OpenH-RF

Dataset Description

OpenH-RF is a community-driven dataset initiative building the open, shared foundation needed to train and evaluate AI models built on pre-beamformed (channel capture) medical ultrasound measurements.

This dataset is a collection of RF samples and metadata in the zea file format from a variety of tasks and applications, including ultrasound localization microscopy, ultrasound computer tomography, b-mode, flow imaging and more.

This dataset is ready for commercial or non-commercial uses.

Dataset Owner

NVIDIA Corporation

Contributing Organizations

NVIDIA, Stanford University, Eindhoven University of Technology, Tel Aviv University, Siemens Healthineers, Resolve Stroke, University of British Columbia, KAIST, Barreleye Inc., Seoul National University Bundang Hospital, us4us Ltd., University of Colorado Boulder, Vanderbilt University, University of North Carolina at Chapel Hill, Weizmann Institute of Science, University of Oslo, Technical University of Munich, Politecnico di Torino, University of Twente, Weill Cornell Medicine, Worcester Polytechnic Institute, University of Strasbourg, University of Basel, Concordia University, Mosaic Intelligence, PATH Lab - Technion - Israel Institute of Technology, University of Waterloo, Dartmouth College, Provost Ultrasound Lab

Dataset Stewardship

OpenH-RF community & Steering Group

Dataset Creation Date

September 2026

Versioning

v1.0.0

Previous Version(s): no previous version

License/Terms of Use

CC-BY-4.0

Intended Usage

Researchers and builders interested in training ultrasound reconstruction models, RF world foundation models, or RF language models.

Dataset Characterization

Data Collection Method

Hybrid: Manually-Collected, Automatic/Sensors, Synthetic

Labeling Method

Hybrid: Manually-Labeled, Automatic/Sensors, Synthetic

Dataset Format

zea HDF5 format >= 0.1.4. This release contains files with zea_version 0.1.4, 0.1.5, and 0.1.6.

Channel capture data and related meta data are saved in HDF5 data files using a specification defined for the zea library.

Dataset Quantification

32 sub-datasets contain 11,491 HDF5 files totaling 38.4 TB (38,366,766,058,064 bytes; 34.9 TiB) of stored HDF5 data. These sizes count the stored files, including metadata and other embedded arrays; they are not decoded RF array sizes or original-source download sizes. TB uses decimal units (10^12 bytes), and TiB uses binary units (2^40 bytes).

Reference(s)

https://github.com/open-h/OpenH-RF

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal developer teams to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns here.