AI tools can do much more valuable things than generate wacky images or help kids cheat on their schoolwork. 

  • Mia could make breast cancer screening both more effective (by catching cases that would’ve otherwise been missed) and less labour-intensive.
  • The NHS, for example, requires two radiologists to review every mammogram — a tool like Mia could someday replace one of those people, effectively doubling productivity.

Yes, but: It’s much more difficult to train AI models to diagnose cancer than write mediocre sonnets.

  • One of the main challenges is that AI models require large amounts of training data to learn, but medical data is usually protected by strict patient privacy rules.
  • Mia took six years to build and train and didn’t have access to individual patients’ medical history — that meant it often flagged things previous doctors had already caught, like harmless cysts.

Zoom out: Other AI tools have also shown promise in the healthcare space. Earlier this year AMIE, a chatbot powered by a large language model, was able to diagnose medical problems in a controlled experiment as accurately as doctors.

Bottom line: With a chronic labour shortage and high costs, healthcare may be one of the most fertile sectors for AI to take root.