waterloo-femoralvein (renamed from waterloo-largeartery): sync data card, pipeline grid and figures with GitHub

.gitattributes

@@ -23472,3 +23472,18 @@ ubc/module_C/acquisitions/session_02/session_02_f4326.hdf5 filter=lfs diff=lfs m
 ubc/module_C/acquisitions/session_02/session_02_f4328.hdf5 filter=lfs diff=lfs merge=lfs -text
 ubc/module_C/acquisitions/session_02/session_02_f4330.hdf5 filter=lfs diff=lfs merge=lfs -text
 ubc/module_C/acquisitions/session_02/session_02_f4332.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq0.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq1.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq10.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq11.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq12.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq13.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq14.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq2.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq3.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq4.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq5.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq6.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq7.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq8.hdf5 filter=lfs diff=lfs merge=lfs -text
+waterloo-femoralvein/data/Acq9.hdf5 filter=lfs diff=lfs merge=lfs -text

{waterloo-largeartery → waterloo-femoralvein}/README.md

@@ -1,4 +1,5 @@
 ---
+name: waterloo-femoralvein
 pretty_name: UW-FemVeinRF
 license: cc-by-4.0
 task_categories:
@@ -22,25 +23,23 @@ size_categories:
 
 # UW-FemVein RF
 
-Dataset consisting of raw RF data and vector velocity measurements of femoral vein acquired in studies conducted by VORTEX @
-University of Waterloo.
+![Vector flow cineloop from Acq0.hdf5](assets/Acq0.gif)
+
+*Cine loop of [`data/Acq0.hdf5`](https://huggingface.co/datasets/nvidia/OpenH-RF/blob/main/waterloo-femoralvein/data/Acq0.hdf5), rendered from provided velocity fields.*
 
 ## Dataset Description
 
-This is a dataset consisting of raw RF frames (plane wave) and vector flow
-profiles of the femoral vein in humans, acquired using a programmable research scanner configured for
-high frame rate vector flow imaging. The data was collected as part of studies
-conducted by the VORTEX research group at the University of Waterloo, focusing on
-venous hemodynamics under muscular contraction and head up tilt.
+Raw RF frames (plane wave) and vector flow profiles of the human femoral vein, acquired by the VORTEX research group at the University of Waterloo with a programmable research scanner configured for high-frame-rate vector flow imaging. The data was collected to study venous hemodynamics under muscular contraction and head-up tilt.
 
 ## Dataset Contributor(s)
 
-Hassan Nahas, Jeremy N. Cohen, Eudoxia Zafiris, Skye H.T. Ling, Alfred C.H. Yu, Jason S. Au
-
-Correspondence emails:
-hassan.nahas@uwaterloo.ca
-jason.au@uwaterloo.ca
-
+- Hassan Nahas <hassan.nahas@uwaterloo.ca>
+- Jeremy N. Cohen
+- Eudoxia Zafiris
+- Skye H.T. Ling
+- Alfred C.H. Yu
+- Jason S. Au <jason.au@uwaterloo.ca>
+- VORTEX, University of Waterloo
 
 ## Dataset Creation Date
 
@@ -48,28 +47,40 @@ jason.au@uwaterloo.ca
 
 ## License / Terms of Use
 
-[Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/legalcode.en).
-
-All human studies were approved by the University of Waterloo’s Human Research
-Ethics Board (ORE #46018). All included data was acquired from participants who
-provided both written and verbal consent prior to participating in the study
-regarding public data sharing.
+[Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/legalcode.en). Retain attribution and identify modifications when reusing the data.
 
 ## Intended Usage
 
-Developing, benchmarking, and evaluating methods for ultrasound image
-reconstruction, motion estimation, clutter filtering, multi-angle Doppler
-processing, and vector flow imaging (VFI) in venous imaging.
+Developing, benchmarking, and evaluating methods for ultrasound image reconstruction, motion estimation, clutter filtering, multi-angle Doppler processing, and vector flow imaging (VFI) in venous imaging.
 
