Skip to content

Ultrasound Post-Processing Filter Design#

Authors: Walter Simson (NVIDIA)
Supported platforms: x86_64, aarch64
Language: python
Last modified: January 21, 2026
Latest version: 0.1.0
Minimum Holoscan SDK version: 3.9.0
Tested Holoscan SDK versions: 3.9.0, 3.10.0
Contribution metric: Level 1 - Highly Reliable

A collection of tools for developing and deploying realtime ultrasound post-processing filters.

Designer UI

Mission#

Enable the ultrasound community (academia and industry) to have reproducible, real-time post-processing.

Overview#

This project includes an Ultrasound Post-Processing Filter Designer and light-weight library that enables you to design and run your ultrasound post-processing filter collection in Holoscan, all from a small and simple YAML config file.

The Filter Designer enables you to create the YAML configs working on Ultrasound File Format (UFF) files (beamformed but not log compressed). You can load and filter your own data locally.

The included Holoscan Runner tool enables you to run those configurations in real-time using the Holoscan SDK.

Prerequisites#

Hardware#

  • Linux device with NVIDIA GPU
  • Tested with x86_64 workstation + RTX Ampere 6000 GPU

Software#

  • CUDA Driver R580 or later
  • See additional HoloHub CLI prerequisites here

Quick Start#

First, clone this repository: 

git clone https://github.com/nvidia-holoscan/holohub.git

Then, run the containerized application.

To run the filter designer webpage for assembling filter pipelines and tuning filter parameters: 

./holohub run ultrasound_postprocessing designer

To run a filter pipeline preset in real time with Holoscan SDK: 

./holohub run ultrasound_postprocessing realtime

Install and Run with Python Packaging#

As an alternative to HoloHub CLI, you can instead design and run filters with project Python entrypoints.

Install this project with your Python package manager: 

python3 -m pip install git+ssh://github.com/nvidia-holoscan/holohub.git@main#subdirectory=applications/ultrasound_postprocessing

Then run your application of choice: 

python3 -m streamlit run applications/ultrasound_postprocessing/ultra_post/app/streamlit_app.py


python3 -m ultra_post.app.holoscan_app [--uff path/to/myfile.uff] [--fps 10]

The sample dataset used in this project is available at http://www.ustb.no/datasets.

Key Concepts#

Glossary#

  • Filters are stateful, CuPy-based Python functions that operate on ultrasound images.
  • The Filter Registry is a collection of pre-defined filters for use in apps such as the Filter Designer and in Holoscan SDK pipelines.
  • Pipelines are a set of one or more filter objects assembled into a serialized graph or "chain".
  • Presets are instructions for configuring a pipeline out of specific filters.
  • Operators are Holoscan SDK "nodes" wrapping filter definitions for real-time serial execution.

Development#

Project Structure#

ultrasound_postprocessing
├── CMakeLists.txt
├── Dockerfile
├── docs
├── external
├── metadata.json
├── plans
├── presets
├── pyproject.toml
├── README.md
├── tests
├── tools
└── ultra_post (Python module)

Review Existing Filters#

./holohub run ultrasound_postprocessing list-filters

Validate a preset#

./holohub run ultrasound_postprocessing validate <presets/my-preset.yml>

Run Tests#

./holohub test ultrasound_postprocessing

Contributing#

We welcome contributions that align with our mission to enable the ultrasound community to have reproducible, real-time post-processing.

Ways to Contribute#

  1. Presets: Create a processing pipeline preset (YAML) and save it to presets/.
  2. Filters:
    • Add your filter implementation to ultra_post/filters/.
    • Register it in ultra_post/filters/registry.py (add to FILTERS and DEFAULT_PARAMS).
  3. Improvements: Bug fixes and performance optimizations are highly encouraged.
    • Our goal is to have code that is easy to understand and highly performant.

Please see the HoloHub Contributing Guide for developer guidance.

Code Style#

  • Concise & Approachable: We prioritize readability so new users can quickly understand the post-processing logic.
  • Performant: We aim for code that is both expressive and performant, enabling real-time processing.

License#

This project is licensed under the Apache-2.0 License. See the HoloHub LICENSE file for details.

Authors#

  • Walter Simson - NVIDIA Holoscan Team

Acknowledgments#

  • NVIDIA Holoscan Team
  • Open source community contributors

Reference Data#

This project uses UFF sample data hosted by the Ultrasound Toolbox.

H. Liebgott, A. Rodriguez-Molares, F. Cervenansky, J. A. Jensen and O. Bernard, "Plane-Wave Imaging Challenge in Medical Ultrasound," 2016 IEEE International Ultrasonics Symposium (IUS), Tours, France, 2016, pp. 1-4, doi: 10.1109/ULTSYM.2016.7728908.