# Multi-Anatomical Ultrafast Ultrasound Dataset: Raw RF Signals acquired with a GE 11L-D probe **Roser Viñals and Jean-Philippe Thiran** Signal Processing Laboratory 5 (LTS5), École Polytechnique Fédérale de Lausanne (EPFL) 1015 Lausanne, Switzerland This repository contains a large-scale ultrasound dataset developed at EPFL. If you use this data in your research, please cite the dataset and the related publications: **Dataset**: R. Viñals and J.-P. Thiran, "Multi-Anatomical Ultrafast Ultrasound Dataset: Raw RF Signals acquired with a GE 11L-D probe," Zenodo, 2026. https://doi.org/10.5281/zenodo.20054331 **Data Paper**: R. Viñals and J.-P. Thiran, "A multi-anatomical in vivo and in vitro ultrafast ultrasound dataset: 20,431 acquisitions with 105 plane waves using the GE 11L-D probe and a vantage system," *Data in Brief*, vol. 69, p. 113259, 2026. https://doi.org/10.1016/j.dib.2026.113259 **Related Publication**: R. Viñals and J.-P. Thiran, "Deep Learning-Based Inpainting for Sparse Arrays in Ultrafast Ultrasound Imaging," *IEEE Transactions on Computational Imaging*, vol. 12, pp. 187-202, 2026. https://doi.org/10.1109/TCI.2025.3648531 --- ## 📊 Dataset Overview | Category | Parameter | Value | |:-----------------|:-------------------------------|:----------------------------------| | **Dataset Size** | **Total Acquisitions** | 20,431 | | | **In Vivo Acquisitions** | 20,121 (16 Volunteers) | | | **In Vitro Acquisitions** | 310 (CIRS 054GS Phantom) | | **Settings** | **System / Probe** | Verasonics Vantage 256 / GE 11L-D | | | **Number of Plane Waves** | 105 angles | | | **Steering Angle Spacing** | 0.32° | | | **Sampling Frequency** | 20.833 MHz | | | **PRF / Peak Voltage** | 9 kHz / 40 V | | **GE 11L-D** | **Center Frequency (f_c)** | 7.3 MHz (75% Bandwidth) | | | **Element Number / Pitch** | 192 elements / 200 μm | | | **Aperture / Elevation Focus** | 38.20 mm / 17 mm | | | **Element Width / Height** | 180 μm / 4 mm | --- ## 🏗️ 1. Dataset Composition The dataset consists of **20,431 total acquisitions** indexed sequentially from `acquisition_00001.npz` to `acquisition_20431.npz`. ### 🔹 In Vivo Data - **Total acquisitions:** 20,121 - **Participants:** 16 volunteers - **File structure:** Each volunteer’s data is provided in a dedicated `.zip` file. **Anatomical Distribution:** - Legs: 4244 acquisitions - Arms: 4422 acquisitions - Abdomen: 2599 acquisitions - Breast: 2875 acquisitions - Neck: 2736 acquisitions - Back: 3246 acquisitions ### 🔹 In Vitro Data (CIRS 054GS) - **Total:** 310 acquisitions - **Archive:** `in_vitro.zip` - **Folder Structure:** - full_speckle/: Uniform phantom regions for the derivation of normalization matrices. - inclusions/: Hyperechoic and anechoic inclusions ### 📂 File Naming Convention Acquisitions are named using a 5-digit ID: `acquisition_00001.npz` -> `acquisition_20431.npz` > **Note:** The numbering is continuous across the entire repository. To identify the specific metadata for any given file ID, refer to the `dataset.csv` master list. > ### 💻 File Format All acquisitions are provided as NumPy arrays. Each `.npz` file contains the RF data and its specific metadata. You can access the arrays using the following keys: ```python import numpy as np # Load the compressed file file = np.load("acquisition_00001.npz") # Access components rf_data = file["data"] # RF raw data region = file["body_region"] # e.g., 'Legs' ("full_speckle" or "inclusions" for in vitro) v_id = file["volunteer_id"] # numerical ID. ("000" for in vitro) ``` --- ## 📋 Metadata & Acquisition Settings All high-level information and hardware configurations are stored in the following files: ### 📄 Global Metadata (`dataset.csv`) - **Description:** A master index mapping every acquisition ID to its source. - **Columns:** `id`, `volunteer_id`, `body_region`. ### ⚙️ Acquisition Parameters (`settings/`) All acquisition and beamforming-related settings are provided in the settings folder: | File | Description | |------|----------------------------------------------------------------------------| | **beamforming_settings.yaml** | Full acquisition, probe, and beamforming metadata (SI units)| | **steering_angles.npy** | 105 steering angles used during plane-wave compounding in radians| | **time_axis.npy** | Global time axis for RF data [s]| | **time_axis_per_angle.npy** | Time axis per steering angle| | **sequence_verasonics_ge11ld_105pws.mat** | Original Verasonics sequence file defining the acquisition parameters| --- ## 📥 Full Dataset Download Links We recommend accessing the dataset using the direct download links listed above or via the S3 interface (see instructions below) to ensure stable and reproducible access. --- ### 🔹 Core Files | File | Direct Download | |------|------------------------------------------------------------------------| | **README** | https://datasets.epfl.ch/epfl_ge11ld_ultrafast_ultrasound/README.md | | **Acquisition metadata (CSV)** | https://datasets.epfl.ch/epfl_ge11ld_ultrafast_ultrasound/dataset.csv | | **Acquisition settings (ZIP)** | https://datasets.epfl.ch/epfl_ge11ld_ultrafast_ultrasound/settings.zip | | **In vitro phantom dataset** | https://datasets.epfl.ch/epfl_ge11ld_ultrafast_ultrasound/in_vitro.zip | --- ### 🔹 In Vivo Volunteer Datasets (ZIP) | Volunteer | Download Link | |----------------------------|------------------------------------------------------------------------------------------| | Volunteer 003 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_003.zip | | Volunteer 005 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_005.zip | | Volunteer 006 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_006.zip | | Volunteer 007 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_007.zip | | Volunteer 008 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_008.zip | | Volunteer 009 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_009.zip | | Volunteer 010 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_010.zip | | Volunteer 011_abdomen | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_011_abdomen.zip | | Volunteer 011_arm | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_011_arm.zip | | Volunteer 011_back_carotid | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_011_back_carotid.zip | | Volunteer 011_breast | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_011_breast.zip | | Volunteer 011_leg | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_011_leg.zip | | Volunteer 012 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_012.zip | | Volunteer 013 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_013.zip | | Volunteer 014 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_014.zip | | Volunteer 015 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_015.zip | | Volunteer 016 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_016.zip | | Volunteer 017 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_017.zip | | Volunteer 018 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_018.zip | | Volunteer 019 | https://datasets.epfl.ch/in_vivo/epfl_ge11ld_ultrafast_ultrasound/volunteer_019.zip | > **Note on Volunteer IDs:** As some volunteers participated in the acquisition of distinct datasets, the volunteer IDs have been synchronized with the ones from the dataset published in: > *R. Viñals and J. -P. Thiran, "Deep Learning-Based Inpainting for Sparse Arrays in Ultrafast Ultrasound Imaging," in IEEE Transactions on Computational Imaging, vol. 12, pp. 187-202, 2026, doi: 10.1109/TCI.2025.3648531.* > **Large File Handling (Volunteer 011):** Due to the high number of acquisitions for **Volunteer 011**, the data has been partitioned into five separate files to facilitate downloading and prevent file size issues. --- ## 🔗 Accessing the dataset via S3 (Using `rclone`) The dataset is hosted on an S3-compatible endpoint and can be accessed directly using **rclone**. ### 1. Create the `rclone` remote Add the following to your `~/.config/rclone/rclone.conf`: `[datasets_epfl] type = s3 provider = Other endpoint = https://datasets.epfl.ch region = us-east-1 env_auth = false access_key_id = secret_access_key =` Or create it automatically: `rclone config create datasets_epfl s3 provider Other endpoint https://datasets.epfl.ch region us-east-1` ### 2. List all files in the dataset `rclone ls datasets_epfl:epfl_ge11ld_ultrafast_ultrasound` ### 3. Download a single file `rclone copy datasets_epfl:epfl_ge11ld_ultrafast_ultrasound/volunteer_005.zip . -P` ### 4. Download the entire dataset `rclone sync datasets_epfl:epfl_ge11ld_ultrafast_ultrasound ./epfl_ge11ld_ultrafast_ultrasound -P` --- ## 🔬 Research Context This dataset was developed to provide a large-scale, multi-anatomical benchmark for ultrafast ultrasound imaging, specifically for training and evaluating deep learning models for beamforming and signal reconstruction. ### 📝 Key Publications * **Data Publication (Primary citation for the dataset):** > R. Viñals and J.-P. Thiran, "A multi-anatomical in vivo and in vitro ultrafast ultrasound dataset: 20,431 acquisitions with 105 plane waves using the GE 11L-D probe and a vantage system," *Data in Brief*, vol. 69, p. 113259, 2026, doi: 10.1016/j.dib.2026.113259. * **Application Study (Subset utilization):** > R. Viñals and J.-P. Thiran, "Deep Learning-Based Inpainting for Sparse Arrays in Ultrafast Ultrasound Imaging," *IEEE Transactions on Computational Imaging*, vol. 12, pp. 187-202, 2026, doi: 10.1109/TCI.2025.3648531. ### ⚠️ Note for Reproducibility (IEEE TCI Study) #### 📐 Angle Selection While the raw `.npz` files in this repository contain **105 plane waves** (the full acquired sequence), the associated publication utilized only **103 plane waves**. * **Methodology:** The two most extreme steering angles were discarded. * **Selection:** Only the 103 plane waves with steering angles closest to **0°** were processed. #### 🧪 Evaluation Subsets To facilitate reproducibility of the results presented in the IEEE TCI paper, the following partitions were used: * **Test Set Volunteers:** 005, 008, and 017. * **Representative Samples:** * **Carotid Artery:** `acquisition_19796.npz` * **Back Muscle:** `acquisition_01509.npz` --- ## 📜 License This dataset is distributed under the **Creative Commons Attribution 4.0 International License (CC BY 4.0)**: https://creativecommons.org/licenses/by/4.0/ --- ## Contact For questions or issues related to this dataset: **Roser Viñals** 📧 roser.vinalsterres@epfl.ch **Jean-Philippe Thiran** 📧 jean-philippe.thiran@epfl.ch