 ## Dataset Characterization
 
 - **Data Collection Method:** In vivo imaging of human femoral veins.
 - **Labeling Method:** Label contains target vessel (Anatomy), view direction (Longitudinal), and physiological condition (Contraction + head up tilt).
-- **Acquisition system:**
-  Raw RF data was acquired from programmable research scanners (US4R, US4US, Warsaw, Poland) equipped with an AL2442 linear array transducer.
+- **Acquisition system:** Raw RF data was acquired from programmable research scanners (US4R, US4US, Warsaw, Poland) equipped with an AL2442 linear array transducer.
+
+## Processing the Dataset
+
+The acquisitions can be processed with the `pipeline.yaml` definition in this folder and the [zea library](https://github.com/tue-bmd/zea).
+
+`zea` streams the data from the Hugging Face Hub and processes it according to the pipeline. You can try it out with the following command:
+
+```bash
+zea process \
+  --dataset hf://nvidia/OpenH-RF/waterloo-femoralvein/data/Acq0.hdf5 \
+  --config hf://nvidia/OpenH-RF/waterloo-femoralvein/pipeline.yaml \
+  --n-frames 1 \
+  --save-as png
+```
+
+Alternatively, you can use the `reconstruct.py` [script](https://github.com/open-h/OpenH-RF/blob/main/datasets/waterloo-femoralvein/reconstruct.py) as provided in the [OpenH-RF GitHub repository](https://github.com/open-h/OpenH-RF).
+
+Swap `--n-frames 1 --save-as png` for `--save-as gif` to get the cine loop. In the script, `ZEA_FILE`, `FRAME`, `POWER_THRESHOLD` (the power-Doppler mask threshold, in dB), `VMAX` (velocity colour-scale maximum) and `NO_DEALIAS` at the top select what is reconstructed and overlaid.
 
 ## Dataset Format
 
+[zea v0.1.5](https://github.com/tue-bmd/zea)
+
 Submitted in the [`zea` file format](https://zea.readthedocs.io/en/latest/) (one HDF5 file per acquisition).
 
 Per-sample contents of the converted HDF5:
@@ -97,30 +108,16 @@ Per-sample contents of the converted HDF5:
 | `data/color_doppler` | `[n_frames, z, x]` (+ `coordinates` `[z, x, 3]`) | float32 | m/s | Color Doppler map |
 | `scan/*` | -- | -- | -- | Probe geometry, sampling/center/demodulation frequency, t0 delays, sound speed, transmit angles, focus distances, transmit origins, apodizations, PRI... |
 
-All `coordinates` arrays are per-pixel Cartesian positions in meters, last axis
-`[x, y, z]` (y = 0 for 2-D maps).
-
-## Shipped Example Acquisitions
+All `coordinates` arrays are per-pixel Cartesian positions in meters, last axis `[x, y, z]` (y = 0 for 2-D maps).
 
-One example acquisition is included under `hdf5/` as a representative subset of
-the full dataset:
 
-| File | Subject | Anatomy | View | Condition | Frames |
-|---|---|---|---|---|---|
-| `hdf5/Acq0.hdf5` | 1 | Femoral Vein | Longitudinal | Contraction 8Kg HUT 40 | 9000-12000 |
 
-Each femoral vein acquisition comprises 2 steered plane-wave transmits (`n_tx = 2`), 2048/3072 axial
-samples, and 192 receive channels. Note that example provided was convrted with start-frame=9000 to get to the interesting part.
-
-The frame count in these examples is truncated for demonstration; full acquisitions contain the frame counts described below.
 
 ## Dataset Quantification
 
 **Current OpenH-RF release:** 15 HDF5 files; 941.96 GB (941,958,804,227 bytes) stored; root `zea_version` **0.1.5**. Sizes include all HDF5 contents and use decimal units (MB = 10^6 bytes, GB = 10^9 bytes, TB = 10^12 bytes), not decoded-array memory or original-source download sizes.
 
-Data was collected from 15 participants and consists of 15 acquisitions (one per participant), containing
-24,000 frames of raw RF data per acquisition. Each participant performed isometric plantarflexion 
-contractions at 8 Kg under head up tilt of 40 degrees.
+Data was collected from 15 participants and consists of 15 acquisitions (one per participant), containing 12,000 frames of raw RF data per acquisition. Each participant performed isometric plantarflexion contractions at 8 Kg under head up tilt of 40 degrees. Each femoral vein acquisition comprises 2 steered plane-wave transmits (`n_tx = 2`), 2048/3072 axial samples, and 192 receive channels. 
 
