IBM Is Teaching Watson To Interpret Medical Images

In the latest sign that the singularity is nigh, IBM announced last week that it would start teaching its ultra-fast computer system “Watson” to be something like a robotic radiologist.

The goal is for Watson — most famous for beating human opponents on “Jeopardy!” — to be able to interpret the results of medical images from sources such as CT scans, electrocardiograms and MRIs, as well as photographs of skin conditions such as melanoma — a job currently handled by MDs.

IBM has already started training Watson to analyze visual data — “to see,” the company says. The supercomputer will soon bring this ability to bear on a trove of 30 billion medical images that IBM acquired in its recent $1 billion purchase of health tech company Merge, to figure out how to distinguish a normal result from an abnormal one.

So does that mean that your radiologist cousin could soon be out of a job? That the next time you get an MRI on your bum knee, you’ll hear the results in C-3PO’s voice?

Not quite. At least not according to radiologists — admittedly not an unbiased group on this issue.

Dr. John Eng, a radiologist and machine learning expert at Johns Hopkins Hospital, was dubious that Watson will actually be able to compete with a human radiologist when it comes to visual diagnoses anytime soon.

“It seems like the claims are being made that Watson is going to look at images and make general diagnoses from those images and that seems like it’s a ways off,” he said.

That’s because interpreting radiologic images is arguably one of the toughest visual reasoning tasks that humans take on — far more difficult than, say, identifying a specific person in a photo posted on Facebook. “The —> Read More