Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative

arXiv:2610.13616 · eess.SP · Submitted 2026-10-09 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.

Rosa: We're looking at the paper: "Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative".

Paper summary: Rosa: So, to wrap up this discussion about Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative, it’s clear that the sheer volume and variety of raw channel data is what matters here.

Dev: The authors are providing a unified format for that forty terabytes of data, which is organized so you can process it reproducibly across different scanners and acquisition schemes <ref:2610.13616#pg3>.

Taro: It’s really about lowering the barrier for researchers who might not be deep in ultrasound to start building things on this foundation.

Rosa: They are opening up possibilities beyond just the images we see, focusing on tasks like tissue characterization and correcting those tricky inaccurate assumptions.

Dev: The next natural step they point towards is training an ultrasound foundation model that uses this raw channel data and conditions it with all that available metadata.

Taro: So it’s moving the field away from just the beamformed image towards methods that actually exploit the full information content of the raw signal.

Rosa: That’s what they are aiming for, using this dataset as a starting point for everyone to build upon and explore new data-driven solutions in ultrasound.

Conclusion: Rosa: So this paper is about Open-H-RF, which is this massive collection of raw ultrasound data they’ve put together from almost forty terabytes across all kinds of scans.

Dev: It’s essentially a shared data initiative, a way for everyone to get access to the actual raw channel information instead of just the final pictures.

Rosa: The authors are bringing together people from academia and industry to contribute this data, spanning everything from living tissue to simulations.

Dev: They’ve built this unified format so you can process it reproducibly no matter what scanner or acquisition setup you started with.

Rosa: What this means is that we’re moving past just looking at the final image and starting to work with the raw signal itself, which opens up ways to do much deeper analysis.

Dev: It’s about building things like foundation models that are trained directly on these raw channel data sets, conditioned by all that acquisition information they include.

Rosa: So it’s a big step toward getting datadriven solutions that really understand the underlying physics of how the ultrasound is made.

Dev: This moves us toward studying things like tissue characterization and motion estimation at a much more fundamental level than we’ve done before.

Rosa: The sheer volume and variety suggest this is going to be a major resource for researchers trying to push past just beamformed images.

NVIDIA Eindhoven University of Technology Department of Computer Science and Engineering, Stanford University

eess.SP

Submitted: 2026-10-09

Updated: 2026-11-17

Comments: Preprint: 15 pages, 15 figures, 3 tables. Dataset (v1.1): https://open-h.github.io/Open-H-RF

Project page: https://open-h.github.io/Open-H-RF

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 100/100

The gist: Medical ultrasound is a widely accessible, real-time imaging modality, yet progress in data-driven ultrasound research is constrained by the scarcity of large…

Key concepts

Open-H-RF
A large, open dataset comprising approximately 40 TB of raw medical ultrasound channel data. It includes acquisition details, probe information, metadata, and reconstruction parameters in a unified format to ensure results are reproducible across different scanners and setups.
Raw Channel Data
The fundamental data collected directly from the ultrasound transducer elements. This includes arrival times and phase information across the acquisition. Using this raw data allows researchers to perform advanced tasks like adaptive beamforming, which goes beyond what standard displayed images can support.
Unified Format (zea file)
A standardized file structure used by Open-H-RF that stores a single acquisition along with all necessary parameters and metadata. This format is designed to capture the raw channel data alongside the specific reconstruction steps applied, making it self-contained for processing.

