Most dermatology algorithms don’t have transparent data

0


Most of the algorithms designed to help people identify skin issues do not allow experts to see the datasets they were developed with and do not share information about the skin tone or ethnicity of patients within. these data sets, according to a new review. This could make it difficult for people to evaluate the programs before using them and understand if they might not work as well for certain groups of people, the authors argue.

These types of tools use images of skin conditions to train a system to recognize these same conditions in new images. Someone could upload a photo of a rash or mole, and the tool would be able to tell what type of rash or mole it was.

The article, published in JAMA Dermatology, analyzed 70 studies that developed a new deep learning model or tested an existing algorithm on a new set of data. Together, the models were developed or tested using over a million images of skin problems. According to the analysis, only a quarter of these images were available to experts or the public. Fourteen of the studies included information on the ethnicity or race of patients in their data, and only seven described their skin type.

The others did not share the demographic distribution of their patients. “I strongly suspect that these datasets are not diverse, but there is no way to find out,” said study author Roxana Daneshjou, dermatology clinician at Stanford University. Twitter.

The analysis also checked whether models aimed at identifying skin cancer were trained on images where cancer was confirmed with a skin sample sent to a lab – the “gold standard” to make sure the diagnosis was correct. Of the included studies, 56 claimed to identify these conditions, but only 36 of them have achieved the gold standard. Those who could not have been less specific, say the authors.

The review included an algorithm from Google, which developed a tool designed to help people identify skin conditions. The company plans to create a pilot version of its web tool, which allows people to upload photos of a skin problem and get a list of possible conditions, later this year. According to the analysis, the Google article includes skin type and an ethnic breakdown, but did not make the data or model used publicly available. He also did not use the gold standard methods to assess a few types of skin cancer, including melanoma and basal cell carcinoma.

Medical algorithms are as good as the data they were developed with, and may not be as effective if used in situations other than what they were trained on. This is why experts argue that data, or descriptions of that data, should be freely available: “The data that is used to train and test a model can determine its applicability and generalizability. Therefore, a clear understanding of the characteristics of data sets… is essential, ”the authors wrote.

Lack of transparency is a constant problem with medical algorithms. Most AI products cleared by the Food and Drug Administration (FDA) do not report important information about the data they were developed with, according to a February 2021 report. New statistics investigation. The FDA said New statistics that its new “action plan” for AI pushes for more transparency.

The limitations do not mean that most dermatology algorithms are useless, wrote Philipp Tschandl, a researcher at the University of Medicine in Vienna, in an accompanying editorial. Doctors are also not perfect and have their own biases or knowledge gaps that can skew their interpretation of a skin problem. “We know this and still manage to practice medicine well,” he wrote. “We need to find ways, through explainability, smart controls and risk mitigation, to allow algorithms to work safely and fairly in medicine. “



Share.

About Author

Leave A Reply