RandStainNA: Learning Stain-Agnostic Features from Histology Slides by Bridging Stain Augmentation and Normalization

Stain variations often decrease the generalization ability of deep learning based approaches in digital histopathology analysis. Two separate proposals, namely stain normalization (SN) and augmentation (SA), have been spotlighted to reduce error, where former alleviates shift across different medical centers using template image latter enriches accessible styles by simulation more variations. However, their applications are bounded selection images construction unrealistic styles. To address problems, we unify SN SA with a novel RandStainNA scheme, which constrains variable practicable range train agnostic model. The is applicable collection color spaces i.e. HED, HSV, LAB. Additionally, propose random space scheme gain extra performance improvement. We evaluate our method two diagnostic tasks tissue subtype classification nuclei segmentation, various network backbones. superiority over both yields that proposed can consistently improve ability, models cope incoming clinical datasets unpredicted codes available at https://github.com/yiqings/RandStainNA.