Terminology

Summary

Medical ultrasound is a widely accessible, real-time imaging modality, yet progress in data-driven ultrasound research is constrained by the scarcity of large, open collections of raw channel data The authors introduce Open-H-RF, a communitydriven open dataset initiative comprising approximately 40 TB of medical ultrasound raw channel data in 19,332 acquisitions contributed by 38 institutions across academia and industry spanning in-vivo, phantom, and simulated data This initiative provides a unified, selfcontained data format that captures raw channel data together with the acquisition, probe, metadata, and reconstruction information required for reproducible processing across scanners and acquisition schemes

The Data Corpus Scope

Open-H-RF comprises 40 TB of raw channel data from 19,332 acquisitions spanning in-vivo, phantom, and simulated sources The dataset is annotated with a number of supported tasks ranging from generalized reconstruction to blood flow imaging and motion estimation Contributions are sorted into five general categories: probe motion compensation, tissue motion estimation, correction of inaccurate image formation assumptions, tissue characterization, and therapy

Data Format and Accessibility

The Open-H-RF initiative uses a unified format built on the zea file specification which stores raw channel data together with all parameters and metadata needed for reproducible processing across scanners Each file holds a single acquisition together with everything needed to process it, organized in a fixed hierarchy of groups including a data group containing the raw channel data Data access requires domain knowledge because reconstructing an image involves a chain of processing steps whose order and settings differ between contributions The corpus is publicly hosted under a permissive CC BY 4.0 license and can be streamed programmatically at scale

Task Coverage and Diversity

The dataset covers a wide range of tasks, with Generalized reconstruction / beamforming being supported by most of the contributions Specific task categories include:

[Probe motion]

[Tissue motion]

[Inaccurate assumptions]

The dataset also includes specialized tasks such as Speed-of-sound estimation, Quantitative ultrasound, and Clutter / reverberation suppression The total task coverage shows broad support across most categories, with the exception of Therapy which contains only one task

Key Research Capabilities

Access to raw channel data enables capabilities beyond what the displayed image can support, such as adaptive beamforming and image reconstruction Channel data preserves the arrival times and phase across the acquisition, allowing for phase-based measures of signal coherence like lag-one coherence (LOC) The variety of data allows for studying how different reconstruction choices affect downstream performance Potential applications include developing a foundation model for raw channel data conditioned on the provided metadata

Conclusion and Future Directions

Open-H-RF is designed to support a number of tasks, as presented in Table I The large volume and variety of raw channel data provided by Open-H-RF opens up new possibilities in developing datadriven solutions that operate on raw channel data A natural next step is the development of an ultrasound foundation model trained on raw channel data and conditioned on the provided metadata Open-H-RF serves as a starting point for the community to build upon, moving ultrasound research beyond the beamformed image and towards methods that exploit the full information content of the raw signal

Pruning redundant reasoning tokens

The authors introduce Open-H-RF, a communitydriven open dataset initiative comprising approximately 40 TB of medical ultrasound raw channel data in 19,332 acquisitions contributed by 38 institutions across academia and industry spanning in-vivo, phantom, and simulated data This initiative provides a unified, selfcontained data format that captures raw channel data together with the acquisition, probe, metadata, and reconstruction information required for reproducible processing across scanners and acquisition schemes

How the training signal is built

The format represents each sequence in an acquisition as a track: a labeled pair of data and scan groups within the same file The pipeline is stored with the contribution and specifies the exact reconstruction steps applied to its data, and each file records the version of the reconstruction library with which it was stored These pipelines are implemented with graphics processing unit (GPU)-accelerated, differentiable operations in a machine learning framework, so that ultrasound reconstruction can be integrated directly into model training

Where the gains come from

The large volume and variety of raw channel data provided by Open-H-RF opens up new possibilities in developing datadriven solutions that operate on raw channel data Research efforts in this direction could explore methods to standardize data across anatomies, probes, etc., as well as neural network architectures that are parameter-efficient and extract useful features The dataset also lowers the barrier for researchers outside the ultrasound community The paper describes v1. p. 6

Where the paper says it

The corpus itself is versioned as well: Open-H-RF is hosted at https://huggingface. co/datasets/ nvidia/Open-H-RF, this paper describes v1.

