The dataset aggregates large‑scale pre‑beamformed (channel capture) ultrasound RF recordings and structured metadata to enable raw‑to‑insight model development: reconstruction, flow estimation, quantitative imaging and ultrasound inverse problems. Its scale and raw channel format target foundation models and reconstruction pipelines that need unprocessed sensor measurements rather than beamformed images.
What Sets It Apart
- Raw channel capture in the zea/HDF5 schema: preserves pre‑beamformed RF waveforms and acquisition metadata so methods can learn physics‑aware reconstructions and device‑specific corrections.
- Multi‑task scope and scale: ~39 TB across 32 contributed sub‑datasets and ~11,491 HDF5 files, covering b‑mode, flow, localization microscopy, transcranial and other application domains — suitable for training large reconstruction or RF foundation models.
- Open, permissive license and community stewardship: released under CC‑BY‑4.0 with a community steering group and multi‑institution contributions to encourage reuse and reproducible benchmarking.
Who It's For and Trade-offs
Great fit if you are training or evaluating ultrasound reconstruction models, RF foundation models, or research on sensor‑level inverse problems and flow estimation. Expect substantial storage, I/O, and preprocessing needs (tens of TB); working with this dataset typically requires domain knowledge of ultrasound acquisition, handling device variability, and attention to clinical/privacy constraints. Not ideal if you only need beamformed B‑mode images or small, lightweight example datasets.