 ## Subject Metadata
 
@@ -129,7 +126,7 @@ contractions at 8 Kg under head up tilt of 40 degrees.
 | **Total Number of Subjects** | 15 |
 | **Total Number of Files (Acquisitions)** | 15 |
 | **Sex Composition** | M: 8, F: 7  |
-| **Total RF Frames** | 360,000 |
+| **Total RF Frames** | 180,000 |
 
 ## Known Issues
 
@@ -140,31 +137,20 @@ contractions at 8 Kg under head up tilt of 40 degrees.
 
 ## Beamforming and Processing
 
-1. **Pre-Filtering:**
-   Channel RF data is pre-filtered to remove hardware artifacts and out-of-band noise using a 5 MHz bandpass filter before beamforming.
-2. **GPU-Accelerated Beamforming (DAS):**
-   Beamforming is carried out via a GPU-accelerated Delay-and-Sum (DAS) module.
+1. **Pre-Filtering:** Channel RF data is pre-filtered to remove hardware artifacts and out-of-band noise using a 5 MHz bandpass filter before beamforming.
+2. **GPU-Accelerated Beamforming (DAS):** Beamforming is carried out via a GPU-accelerated Delay-and-Sum (DAS) module.
    - **Aperture & Apodization:** 128-element Hanning window apodization and an F-number of 1.5.
    - **Dual Angle-Compounding:** Beamforming for B-mode and power Doppler is performed twice with opposite receive angles ($+15^{\circ}$ and $-15^{\circ}$). The final high-resolution beamformed image (HRI) is the average of these two acquisitions:
      $$HRI = \frac{HRI_{+15^{\circ}} + HRI_{-15^{\circ}}}{2}$$
    - **Reconstruction Grid:** Cartesian coordinates mapped by a `PixelMap` representing a lateral range of $[-19, 19]\text{ mm}$ and axial depth of $[10, 60]\text{ mm}$ at $0.1\text{ mm}$ spatial resolution.
-3. **Clutter Filtering:**
-   Clutter filtering is performed on the beamformed ensemble using a high-pass wall filter (normalized cut-off frequencies of 0.05 and 0.1, attenuation of 100 dB).
-4. **Multi-Angle Doppler Frequency Estimation:**
-   Angle-specific Doppler frequencies are computed using an ensemble size of 64 frames with a step size of 1.
-   - For conventional vector velocity estimation, we used the following Tx-Rx angles:
-     Tx: [-10, -10, 10, 10]; Rx: [-10, 10, -10, 10]
-   - For dealiased vector velocity estimation, we used the following Tx-Rx angles:
-     Tx: [-10, -10, -10,-10, 10, 10]; Rx: [-10, -3, 6, 10, -6, 3,10]
-   - Color Doppler map is selected as the first of these (Tx: -10, Rx = -10)
-5. **Vector Doppler Velocity Estimation:**
-   Lateral ($v_x$) and axial ($v_z$) velocity components are computed from the multi-angle Doppler frequency estimates using GPU-accelerated least-squares estimation.
-   Lateral ($v_x$) and axial ($v_z$) dealiased velocity components are computed from the multi-angle Doppler frequency estimates using GPU-accelerated extended least-squares estimation.
-
-The full LITMUS processing pipeline (GPU DAS beamforming + multi-angle vector
-Doppler) is documented in [`convert.py`](convert.py). That script is included for
-provenance and reproducibility; it depends on the LITMUS core Python package and
-the raw acquisition frames, so it is not runnable from this folder alone.
+3. **Clutter Filtering:** Clutter filtering is performed on the beamformed ensemble using a high-pass wall filter (normalized cut-off frequencies of 0.05 and 0.1, attenuation of 100 dB).
+4. **Multi-Angle Doppler Frequency Estimation:** Angle-specific Doppler frequencies are computed using an ensemble size of 64 frames with a step size of 1.
+   - For conventional vector velocity estimation, we used the following Tx-Rx angles: Tx: [-10°, -10°, 10°, 10°]; Rx: [-10°, 10°, -10°, 10°]
+   - For dealiased vector velocity estimation, we used the following Tx-Rx angles: Tx: [-10°, -10°, -10°,-10°, 10°, 10°]; Rx: [-10°, -3°, 6°, 10°, -6°, 3°,10°]
+   - Color Doppler map is selected as the first of these (Tx: -10°, Rx = -10°)
+5. **Vector Doppler Velocity Estimation:** Lateral ($v_x$) and axial ($v_z$) velocity components are computed from the multi-angle Doppler frequency estimates using GPU-accelerated least-squares estimation. Lateral ($v_x$) and axial ($v_z$) dealiased velocity components are computed from the multi-angle Doppler frequency estimates using GPU-accelerated extended least-squares estimation.
+
+The full LITMUS processing pipeline (GPU DAS beamforming + multi-angle vector Doppler) is documented by the contributors. That documentation is provided for provenance and reproducibility; it depends on the LITMUS core Python package and the raw acquisition frames, so it is not runnable from this folder alone.
 
