KiTS23 — Kidney Tumor Segmentation Challange

This dataset is derived from the 2023 Kidney and Kidney Tumor Segmentation Challenge (KiTS23), the third iteration of the KiTS challenge series, organized by the University of Minnesota Robotics Institute, Helmholtz Imaging (DKFZ), and the Cleveland Clinic in conjunction with MICCAI 2023. The original cohort comprises 489 publicly released contrast-enhanced preoperative CT scans from patients who underwent cryoablation, partial nephrectomy, or radical nephrectomy for suspected renal malignancy between 2010 and 2022 at an M Health Fairview medical center. Scans are in either corticomedullary or nephrogenic contrast phase, and each was manually annotated for kidney, tumor, and cyst regions by a team of trainees under expert radiologist and urologic oncologist supervision.

Here we provide a Hugging Face ready, 2D slice-level version of the dataset: every 3D volume is decomposed into its axial slices, each paired with its segmentation mask and the patient-level clinical metadata that accompanies the original release.

📦 Dataset Structure

Each entry corresponds to a single axial CT slice:

• volume_id → Identifier of the source case/volume (e.g. 210 for case_00210)
• slice_id → Index of the axial slice within that volume
• image → Axial CT slice
• mask → Segmentation mask for the same slice (see class table below)
• gender → Patient gender
• age_at_nephrectomy → Patient age at time of surgery, in years
• bmi → Body mass index
• malignant → Whether the mass was pathologically confirmed malignant
• tumor_histologic_subtype → Histologic subtype (e.g. clear_cell_rcc, papillary, chromophobe, oncocytoma, angiomyolipoma, …)
• pathology_t_stage → Pathologic T stage (e.g. 1a, 1b, 2, 3, …)
• radiographic_size → Largest tumor diameter measured radiographically, in cm
• pathologic_size → Largest tumor diameter measured on the surgical specimen, in cm
• aua_risk_score → AUA risk category (low_risk, intermediate_risk, high_risk, very_high_risk, benign)

Slices from the same patient share a volume_id, split by volume_id, never by row, or you will leak the same patient across train and test.

🏷️ Segmentation Classes

The mask field encodes four classes, matching the original KiTS23 label convention:

Value	Class	Definition
0	Background	Everything not belonging to a class below
1	Kidney	All parenchyma and non-adipose tissue within the hilum
2	Tumor	Masses on the kidney preoperatively suspected of being malignant
3	Cyst	Kidney masses radiologically (or pathologically) determined to be cysts

The KiTS challenge evaluates using Hierarchical Evaluation Classes (HECs) rather than raw labels, which is worth mirroring if you want comparable numbers:

HEC	Composition
Kidney and Masses	Kidney + Tumor + Cyst
Kidney Mass	Tumor + Cyst
Tumor	Tumor only

⚙️ Preprocessing

• 3D volumes (imaging.nii.gz) and label maps (segmentation.nii.gz) were decomposed into axial slices
• Hounsfield Unit intensities were windowed and normalized to 8-bit for image storage
• Masks were kept as nearest-neighbor, lossless single-channel images preserving the integer class values {0, 1, 2, 3}
• Slice-level rows inherit the patient-level clinical fields from the KiTS23 clinical metadata; missing values are left as nulls
• Only the publicly released training cohort (489 cases) is included — the 110-case KiTS23 test set was never publicly distributed with labels

🚀 Usage

from datasets import load_dataset
import matplotlib.pyplot as plt
import numpy as np

ds = load_dataset("chehablab/KiTS23", split="train")
sample = ds[1500]

img = np.array(sample["image"])
mask = np.array(sample["mask"])

fig, ax = plt.subplots(1, 2, figsize=(10, 5))
ax[0].imshow(img, cmap="gray")
ax[1].imshow(img, cmap="gray")
ax[1].imshow(np.ma.masked_where(mask == 0, mask), cmap="jet", alpha=0.5, vmin=0, vmax=3)
for a in ax:
    a.axis("off")
fig.suptitle(
    f"case_{sample['volume_id']:05d} | slice {sample['slice_id']} | "
    f"{sample['gender']}, {sample['age_at_nephrectomy']}y | "
    f"{sample['tumor_histologic_subtype']} | T{sample['pathology_t_stage']}"
)
plt.show()

Patient-level splitting:

import numpy as np

vols = np.array(sorted(set(ds["volume_id"])))
rng = np.random.default_rng(0)
rng.shuffle(vols)
val_vols = set(vols[: int(0.2 * len(vols))].tolist())

val_ds   = ds.filter(lambda x: x["volume_id"] in val_vols)
train_ds = ds.filter(lambda x: x["volume_id"] not in val_vols)

📚 Citation

If you use this dataset, please acknowledge Chehab Lab and cite the original KiTS publications:

@article{heller2019kits19,
  title   = {The KiTS19 Challenge Data: 300 Kidney Tumor Cases with Clinical Context, CT Semantic Segmentations, and Surgical Outcomes},
  author  = {Heller, Nicholas and Sathianathen, Niranjan and Kalapara, Arveen and Walczak, Edward and Moore, Keenan and Kaluzniak, Heather and Rosenberg, Joel and Blake, Paul and Rengel, Zachary and Oestreich, Makinna and others},
  journal = {arXiv preprint arXiv:1904.00445},
  year    = {2019},
  url     = {https://arxiv.org/abs/1904.00445}
}

@article{heller2021state,
  title   = {The state of the art in kidney and kidney tumor segmentation in contrast-enhanced {CT} imaging: Results of the {KiTS19} challenge},
  author  = {Heller, Nicholas and Isensee, Fabian and Maier-Hein, Klaus H. and Hou, Xiaoshuai and Xie, Chunmei and Li, Fengyi and others},
  journal = {Medical Image Analysis},
  volume  = {67},
  pages   = {101821},
  year    = {2021},
  doi     = {10.1016/j.media.2020.101821}
}

@article{heller2023kits21,
  title   = {The {KiTS21} Challenge: Automatic segmentation of kidneys, renal tumors, and renal cysts in corticomedullary-phase {CT}},
  author  = {Heller, Nicholas and Isensee, Fabian and Trofimova, Dasha and Tejpaul, Resha and Zhao, Zhongchen and Chen, Huai and others},
  journal = {arXiv preprint arXiv:2307.01984},
  year    = {2023},
  url     = {https://arxiv.org/abs/2307.01984}
}

Challenge homepage: https://kits-challenge.org/kits23/

Original data repository: https://github.com/neheller/kits23

---

📜 License

The KiTS image and segmentation data are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license, and this derivative dataset inherits the same terms.

You may share and adapt the data provided you give appropriate credit, do not use it for commercial purposes, and distribute any derivative works under the same license.

Chehab Lab @ 2026