Papers

PatchAlign: Fair and accurate skin disease image classification by alignment with clinical labels

Fairness Medical Imaging Optimal Transport

MICCAI, 2024

PatchAlign improves skin condition image classification accuracy and fairness by aligning dermatology image representations with clinical text representations using Graph Optimal Transport.

Paper Slides
TL;DR

Skin-condition classifiers trained on images alone tend to be less accurate on darker skin tones. PatchAlign fixes this by aligning image patches with clinical-text embeddings through graph optimal transport, closing much of that gap without hurting overall accuracy.