PracticalDermatology

A growing number of apps claim to help individuals determine if a lesion or mole is concerning and requires further medical evaluation. The upside would be earlier identification of skin cancer, but these apps are not without risks, namely unnecessary biopsies or false security. Research in the April 2022 issue of the British Journal of Dermatology1 suggests that these apps are not ready for prime time yet.

Specifically, apps were, on average, just 59 percent accurate, and when presented with types of images and skin conditions they had never seen before, they were less accurate than they were with more familiar images. Algorithms were only six percent accurate for a diagnosis they had not seen in training, the study showed.

Lead author Veronica Rotemberg, MD, PhD, a dermatologist at Memorial Sloan Kettering Cancer Center in New York City, is developing and evaluating artificial intelligence (AI) and other computer-based tools for detecting and diagnosing melanoma. Here, she discusses the results of the study and what they mean for patients.

What should dermatologists be telling patients about these apps?

Veronica Rotemberg, MD, PhD: AI for dermatology applications, especially diagnosis of skin cancer, is a very promising research area. That said, there have been very few prospective studies of these apps in their intended use setting to date. When we have tried to benchmark these apps ourselves, they have not performed well. It’s really hard to make a blanket statement about every single app, since new ones are being developed all the time and need to be individually tested against accepted benchmarks. To date, I haven’t seen any that I would be able to recommend for personal use.

What are the risks of using these consumer-facing apps to screen for melanoma?

Dr. Rotemberg: In our study, we found two risks. Because the sensitivity was so low overall, these apps have the potential to miss melanomas and give people a false sense of security. Of course, our study was small, and we need bigger studies to truly quantify this risk.

We also found fairly low specificity for many of the apps, which means that they have the potential to overwhelm dermatologists with benign lesions if used by hundreds of millions of people who may have access to unvetted apps on app stores.

How can these apps be improved?

Dr. Rotemberg: I think we need a lot more thought into the entire development process from the beginning to ensure algorithms aren’t just tested at the very end of the process when some of these decisions have already been made.

Were you at all surprised by the accuracy and sensitivity?

Dr. Rotemberg: Yes I was! But I do want to say, of course, that the study is small, and I would like to see other investigators independently vet these apps and see their findings.

Are further studies planned?

Dr. Rotemberg: Yes, this is just an initial, small, study with a small number of images. These results may not be exactly the same if we have a larger sample across all skin tones and disease types, and that’s important before we can start recommending apps for dermatologic applications.

Dr. Rotemberg is an expert consultant for Inhabit Brands, Inc.

Veronica Rotemberg, MD, PhD is a dermatologist at Memorial Sloan Kettering Cancer Center in New York City.

Vishal Anil Patel, MD is director of cutaneous oncology at GW Cancer Center, director of dermatologic surgery at GW department of dermatology, and an associate professor of dermatology & of medicine/oncology at George Washington University School of Medicine & Health Sciences in Washington, DC. He will serve as interim editor as Mark D. Kaufmann, MD fulfills his one-year term as President of the American Academy of Dermatology.

1. Rotemberg V, et al. Accuracy of commercially available smartphone applications for the detection of melanoma. BJD. 2022; 186( 4): 744-746.

More on AI