Dermatology faces specialist shortages and relies heavily on visual assessment, making it a natural target for AI. This review summarizes four main AI approaches in dermatology—image recognition, large language models (LLMs), multi-omics models and multimodal models—and assesses their integration into diagnosis, treatment decisions, personalized management and remote care.

Image-based models can classify skin lesions and support early detection of melanoma and basal cell carcinoma. LLMs such as GPT-4 and SkinGPT-4 can analyze patient histories and generate diagnostic reports, while multi-omics and multimodal models help with disease subtyping, biomarker discovery and integration of images, clinical records, omics data and wearable monitoring. AI is also being applied to drug development, treatment response prediction and drug repurposing, and telemedicine platforms and mobile apps are being used for remote assessment and follow-up.

The review highlights persistent barriers: mainstream datasets such as ISIC and HAM10000 are dominated by lighter skin tones, and HAM10000 lacks clear Fitzpatrick skin type labels. The SCIN dataset was designed with more skin-color diversity. The authors call for targeted data collection, fairness-aware training, external multi-ethnic validation and post-market monitoring, and stress that AI should remain an assistive tool rather than replace clinicians.

The work titled “Up-to-Date Progress, Future Challenges and Future of Artificial Intelligence in Dermatology: From Diagnosis, Drug Discovery and Clinical Pathway” was published on Skin (published on September 23, 2026).
DOI:10.2738/SKIN.2026.0011