The CoNIC challenge data is released by its authors under CC BY-NC-SA 4.0 for research purposes only. By requesting access you agree to those terms, to cite the CoNIC and Lizard papers, and to share any derivative under the same licence.

CoNIC in the nuclei store format

A faithful conversion of the CoNIC 2022 challenge training data (Graham and others, 2021) into one store format shared by several nuclei datasets: images and masks in sharded Zarr, one row per patch and one row per nucleus in Parquet. Every pixel, every instance and every class label was checked equal to the original images.npy and labels.npy, and the challenge's counts.csv was reproduced on every patch under its own counting rule, on 31-10-2026. The original data, its licence and its citation requirements are unchanged.

CoNIC is Lizard cut into patches: the same regions, labels and six classes, served as 4,981 non-overlapping patches of 256 px at 0.5 um/px for the challenge. It adds no tissue; it fixes the patches, the evaluation size and the leaderboard that published methods compare against. The companion store nuclei-lizard holds the full regions.

Source and licence

Source: CoNIC, Colon Nuclei Identification and Counting Challenge 2022, Tissue Image Analytics centre, University of Warwick, https://conic-challenge.grand-challenge.org, the public training patches as distributed. The challenge's test set is not public.
Licence: CC BY-NC-SA 4.0, research purposes only, as the release's README and its bundled licence text state. This copy carries the same licence. Commercial use is not allowed.
Cite, as the source asks, both the challenge and the dataset it is cut from: Graham, S., Jahanifar, M., Vu, Q. D., Hadjigeorghiou, G., Leech, T., Snead, D., Raza, S. E. A., Minhas, F., Rajpoot, N. (2021). CoNIC: Colon Nuclei Identification and Counting Challenge 2022. arXiv:2111.14485. Graham, S., Jahanifar, M., Azam, A., and others (2021). Lizard: a large-scale dataset for colonic nuclear instance segmentation and classification. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 684 to 693.

What changed in the conversion

Pixels: images.npy is stored unchanged.
Labels: the instance channel of labels.npy is stored as it is, and each nucleus's class is the class channel's value over its pixels. Every nucleus owns pixels, so none was dropped: 569,861 nuclei in all. Each nucleus's centre is kept as its geometry.
Each patch keeps its challenge name in key (consep_1-0000) and its Lizard region in group (consep_1), from patch_info.csv, so a hold-out can keep a region's patches together and keep the pannuke_* regions apart from PanNuke.
Every patch is train, since every public patch is challenge training data.
counts.csv is not copied: its numbers are reproduced exactly from the store by counting each class's nuclei with any pixel in the central 224 px window.

Layout

native/conic/card.json          what the dataset is: source, licence, scale and its
                                evidence, label origin, class names, arrays present
native/conic/data.zarr          Zarr v3, sharded, Zstd level 3, CRC32C checksums
    images     uint8  (4981, 3, 256, 256)   RGB
    instances  uint32 (4981, 256, 256)      0 background, one id per nucleus
native/conic/images.parquet     one row per patch
native/conic/objects.parquet    one row per nucleus (GeoParquet, point geometry)
patches/conic/                  the same, doubled to 0.25 um/px: four 256 px patches
                                per challenge patch, 19,924 in all, for pooling
                                with datasets at that scale; the challenge is defined at
                                0.5, so the native tier is the one to compare against
views/conic_splits.parquet      every patch as role train

images.parquet columns: image (row number, the index into the arrays and the row of counts.csv), key (the challenge patch name), height, width, um_per_px, split, patient (not reported), slide (not reported), organ (colon), group (the Lizard region), annotated_classes, label_origin, intensity_low and intensity_high (per-channel percentiles), and parent_image, parent_x, parent_y, parent_scale (filled in the patches tier).

objects.parquet columns: image, object (the nucleus's id in instances), native_class, label_origin, x_min, y_min, x_max, y_max (bounding box in pixels) and geometry (the nucleus centre, WKB).

Class names are the challenge's own, never regrouped. A class code is the position in card.json's native_classes, counted from 1, the same order as the class channel and the counts.csv columns; 0 means no class:

code	native_class	nuclei
1	neutrophil	5,082
2	epithelial	282,082
3	lymphocyte	120,933
4	plasma	31,965
5	eosinophil	3,849
6	connective	125,950

Patches per Lizard source:

source	patches
crag	2,304
dpath	1,799
glas	702
pannuke	112
consep	64

Nuclei per patch: minimum 0, median 112, maximum 579; 140 patches hold no nucleus.

Reading the store

With zarr (3.x) and pandas:

import pandas as pd
import zarr

root = zarr.open_group("native/conic/data.zarr", mode="r")
image = root["images"][0]          # (3, 256, 256) uint8
instances = root["instances"][0]   # (256, 256) uint32
images = pd.read_parquet("native/conic/images.parquet")
objects = pd.read_parquet("native/conic/objects.parquet")
classes = objects[objects["image"] == 0].set_index("object")["native_class"]

With pathomix[nuclei-datasets], the package this format belongs to, a store opens with Store.open, a view loads with load_role, and a PyTorch loader comes from pathomix.nuclei.datasets.loading; both read the patches tier.

Limits

Labels are Lizard's, model output refined by pathologists; nothing was corrected.
The content duplicates nuclei-lizard, by design; do not pool the two as if they were different data. The pannuke_* regions duplicate PanNuke tiles at half resolution.
Models trained on this data inherit the non-commercial licence.

Downloads last month: -

Total file size:

3.76 GB

Paper for saahilw/nuclei-conic

CoNIC: Colon Nuclei Identification and Counting Challenge 2022

Paper • 2111.14485 • Published Nov 29, 2021