Team Develops Skin-Tone Conversion AI for Skin Diagnosis
A Seoul Asan Medical Center research team developed generative AI that changes apparent skin tone while preserving the shape of a skin lesion, aiming to address diagnostic performance gaps across skin tones.
This article was automatically drafted and passed editorial compliance checks. It synthesizes public source metadata and is not a substitute for individual care or clinical review.
The team included Professor Kim Nam-kook, dermatology resident Kim Kyung-hoon and dermatologist Moon Ik-jun. Their technology converts images of lighter skin to resemble darker skin while retaining the lesion’s morphology.
Why skin-tone conversion matters
The research team targeted a known imbalance in skin-disease AI: clinical training data are concentrated largely on lighter skin, including Fitzpatrick skin types I and II, and diagnostic accuracy can vary with a patient’s skin tone and race.
By changing skin tone without altering lesion structure, the team designed the generative AI to help supplement images of darker skin for diagnostic-AI development.
What the development means
Some skin-disease AI systems have reached dermatologist-level performance in certain areas, but the team’s work addresses the separate question of whether that performance extends consistently across different skin tones.
Creating a skin-tone conversion method is not, by itself, evidence that a diagnostic system will achieve equal accuracy in real clinical populations. That conclusion requires validation across diverse patients and direct comparison of diagnostic performance by skin tone.
MediIndex articles are for general information only and are not medical advice, diagnosis, or advertising. Outcomes vary by individual — consult a board-certified specialist for personal decisions.
