Mannheim Urology HyperSPECTRAL Imaging Atlas (ManUSPECTRAL)

Welcome to ManUSPECTRAL, a human intraoperative hyperspectral imaging atlas acquired during 159 urological procedures. It contains 1,095 recordings with expert polygon annotations covering 17 anatomical organ classes.

The recordings were acquired using medically certified TIVITA systems from Diaspective Vision: The extracoporeal TIVITA Surgery and the laproscopic TIVITA Mini camera. TIVITA Surgery images have a spatial shape of 480 × 640 pixels and TIVITA Mini images have a spatial shape of 540 × 720 pixels. Both systems record 100 spectral channels from 500 nm to 1,000 nm at approximately 5 nm spectral resolution. The release contains raw hyperspectral data, reconstructed RGB and tissue parameter images, annotations, metadata, and preprocessed files for use with the htc framework.

The public representation distinguishes 17 overall classes, which result in 22 fine-grained labels. The three classes artery, vein, and fat are further separated into multiple labels, as can be seen in the Dataset labels table below.

📝 More details can be found in our publication: Rathenborg, A., Klütz, V., Grützner, P. et al. ManUSPECTRAL – the Mannheim Urology HyperSPECTRAL Imaging Atlas of 17 Physiological Human Organs. BJU International (2026). https://doi.org/10.1111/bju.70467

Dataset labels

The numbering follows the 17 classes used in the publication. Anatomically distinct artery, vein, and fat subclasses are stored as separate labels, resulting in 22 dataset labels.

Download

The public archive contains the raw data, annotations, metadata, and precomputed intermediates. The full dataset can be downloaded at:

• https://e130-hyperspectal-tissue-classification.s3.dkfz.de/data/ManUSPECTRAL.zip
• SHA256: d190ecea364ceddffeb3f1a4f527556b651e35df93ea584600ff4743c3e0aa64

For large downloads, we recommend a resumable download tool such as wget and verification of the SHA256 checksum before extraction.

Usage

We recommend using the dataset with the htc framework, which provides efficient access to HSI cubes, annotations, metadata, and tissue parameter images.

# Make the dataset available to htc
export PATH_Tivita_ManUSPECTRAL=/path/to/ManUSPECTRAL

# Install the framework
pip install imsy-htc

The following example selects one recording and loads its main data products:

import numpy as np

from htc import DataPath, LabelMapping, settings

path = next(DataPath.iterate(settings.data_dirs.manuspectral))

# Device-dependent HSI cube shape:
# TIVITA Mini: (540, 720, 100)
# TIVITA Surgery: (480, 640, 100)
cube = path.read_cube()
print(path.device_name, cube.shape)

# Expert polygon annotation
segmentation = path.read_segmentation("polygon#annotator1")
print(path.annotated_labels("polygon#annotator1"))

# Example: median tissue oxygenation in one annotated class
label_name = path.annotated_labels("polygon#annotator1")[0]
label_index = LabelMapping.from_path(path).name_to_index(label_name)
sto2 = path.compute_sto2(cube)
print(float(np.median(sto2[segmentation == label_index])))

Dataset structure

The two acquisition systems share one label mapping but use device-specific settings because their spatial image dimensions differ.

ManUSPECTRAL/
├── data/
│   ├── dataset_settings.json
│   ├── meta.schema
│   ├── Tivita_Mini/
│   │   ├── dataset_settings.json
│   │   └── subjects/MXXXX/YYYY_MM_DD_HH_MM_SS/
│   │       ├── YYYY_MM_DD_HH_MM_SS_SpecCube.dat
│   │       ├── <RGB and tissue parameter images>
│   │       └── annotations/
│   │           ├── YYYY_MM_DD_HH_MM_SS_meta.json
│   │           ├── YYYY_MM_DD_HH_MM_SS#polygon#annotator1#label#binary.png
│   │           └── YYYY_MM_DD_HH_MM_SS#polygon#annotator1#label#coordinates.csv
│   └── Tivita_Surgery/
│       ├── dataset_settings.json
│       └── subjects/PXXXX/YYYY_MM_DD_HH_MM_SS/
│           ├── YYYY_MM_DD_HH_MM_SS_SpecCube.dat
│           ├── <RGB, camera metadata, and tissue parameter images>
│           └── annotations/
└── intermediates/
    ├── preprocessing/
    │   ├── L1/
    │   └── parameter_images/
    ├── segmentations/
    ├── tables/
    ├── view_all/
    └── profiles/

Image names include the acquisition system so that every recording remains uniquely addressable:

SUBJECT#YYYY_MM_DD_HH_MM_SS#Tivita_Mini
SUBJECT#YYYY_MM_DD_HH_MM_SS#Tivita_Surgery

Frequently asked questions

If you have any additional questions about the dataset or its usage, feel free to head over to htc repository and connect with us 😊. Alternatively, you can contact us at info@spectralverse-heidelberg.org.

License

The ManUSPECTRAL publication is available under the Creative Commons Attribution 4.0 International License.

If you use ManUSPECTRAL, please cite the accompanying publication: ManUSPECTRAL – the Mannheim Urology HyperSPECTRAL Imaging Atlas of 17 Physiological Human

@article{rathenborg_manuspectral_2026,
  author       = {Rathenborg, Alexander and Klütz, Vera and Grützner, Paula and Hartung, Luisa and Demmel, Greta and Bender, Christoph Julien and Sellner, Jan and Seidlitz, Silvia and Haney-Aubert, Caelan Max and Fattal, Wahid and Clanget, Tilmann and Özdemir, Berkin and Knoedler, Samuel and Salg, Gabriel Alexander and Renner, Luisa and Carl, Nicolas and Yu, Cuyler and Leung, Charlene and Edelmann, Dominic and Schreck, Nicholas and Michel, Maurice Stephan and Maier-Hein, Lena and Kowalewski, Karl-Friedrich},
  url          = {https://spectralverse-heidelberg.org/ManUSPECTRAL},
  date         = {2026-10-07},
  doi          = {10.1111/bju.70467},
  issn         = {1464-410X},
  journaltitle = {BJU International},
  title        = {ManUSPECTRAL -- the Mannheim Urology HyperSPECTRAL Imaging Atlas of 17 Physiological Human Organs},
}

Acknowledgements

The authors gratefully acknowledge the data storage service SDS@hd supported by the Ministry of Science, Research and the Arts Baden-Württemberg (MWK) and the German Research Foundation (DFG) through grant INST 35/1314-1 FUGG and INST 35/1503-1 FUGG. Furthermore, the authors gratefully acknowledge the support from the NCT (National Center for Tumor Diseases in Heidelberg, Germany) through its structured postdoc program and the Surgical Oncology program. We also acknowledge the support through state funds approved by the State Parliament of Baden-Württemberg for the Innovation Campus Health + Life Science Alliance Heidelberg Mannheim from the structured postdoc program for Alexander Rathenborg (born Studier-Fischer): Artificial Intelligence in Health (AIH) - A collaboration of DKFZ, EMBL, Heidelberg University, Heidelberg University Hospital, University Hospital Mannheim, Central Institute of Mental Health, and the Max Planck Institute for Medical Research. This project was supported by the German Cancer Research Center (DKFZ) and the Helmholtz Association under the joint research school HIDSS4Health (Helmholtz Information and Data Science School for Health) and by the Else-Kröner-Fresenius-Stiftung (2021_EKEA.166). Furthermore, we acknowledge the support through the DKFZ Hector Cancer Institute at the University Medical Center Mannheim.