 Papers relevant to our pipeline:
 
@@ -176,42 +162,17 @@ B. Y. S. Yiu and A. C. H. Yu, "Least-Squares Multi-Angle Doppler Estimators for
 
 ## Data Validation
 
-[`reconstruct.py`](reconstruct.py) builds a `zea.Pipeline` of DAS beamforming →
-envelope detection → normalization → log-compression **in code** and
-reconstructs a B-mode directly from `raw_data`, showing the raw-to-image flow
-without any config file. It also saves the pipeline to
-[`pipeline.yaml`](pipeline.yaml) as a shareable recipe. Comparing the
-reconstruction against the stored (LITMUS) B-mode is a sanity check that the
-acquisition parameters and probe geometry are recorded correctly, and serves as
-a reproducible reference reconstruction.
-
-When the vector-flow fields (`vector_velocity_x/z`/`vector_velocity_x/z_deal` + `power_doppler`) are present,
-a third panel overlays the vector velocity field on the stored B-mode. The
-overlay uses `draw_velocity_field`, a single self-contained (numpy + matplotlib)
-helper reproduced inside `reconstruct.py` from the LITMUS core Python package
-(`litmus.core_py.visualization`), so the script has no dependency on the full
-LITMUS GPU stack.
+`reconstruct.py` builds a `zea.Pipeline` of DAS beamforming → envelope detection → normalization → log-compression **in code** and reconstructs a B-mode directly from `raw_data`, showing the raw-to-image flow without any config file. It also saves the pipeline to `pipeline.yaml` as a shareable recipe. Comparing the reconstruction against the stored (LITMUS) B-mode is a sanity check that the acquisition parameters and probe geometry are recorded correctly, and serves as a reproducible reference reconstruction.
 
-The result is written to `reconstruct_output.png`:
+When the vector-flow fields (`vector_velocity_x/z`/`vector_velocity_x/z_deal` + `power_doppler`) are present, the vector velocity field is overlayed on the stored B-mode. The overlay uses `draw_velocity_field`, a single self-contained (numpy + matplotlib) helper reproduced inside `reconstruct.py` from the LITMUS core Python package (`litmus.core_py.visualization`), so the script has no dependency on the full LITMUS GPU stack.
 
-![reference reconstruction](reconstruct_output.png)
-
-### Example Usage of reconstruct.py
-
-```bash
-# Reconstruct the default file (hdf5/Acq0.hdf5) at frame 100
-python reconstruct.py
+The result is written to `reconstruct_output.png`:
 
-# Reconstruct a specific file and frame, and adjust the power-Doppler mask
-python reconstruct.py --input "hdf5/Acq0.hdf5" --frame 700 --power-threshold 58 --vmax 1
-```
+![reference reconstruction](assets/reconstruct_output.png)
 
 ## Ethical Considerations
 
-All human studies were approved by the University of Waterloo’s Human Research
-Ethics Board (ORE #46018). All included data was acquired from participants who
-provided both written and verbal consent prior to participating in the study
-regarding public data sharing.
+All human studies were approved by the University of Waterloo’s Human Research Ethics Board (ORE #46018). All included data was acquired from participants who provided both written and verbal consent prior to participating in the study regarding public data sharing.
 
 ## Citation
 

{waterloo-largeartery → waterloo-femoralvein}/pipeline.yaml

@@ -18,6 +18,14 @@ pipeline:
             - 1.0
     - log_compress
 parameters:
+    xlims:
+    - -0.019
+    - 0.019
+    zlims:
+    - 0.01
+    - 0.06
+    grid_size_x: 381
+    grid_size_z: 501
     dynamic_range:
     - -50
     - 0