Dermatologists urged to be more involved in AI research
Dermatologists need to be more involved in the development of AI technologies designed to evaluate skin cancer, according to a new analysis published in the Journal of the American Academy of Dermatology.
The authors performed an online search for academic studies that included the terms “machine learning” or “artificial intelligence” combined with “skin cancer” or “melanoma.” To be included in their analysis, a study had to include three things: full-length clinical trials, the use of machine learning specifically to screen or diagnose skin cancer, and the use of dermatoscopic images or actual photographs of gross lesions. Studies were excluded if they duplicated prior research or only focused on animal models.
After making hundreds of exclusions for the previously listed reasons, the team included 51 articles in its final analysis. Overall, 41% of the included articles had at least one dermatologist listed as an author.
“Since 2017, there has been a sharp rise in the number of publications describing new models in the literature, and the majority have been built using convolutional neural networks—models consisting of several layers of simple algorithms that extract and assess features automatically to classify input images,” wrote lead author George A. Zakhem, MD, MBA, New York University School of Medicine, and colleagues. “However, this increase in publications has been driven largely by teams of non-dermatologists.”
Studies including dermatologists as authors had larger image datasets by a significant margin, which the authors said was one primary reason dermatologist input is so crucial. Zakhem et al. also noted that the models were typically designed as if they were being used by dermatologists, even if a dermatologist wasn’t specifically invited to assist with the research. Studies designed for dermatologists and performed by someone else—a primary care provider, for example—could end up with “high rates of false positive screens.”
“Moreover, the majority of models are built using dermatoscopic images; dermatoscopes are much less likely to be available to a primary care practitioner or patient that to a dermatologist,” the authors wrote. “Although decision support for dermatology is a viable use case for these technologies, it should be dermatologists, rather than data scientists, who define how this technology is used in the clinical setting. A robust understanding of the clinical context is absolutely essential for effective implementation of these technologies.”
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.