AI in skin science: different questions, different evidence
Image labels, drug transport and clinical image assessment each demand a different evaluation.
Three distinct research settings
A 2026 skin-tone dataset preprint examines labeling and benchmarking. A 2023 microneedle study compares machine-learning approaches for drug-permeation prediction and identifies drug loading, permeation time and microneedle surface area as influential variables. A September 2026 preprint compares three multimodal AI models with three dermatologists using patient-captured images; results depend on the metric, and model confidence is not consistently well calibrated.
Keep the claims attached to the task
These studies do not measure the same capability. Predicting permeation is different from classifying skin tone or interpreting a lesion image. High performance in one dataset does not establish reliability across a new population, acquisition method or intended use.
Research and regulation answer different questions
The FDA’s 2026 melanoma-device order concerns specific adjunctive device types. It does not validate the research systems discussed here. Our editorial recommendation is to describe each model through its task, reference standard, evaluation population and limitations before discussing practical use.
Sources & further reading
Read the original company post on LinkedInSkin-tone benchmark — arXiv:2603.02475Microneedle permeation study, 2023Multimodal AI benchmark — arXiv:2609.24190FDA final order on two melanoma device typesAdapted from a public company post, with sources checked on 8 October 2026. Original post dates are preserved; the website summary is not attributed to an individual employee.
