MedGenAI: Synthetic medical images for fair AI

MedGenAI from DFKI is a prototype service that generates highly realistic, customizable synthetic medical images, enabling fair and reproducible testing of AI systems. The focus is on applications such as skin cancer analysis and radiology, where the service simulates underrepresented patient groups or systematically varies real data – for example, by age, skin type, gender, or image quality.

In the demonstrator, users select synthesis parameters via an intuitive web UI, define target groups, and start generating images. The web UI visualizes the results directly: comparisons of original and synthetic data, bias analyses, and behavior curves of AI models under controlled conditions. This approach precisely identifies weaknesses, such as demographic biases or robustness gaps, without requiring sensitive patient data.

MedGenAI addresses core challenges in trustworthy medical AI: data protection through synthetic data, transparency via configurable parameters, and reproducibility for scientific validation. The demonstrator integrates with the DFKI competence field Trusted AI for Healthcare and supports developers in creating robust, ethically secure systems in the face of scarce real training data.