In the recently published article in the journal JEADV analyze Theresa Willem, Prof. Alena Buyx and colleagues discuss the potential risks and benefits of dermatological machine learning applications for different user groups (patients, physicians) and place them in the context of society as a whole. In their article, they show that medical ethics research on the development of machine learning models in dermatology is not yet sufficiently developed for the standardized use and regulation of these new technologies.

Visual data is particularly well suited for machine learning. As clinical photography is well established as a visual method for skin surveillance, documentation and progress monitoring, dermatology represents an ideal development and application area for machine learning health care applications (ML-HCAs). Several ML-HCAs already detect malignant skin lesions at a similar level as Expert:in and find visual patterns that correlate with specific dermatological diseases. In parallel to these developments, the ethical and social implications of ML-HCAs are being discussed in the professional community; this paper provides an overview of the debate.

Based on a thematic, keyword-based literature search, we conducted an ethical analysis using established medical ethics frameworks and combined our findings with current, relevant normative literature from machine learning ethics to determine the status quo of the ethics of ML-HCAs in dermatology: The potential benefits of ML-HCAs range from improved treatment outcomes, to increased access to knowledge for medical professionals, to reduced health care disparities, i.e., standards of care for different populations. The risks associated with ML-HCAs range from confidentiality issues to potentially worsened treatment outcomes to exacerbation of existing health care disparities. In our article, we discuss the practical implications for all phases of dermatologic ML-HCA development.

We find that ML-HCAs pose specific risks to patients, health professionals, and society that need to be considered separately. We conclude that there is a lack of discipline-specific medical ethics research that sufficiently accompanies the different development stages of ML-HCAs (design, implementation, use, and regulation) of ML-HCAs to ensure ethically and socially responsible development and clinical implementation, especially in the field of dermatology.

The article is available at: https://doi.org/10.1111/jdv.18192