Dr Jonathan Kentley Discusses Training Algorithms to Detect Melanoma

Many algorithms claiming to identify melanoma better than dermatologists are trained on retrospective datasets, which is not reflective of real-world clinical practice, said Jonathan Kentley, MBBS, MSc, research fellow at Memorial Sloan Kettering Cancer Center.

Many algorithms claiming to identify melanoma better than dermatologists are trained on retrospective datasets, which is not reflective of real-world clinical practice, said Jonathan Kentley, MBBS, MSc, research fellow at Memorial Sloan Kettering Cancer Center.

Transcript

What can you tell us about the Skin Analytics study evaluating the effectiveness of an image-analyzing algorithmin identifying certain skin conditions?

The issue with a lot of these algorithms that are claiming they're much better than dermatologists, they've all been trained on retrospective data sets of images—which doesn't really mirror clinical practice and how they're going to be rolled out in reality—whereas Skin Analytics is one of the few companies that have really gone through the effort to make a well-controlled prospective trial. They've already done one in the UK, across 7 different hospitals, with about 500 lesions that they imaged, and showed that, actually, it was sort of on par with dermatologists at detecting melanoma.

But now that they have a breakthrough FDA authorization and a CE mark, they're embarking on a much bigger trial across Europe and the US. I think it'll be really interesting to see the results of this, and as with any medical device or medication, we really need rigorous clinical trials to validate them in clinical practice, so I'm excited to see the results.