0, and any release can be directly loaded through zea by its tag, so follow-up work can state exactly which version of the corpus was used

Open-H-RF is designed to support a number of tasks, as presented in Table I The large volume and variety of raw channel data provided by Open-H-RF opens up new possibilities in developing datadriven solutions that operate on raw channel data A natural next step is the development of an ultrasound foundation model trained on raw channel data and conditioned on the provided metadata Open-H-RF serves as a starting point for the community to build upon, moving ultrasound research beyond the beamformed image and towards methods that exploit the full information content of the raw signal

The dataset is annotated with a number of supported tasks ranging from generalized reconstruction to blood flow imaging and motion estimation Contributions are sorted into five general categories: probe motion compensation, tissue motion estimation, correction of inaccurate image formation assumptions, tissue characterization, and therapy

The dataset covers a wide range of tasks, with Generalized reconstruction / beamforming being supported by most of the contributions Specific task categories include:

The dataset also includes specialized tasks such as Speed-of-sound estimation, Quantitative ultrasound, and Clutter / reverberation suppression The total task coverage shows broad support across most categories, with the exception of Therapy which contains only one task

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Adaptive Beamforming using Deep Learning: The system can perform adaptive beamforming by operating directly on raw channel data, which allows for methods that are beyond what the displayed image can support.

  2. Speed-of-Sound Estimation: The AI system can estimate sound speed in tissue by computing phase-based measures of signal coherence, such as common midpoint phase error (CMPE), which is a capability enabled by access to channel data.

  3. Aberration Correction: The system can correct defocusing caused by an incorrect assumed sound speed through methods that utilize the arrival times and phase across the acquisition.

  4. Clutter Suppression: The AI system can implement clutter suppression using aperture-domain methods that exploit differences in coherence and arrival structure across elements, which are lost after summation.

  5. Tissue Characterization: The system can estimate tissue properties like attenuation or backscatter coefficients by accessing the frequency content of the backscattered signal, which is preserved in raw channel data but not in B-mode images.

  6. Quantitative Ultrasound: The AI system can perform quantitative ultrasound tasks such as estimating attenuation or echo-envelope statistics by leveraging the full information content of raw channel data, allowing for distinguishing benign from malignant breast masses and grading liver steatosis.

  7. 3D Reconstruction: The system can perform tracked or matrix imaging to reconstruct volumes in three dimensions, enabled by tracking probe positions and acquiring data from various geometries like IVUS catheters or ring arrays.

  8. Blood Flow Estimation: The system can estimate blood motion with Doppler methods or localize microbubbles to map the microvasculature, optimizing beamforming and angle compounding using channel data access.

Abstract

Medical ultrasound is a widely accessible, real-time imaging modality, yet progress in data-driven ultrasound research is constrained by the scarcity of large, open collections of raw channel data. Existing public datasets are predominantly based on reconstructed images, while the smaller number of open raw channel datasets is generally specialized to individual reconstruction or imaging tasks, limiting the development of general-purpose methods and foundation models that operate directly on ultrasound signals. While interpretive tasks can be performed on reconstructed images, many of the key technical challenges in ultrasound imaging, such as aberration correction and reverberation reduction, require the raw channel data itself. We introduce Open-H-RF, a community-driven open dataset initiative comprising approximately 40 TB of medical ultrasound raw channel data in 19,332 acquisitions, contributed by 39 institutions across academia and industry, spanning in-vivo, phantom, and simulated data. Each dataset is annotated with a number of supported tasks, ranging from generalized reconstruction, speed-of-sound estimation, and aberration correction to blood flow imaging, motion estimation, quantitative ultrasound, and image interpretation. Open-H-RF provides a unified, self-contained data format that captures raw channel data together with the acquisition, probe, metadata, and reconstruction information required for reproducible processing across scanners and acquisition schemes. The corpus is publicly hosted under a permissive CC BY 4.0 license and can be streamed programmatically and selectively at scale. By bringing heterogeneous ultrasound data into a common representation, backed by modern data infrastructure, Open-H-RF lowers the barrier to working with raw channel data and aims to open medical ultrasound research to new communities. See https://open-h.github.io/Open-H-RF.

Sources

Related papers