Can AI learn rare childhood diseases from fake faces alone?

Diagnosing rare genetic diseases in children from facial features is powerful but nearly impossible—there's almost no real training data due to privacy rules and the diseases' rarity. Researchers trained computer vision models exclusively on high-quality synthetic faces and found they performed as well as models trained on real pediatric images across multiple architectures. The synthetic-only approach solves a genuine clinical bottleneck: it enables privacy-preserving visual resources for genetic counseling and clinician education without requiring access to actual patient photos.