AI in Dermatology
Image models match dermatologists on tests of melanoma and lesion classification, but real-world proof is thinner. Skin tone bias is a known problem. Dermatology bodies now publish position statements on AI, and large language models and AI scribes are the newest topic.
What AI does in Dermatology
Scores a lesion image as benign or malignant to support referral and biopsy decisions.
Examples: Research CNNs; commercial dermoscopy tools
Helps non-specialists decide who needs a dermatologist.
Examples: Teledermatology tools
Compares photos over time to find new or changing lesions.
Examples: Mapping systems
Drafts notes from the consultation. Discussed in recent position statements.
Examples: General ambient scribes
Lists likely skin conditions from a clinical photo.
Examples: Google deep learning system for skin disease
Scores acne, psoriasis or hair loss from images for trials and follow-up.
Examples: Research tools
routine = in everyday use in many centres · early adoption = in use at some sites, evidence still building · research = studies only
Key ideas for this specialty
New to the basics? Start with the theory lessons.
Tools
Competition dataset for melanoma classification practice.
Regulatory status: Not listed
A listing here is not an endorsement. Ask any vendor for its CDSCO licence and class before clinical use.
Where to learn
Recent society position on safe AI use.
The paper that started the field.
Shows AI help changes doctor accuracy, for better and worse.
Skin tone performance gap.
Multi-condition model.
General medical AI grounding.
Do it yourself
Download the Kaggle melanoma set. Fine-tune a small pretrained network in a free notebook and report AUC.
Open the dataset or tool →Read the Science Advances disparities paper, then test a public skin model on a few de-identified images of darker skin from your clinic, with consent. Note the errors.
Open the dataset or tool →Use public or properly de-identified data only, and get ethics approval before any project on your own patients.
Papers worth reading
- Dermatologist-level classification of skin cancer with deep neural networks Nature, 2017
- Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists Annals of Oncology, 2018
- Human-computer collaboration for skin cancer recognition Nature Medicine, 2020
- Disparities in dermatology AI performance on a diverse, curated clinical image set Science Advances, 2022
Societies and guidance
Position statement on AI adoption (2026).
Open skin image archive and challenges.
Recent updates
JAMA Dermatology: AI tool raised lay people's skin diagnosis accuracy, below dermatologist panels
In a three-arm randomised trial of 2,345 people, correct naming of a skin condition was 7.9% with web search, 22.8% with an AI tool and 36.2% with dermatologist-curated differentials. Understanding of next steps did not improve.
Source: Managed Healthcare ExecutiveLinks checked 3 October 2026. “Not listed” means we could not confirm the detail, not that it is absent. Educational material, not clinical advice.