Twenty-one skin cancer image datasets are publicly available, and most are from Europe, North America, and Oceania, according to the results of a review presented at the National Cancer Research Institute Festival and published online Nov. 9 in The Lancet Digital Health.

David Wen, B.M.B.Ch., from the University of Oxford in the United Kingdom, and colleagues conducted a systematic review to identify and evaluate all publicly available skin image datasets used for skin cancer diagnosis.

The researchers identified 21 open-access datasets containing 106,950 skin lesion images, 17 open-access atlases, eight regulated access datasets, and three regulated access atlases in a combined search. Two independent reviewers evaluated images and accompanying data from open-access datasets. Eleven of the 14 datasets that reported country of origin were from Europe, North America, and Oceania. Nineteen of the datasets contained dermoscopic images or macroscopic photographs only. Clinical information was available regarding age, sex, and body site for 76.4, 77.5, and 74.4 percent of the images, respectively. For 1.3 and 2.1 percent of the images, subject ethnicity and Fitzpatrick skin type data were available, respectively. Among datasets, there was substantial underrepresentation of darker skin types.

"Artificial intelligence programs hold a lot of potential for diagnosing skin cancer," Wen said in a statement. "However, it's important to know about the images and patients used to develop programs, as these influence which groups of people the programs will be most effective for in real-life settings. Research has shown that programs trained on images taken from people with lighter skin types only might not be as accurate for people with darker skin, and vice versa."

Several authors were employed by Databiology.

Abstract/Full Text

More Information