A global atlas of digital dermatology to map innovation and disparities

Nature
Read full story· 3 min
Researchers compiled 1.1 million dermatology images from 29 public datasets into a semantic atlas, revealing significant biases: only 11% depict darker skin tones and 2.3% pediatric cases, with data redundancy increasing despite growth.

What happened

On November 21, 2026, a team of researchers published a study that combined 1.1 million dermatology images from 29 public datasets into a single, searchable atlas. This project aimed to provide a comprehensive overview of the available digital dermatology data worldwide. The analysis revealed that the majority of images represent lighter skin tones and adults, with significant underrepresentation of darker skin tones and pediatric cases. The study also found that, despite the rapid increase in available images, much of the new data repeats information already present in existing datasets.

Key details

  • The atlas includes 1.1 million images from 29 public datasets.
  • Only 11% of the images show darker skin tones (Fitzpatrick V–VI).
  • Just 2.3% of the images are of pediatric patients.
  • The growth in dataset size has led to increased redundancy rather than new information.
  • The atlas is designed to be queryable for research and analysis.

Why it matters

Developers and researchers using these datasets to train AI models for dermatology may face limitations in accuracy and fairness, especially for people with darker skin or children. Healthcare providers and patients in underrepresented groups could experience less reliable AI-driven diagnostics. The findings highlight a need for targeted data collection to improve representation and reduce bias in future AI tools for dermatology.


More in Healthcare