MoNuSeg is released by its authors under CC BY-NC-SA 4.0. By requesting access you agree to those terms, to cite the MoNuSeg papers, and to share any derivative under the same licence.

MoNuSeg in the nuclei store format

A faithful conversion of the MoNuSeg dataset (Kumar and others, 2017 and 2019) into one store format shared by several nuclei datasets: images and masks in sharded Zarr, one row per image and one row per nucleus in Parquet. Every image's pixels were checked equal to the original TIFF and every stored nucleus checked against its original XML outline on 31-10-2026. The original data, its licence and its citation requirements are unchanged.

Source and licence

Source: MoNuSeg 2018 challenge, https://monuseg.grand-challenge.org, training and test data as distributed: 37 training and 14 test images of 1000 by 1000 pixels from TCGA whole-slide images, every nucleus outlined by hand in Aperio ImageScope, no classes.
Licence: CC BY-NC-SA 4.0, as stated on the source page. This copy carries the same licence. Commercial use is not allowed.
Cite, as the source asks: Kumar, N., Verma, R., Sharma, S., Bhargava, S., Vahadane, A., Sethi, A. (2017). A dataset and a technique for generalized nuclear segmentation for computational pathology. IEEE Transactions on Medical Imaging, 36(7), 1550 to 1560. Kumar, N., Verma, R., Anand, D., and others (2020). A multi-organ nucleus segmentation challenge. IEEE Transactions on Medical Imaging, 39(5), 1380 to 1391.

What changed in the conversion

Pixels: the TIFFs are read as RGB and stored unchanged.
Labels: each XML outline is drawn into one instance map per image, larger outlines first so a small nucleus drawn inside a larger one keeps its pixels; a region with no pixel of its own or no centre inside the image is dropped. 10 of the 30,837 XML regions were dropped this way, leaving 30,827 nuclei. The outline itself is kept as each nucleus's geometry.
Pixel size: each XML states MicronsPerPixel and it is kept per image; four files omit it and take 0.25 from the stated 40x. Three training images state 0.5005 um/px, and their nuclei measure about a third of the usual pixel area, so that value is true.
Patient and slide come from the TCGA file names; organ is not reported by the source.

Layout

native/monuseg/card.json         what the dataset is: source, licence, scale and its
                                 evidence, label origin, class names, arrays present
native/monuseg/data.zarr         Zarr v3, sharded, Zstd level 3, CRC32C checksums
    images     uint8  (51, 3, 1000, 1000)   RGB
    instances  uint32 (51, 1000, 1000)      0 background, one id per nucleus
native/monuseg/images.parquet    one row per image
native/monuseg/objects.parquet   one row per nucleus (GeoParquet, outline polygon)
patches/monuseg/                 the same, cut into 960 patches of 256 px at
                                 0.25 um/px: images within 5 % of that scale are cut as
                                 they are, the three 0.5005 images are doubled first;
                                 a nucleus crossing a patch edge appears in both patches
views/monuseg_splits.parquet     the source's own split as roles: train and test; a
                                 validation split is the user's choice

images.parquet columns: image (row number, the index into the arrays), key (train/<TCGA id> or test/<TCGA id>), height, width, um_per_px (per image), split, patient (TCGA participant), slide (not reported), organ (not reported), group, 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: the native row, the patch offset in resampled pixels and the resampling factor).

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

Class codes: native_classes in card.json holds one class, nucleus, code 1; 0 means background.

split	images	nuclei
train	37	24,130
test	14	6,697

Nuclei per image: minimum 249, median 442, maximum 1862.

Stated pixel sizes:

um/px	images
0.2456	2
0.2457	2
0.2462	2
0.2468	7
0.2472	1
0.248	3
0.2485	3
0.25	4
0.252	13
0.2525	5
0.2527	6
0.5005	3

Reading the store

With zarr (3.x), pandas and shapely:

import pandas as pd
import shapely.wkb
import zarr

root = zarr.open_group("native/monuseg/data.zarr", mode="r")
image = root["images"][0]          # (3, 1000, 1000) uint8
instances = root["instances"][0]   # (1000, 1000) uint32
images = pd.read_parquet("native/monuseg/images.parquet")
objects = pd.read_parquet("native/monuseg/objects.parquet")
outline = shapely.wkb.loads(bytes(objects["geometry"][0]))

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

Outlines are hand-drawn and inherit the source's judgement calls; nothing was corrected. Where the store and a plain polygon rasteriser differ, it is by the boundary pixels: the store fills by pixel centre.
Organ is not reported per image by the source.
Models trained on this data inherit the non-commercial licence.

Downloads last month: -

Total file size:

334 MB