Standardized CycleGAN training for unsupervised stain adaptation in invasive carcinoma classification for breast histopathology

Generalization is one of the main challenges of computational pathology. Slide preparation heterogeneity and the diversity of scanners lead to poor model performance when used on data from medical centers not seen during training.