The Colorectal_Cancer_IHC_CISH_HE_Epithelium_Segmentation dataset

Dataset Overview One of the largest (colorectal) epithelium segmentation datasets to date, featuring H&E and IHC images with pathologist-annotated segmentations. (full dataset can be downloaded at https://dataverse.no/dataset.xhtml?persistentId=doi:10.18710/DIGQGQ). With over 500,000 epithelium annotations, it represents a significant advance in scale: ~15x larger than our previous IBDColEpi dataset (31k vs >500k annotations; https://dataverse.no/dataset.xhtml?persistentId=doi:10.18710/TLA01U&version=2.1) ~10x larger than SegPath (https://www.sciencedirect.com/science/article/pii/S2666389923000193) in total annotated pixel area (1,300 billion vs 154 billion pixels) Dataset Contents The collection includes Tissue Microarray (TMA) cores from 100 patients, featuring both normal colorectal mucosa and cancer tissue. Each patient has: 3 normal colorectal mucosa cores 3 colorectal cancer cores Each core includes original image and corresponding epithelium segmentation mask Imaging Markers All images are approximately 10,000 x 10,000 pixels at 40X magnification. Each marker includes matched pairs of JPG images and PNG segmentation masks: H&E stained cores (552 image/mask pairs) Immunohistochemistry for 13 proteins: E-Cadherin (558 pairs) Vimentin (559 pairs) Smooth Muscle Actin (SMA) (558 pairs) Ki-67 (557 pairs) SMAD3 (546 pairs) MACC1 (531 pairs) LASP1 (535 pairs) CD44 (526 pairs) NAIP (553 pairs) KLF5 (553 pairs) FSCN1 (553 pairs) CTNND1 (556 pairs) KRAS (547 pairs) ISH stains: miR-143 (2,195 pairs) miR-145 (2,189 pairs) ISH Positive Control (U6 snRNA) (555 pairs) ISH Negative Controls (Scrambled probe) (556 pairs) Total: 13,179 image/mask pairs (26,358 files) Data Format High-resolution histopathological images Full epithelium segmentation masks for both normal and cancer tissue Quantitative measurements (SPSS datafile) for selected markers All annotations validated by pathologists Applications Ideal for AI training in: Colorectal cancer analysis Epithelium segmentation Digital pathology Deep learning-based image analysis Dataset use must adhere to the CC0 licence and cited as: Pettersen, Henrik Sahlin; Wiik, Erik Nesje, 2025, "The Colorectal_Cancer_IHC_CISH_HE_Epithelium_Segmentation dataset", https://doi.org/10.18710/DIGQGQ, DataverseNO, V1