README.md

metadata

license: other
license_name: warwick-tia-citation-required
task_categories:
  - image-segmentation
tags:
  - medical
  - histopathology
  - ihc
  - immunohistochemistry
  - lymphocytes
  - nuclei
  - nuclick
pretty_name: NuClick-IHC

NuClick-IHC (Lymphocyte Segmentation in IHC)

Immunohistochemistry (IHC) stained histopathology patches of lymphocytes with per-nucleus instance segmentation masks. Released by the Warwick TIA Centre as the IHC component of the NuClick framework's training/validation data, with ROIs sourced from the LYON19 cohort (CD3/CD8 IHC of breast, colon, prostate).

Overview

• Modality: Histopathology (IHC, RGB microscopy)
• Tissue: Lymphocytes in CD3/CD8-stained breast/colon/prostate
• Image size: 256x256 RGB
• Samples: 671 train + 200 validation = 871
• Ground truth: Per-nucleus instance segmentation masks generated by the NuClick interactive tool and refined for training. The paper validates these by showing a model trained on them placed first on LYON19.

Columns

Column	Type	Notes
id	string	ROI identifier (e.g. ROI_100_1)
image	Image (RGB)	256x256 IHC patch
mask	Image (mode L)	256x256 uint8 instance map: 0 = background, 1..N = instance IDs
num_nuclei	int32	Number of nuclei instances in the patch (0 if empty)

Notes

• Approximately 30% of training patches and 25% of validation patches contain no nuclei (num_nuclei == 0, mask is all-zero). This matches the source release.
• Max instances per patch in this release is 69, so a uint8 mask losslessly preserves all instance IDs.
• For semantic (foreground/background) use, threshold the mask with mask > 0.

Derivation

Source: ihc_nuclick.zip from https://warwick.ac.uk/fac/cross_fac/tia/data/nuclick/ (IHC subset). The source ships 256x256 PNG images and uint32 .npy instance maps; we re-encode masks as uint8 PNG (lossless under the observed instance count). The companion IHC_xml_asap/ folder contains the raw ASAP-compatible polygon annotations and is not included here.

License

The Warwick TIA release does not provide an explicit dataset license. Users must cite the NuClick paper when publishing work derived from it.

Citation

• Alemi Koohbanani N., Jahanifar M., Zamani Tajadin N., Rajpoot N. NuClick: A deep learning framework for interactive segmentation of microscopic images. Medical Image Analysis, 65:101771, 2020. doi:10.1016/j.media.2020.101771