I mentioned in a previous post that I work for a big data platform company, which I am hopeful can be used to really allow data to help make people’s lives better, healthier, and safer. One of the recent articles that has come out regarding artificial intelligence has a deeply personal aspect, skin cancer detection.
Now, I know I wrote a post a little while ago that outlined how doctors were saying that smartphone apps were no match for a dermatologist for detecting cancerous skin lesions. In fact, amelanotic nodular melanoma was one type that the apps routinely misdiagnosed as benign.
But algorithms help machines learn and those algorithms are pretty powerful when machine learning happens. I’m not smart enough to explain how machine learning actually works, other than to say that providing data with an algorithm allows the software to actually refine and learn characteristics of whatever you’re trying to teach it. And machine learning has really begun to soar once we were able to harness the power of using petabytes of data.
Anyway, long introduction to the article that summarizes the research carried out at Stanford who developed a system that was “able to distinguish between cancerous moles and harmless ones with more than 90 percent accuracy.” That’s quite an impressive result. But even more impressive was when they put this system up against trained dermatologists where the system “achieved performance on par with all tested experts.”
Now, does this mean that your dermatologist will be soon replaced by a machine? No, absolutely not. First of all, no one has developed a robot doctor yet to perform excisions (although I’m now picturing Rosie the maid robot from the Jetsons wielding a scalpel and that is a creepy mental image). Secondly, the system was fed almost 130,000 images to learn the characteristics of skin cancer; but those images were high-quality, high-resolution. As I learned, my Loki did not want its picture taken. (Seriously, every picture I took made it look like a weird pimple rather than pink melanoma.) So, the leap from this system to an app that can diagnose melanoma – let alone abnormal melanoma – on the fly as an app using your smartphone camera is not going to happen in the next few months. But it is cool to see researchers using machine learning to advance the tools that can spot skin cancer.
It almost feels like we’re finally moving into the promise of the space age…just leave Rosie out of the doctor’s office.