Staining normalization in histopathology: Multi-center dataset and method benchmarking

The H&E-stained tissue image dataset was obtained as part of an external quality assessment (EQA) initiative coordinated by Labquality, a firm specializing in external quality assessment programs for clinical laboratories located in Helsinki, Finland. These slides featured a tissue microarray section comprising three 6mm punch biopsies, each extracted from normal human skin, kidney, and colon tissue specimens. These tissue samples were obtained from anonymized, formalin-fixed, and paraffin-embedded histological specimens from a reference pathology laboratory.

During the EQA round, slides with 3-micron unstained tissue sections were dispatched to EQA participant laboratories. These laboratories were instructed to apply their routine H&E staining methodology, typically used in their daily diagnostic practices. In total, 66 laboratories from 11 different countries participated in the evaluation process. Subsequently, these slides were digitized using a Hamamatsu Photonics NanoZoomer-XR slide scanner, employing a 20× objective lens, resulting in a scanning resolution of 0.46 μm per pixel. For stain normalization experiments, the whole slide images (WSI) were resampled to 10×.

This WSI dataset is structured in three folders, namely, tissue_a, tissue_b, and tissue_c, containing skin, kidney, and colon samples, respectively. Each folder contains 66 WSIs in TIF format.