DFKI project MedGenAI uses synthetic images to counter bias in skin cancer detection (heise.de)

• AI models for skin cancer detection are biased due to imbalanced training data.
• DFKI's MedGenAI generates synthetic skin images controllable by age, gender, and skin type.
• Counterfactuals measure whether a model gives the same diagnosis when an aspect changes.

"The DFKI's MedGenAI project uses generative AI to reduce bias in medical image recognition. By creating synthetic skin images controllable by age, gender, and skin type, it generates more balanced test datasets. Counterfactuals reveal when predictions become unfair. The goal is to improve fairness of AI diagnostics, not to eliminate bias completely. Synthetic data complements real data; clinical validation is required before deployment."