It's done...

With these words – which are now engraved in the common memory – presenter Erik Van Looy ended every episode of the immensely popular TV quiz ' The Smartest Man in the World’. These words recently popped into my head when I read an article about a Chinese research team that had built an AI algorithm to do better cancer diagnostics. And succeeded at correctly interpreting images better than an army of radiologists...

The numbers are hallucinating. In several rounds, several scans were shown to a group of renowned radiologists, as well as an artificial intelligence program, called Biomind. Of the total of 225 scans, the radiologists could make the correct diagnosis in 66 percent of cases in 30 minutes. Biomind did it gigantically better with 87 percent in half the time. Auch…

So, it happened. Once again, an AI manages to defeat man. Chess, Go and now radiology... What's Next?

“Yes, but...", I hear people say very often when they hear this kind of news. Followed by a whole series of arguments to proof their point: why it is not fair and that the computer has more computing capacity and that the radiologists have gotten too little time and that the light was not ideal to make a correct diagnosis and that the coffee (or tea) was too lukewarm and... and... Even though some of these arguments may be correct or not, it doesn't matter. It is about estimating trends and interpreting what is happening. 10 years ago, most people had Never Heard of AI and now it can do this. What does that mean for the next 10 years or 20 or what about 2050?

Does this mean this is the end of radiology? Doom thinkers will claim so and state this with a lot of verve. My estimation is that this is not the case but, the role will fundamentally change – and in extension for many doctors. There are now another 13% of the cases where the computer does not know the diagnosis. Which means that AI can be used to interpret the 'normal' cases, and that leaves ample time to deal with the complex cases. But that requires an enormous amount of knowledge and expertise. Do we train our doctors of the future to diagnose the 87% or the remaining 13%? I 'm not sure whether I want to know the answer.

Tom is a regular columnist at De Artsenkrant. This article appeared first in Dutch on July 24th, 2018 in De Artsenkrant.