Fig. 4: Adaptive stain normalization from Center 2 to Center 1 with StainLUT. | npj Digital Medicine

Fig. 4: Adaptive stain normalization from Center 2 to Center 1 with StainLUT.

From: Self-supervised stain normalization empowers privacy-preserving and model generalization in digital pathology

Fig. 4

A The StainLUT model trained in Center 1 infers the test source images in Center 1. B The StainLUT model trained in Center 1 adaptively converts the pathological images in Center 2 to the staining style of Center 1. C The UMAP is used to visualize the pathological color features of the two centers before and after executing StainLUT. Left panel: Tumor (T) and non-tumor (NT) images from the same center have similar color characteristics, with significant differences between centers. Right panel: After StainLUT normalizes the staining style of Center 2 to that of Center 1, their feature distributions become closer together and cannot be distinguished as clearly as before.

Back to article page