SUPRA: Open Source Software Defined Ultrasound Processing for Real-Time Applications
By the Chair for Computer Aided Medical Procedures
Main contributors:
- Rüdiger Göbl
- Dr. Christoph Hennersperger
Supported by EDEN2020
A 2D and 3D Pipeline from Beamforming to B-mode
SUPRA is an open-source pipeline for fully software defined ultrasound processing for real-time applications. Covering everything from beamforming to output of B-Mode images, SUPRA can help reproducibility of results and allows modifications to the image acquisition.
Including all processing stages of a usual ultrasound pipeline, it can be executed in 2D and 3D on consumer GPUs in real- time. Even on hardware as small as the CUDA enabled Jetson TX2 SUPRA can be run for 2D imaging in real-time.
Getting started
To get an overview of the concepts behind SUPRA and how you can use and modify it, head over to the wiki, or have a look at the recording of SUPRA-con at YouTube SUPRA-con Playlist.
License
LGPL v2.1 see LICENSE
Publication
If you use SUPRA for your research, please cite our work https://doi.org/10.1007/s11548-018-1750-6
Göbl, R., Navab, N. & Hennersperger, C. , "SUPRA: Open Source Software Defined Ultrasound Processing for Real-Time Applications" Int J CARS (2018). https://doi.org/10.1007/s11548-018-1750-6
@Article{Goebl2018supra,
author="G{\"o}bl, R{\"u}diger and Navab, Nassir and Hennersperger, Christoph",
title="SUPRA: open-source software-defined ultrasound processing for real-time applications",
journal="International Journal of Computer Assisted Radiology and Surgery",
year="2018",
month="Mar",
day="28",
issn="1861-6429",
doi="10.1007/s11548-018-1750-6",
url="https://doi.org/10.1007/s11548-018-1750-6"
}
Building
Requirements
- cmake ≥ 3.4
- gcc ≥ 4.8 or min. Visual Studio 2015 (Compiler needs to be supported by CUDA! For that, see the CUDA installation instructions.)
- QT ≥ 5.5
- TBB
- CUDA ≥ 10.0
Build instructions (Ubuntu 16.04 / 18.04)
Install CUDA (≥ 10.0) as described by NVIDIA https://developer.nvidia.com/cuda-downloads . Keep in mind that the C++ host compiler has to be supported by the CUDA version. (Check http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html and http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html for details.)
Build requirements
apt-get install cmake cmake-gui qt5-default libtbb-dev libopenigtlink-dev git
SUPRA
mkdir -p $HOME/git && cd $HOME/git #(or your favorite directory for repositories)
git clone https://github.com/IFL-CAMP/supra.git
cd supra
mkdir -p build && cd build
cmake-gui ..
-
Configure
-
For systems with multiple gcc versions, make sure to select one supported by the installed CUDA version
-
You might need to specify the CUDA toolkit directory (usually "
/usr/local/cuda
") -
Configure & Generate, then close cmake and build
-
Build SUPRA
make -j5
-
Start SUPRA: See below
Building with PyTorch inference (via libtorch)
-
Download the stable libtorch for CUDA 10.0 from pytorch.org
https://download.pytorch.org/libtorch/cu100/libtorch-win-shared-with-deps-latest.zip https://download.pytorch.org/libtorch/cu100/libtorch-shared-with-deps-latest.zip
Tested version: 1.1.0:
https://download.pytorch.org/libtorch/cu100/libtorch-win-shared-with-deps-1.1.0.zip https://download.pytorch.org/libtorch/cu100/libtorch-shared-with-deps-1.1.0.zip
-
Install cuDNN
-
Unpack libtorch (e.g. to
supra/external
on windows or/opt/
on linux) -
Activate
SUPRA_TORCH
in cmake (e.g. in the GUI, or via-DSUPRA_TORCH=ON
) -
Point cmake to the libtorch you just extracted (e.g.
supra/external/libtorch/share/cmake/Torch
or/opt/libtorch/share/cmake/Torch
) -
Configure and build =======
Demo (No US-system required!)
Change to your build directory. If you used the commands above, you can execute
cd $HOME/git/supra/build
Start the SUPRA GUI with a demo config file
src/GraphicInterface/SUPRA_GUI -c data/configDemo.xml -a
Where -c
defines the config file to load and -a
is autostart.
This shows a complete ultrasound pipeline running on your computer from raw channel data recorded with a Cephasonics system and a 7MHz linear probe. With the dropdown menu "Preview Node", you can select which stage of the pipeline to inspect. For the final state of the image, select "SCAN", which shows the output of the scan-converter - the B-mode.
Start the SUPRA GUI with a demo 3D config file
src/GraphicInterface/SUPRA_GUI -c data/configDemo3D.xml -a
Used libraries
SUPRA uses tinyxml2 which is awesome and distributed under the zlib-license. For more details see the tinyxml2 README and (http://grinninglizard.com/tinyxml2/index.html and https://github.com/leethomason/tinyxml2)
SUPRA also uses jsoncpp for more structured data handling which is distributed under the MIT license. For more details see the jsoncpp README
On windows, ROS-message headers generated with rosserial are used and are included in the source. On Linux, the usual ROS-libraries are used during build. (roscpp, geometry_msgs)
SUPRA additionally uses the Intel Thread Building Blocks (but does not provide them) in their Apache 2.0 licensed form. https://www.threadingbuildingblocks.org/
Finally, it can be built against
- QT (LGPLv3)
- IGTL (BSD 3clause)
- CAMPVis (Apache 2.0) (unfortunately, the respective QT5 version is not yet public)
Alternate Builds
REST interface instead of graphical interface
Build requirements
apt-get install cmake cmake-gui libtbb-dev libopenigtlink-dev libcpprest-dev libboost-all-dev git
SUPRA
mkdir -p $HOME/git && cd $HOME/git #(or your favorite directory for repositories)
git clone https://github.com/IFL-CAMP/supra.git
cd supra
mkdir -p build && cd build
cmake-gui .. -DSUPRA_INTERFACE_REST=ON -DSUPRA_INTERFACE_GRAPHIC=OFF
-
Configure
-
For systems with multiple gcc versions, make sure to select one supported by the installed CUDA version
-
You might need to specify the CUDA toolkit directory (usually "
/usr/local/cuda
") -
Configure & Generate, then close cmake and build
-
Build SUPRA
make -j5
-
Start SUPRA: See below
Rest Interface Queries
SUPRA accepts GET and POST requests.
The IP address / hostname SUPRA can be reached with is referred as SUPRA_URL
below.
GET REQUESTs
SUPRA_URL/nodes/[var]
where var can be input
to return all input nodes, output
to get only the output nodes and empty or all
to return all nodes regardless of their types.
The shape of the object in response's body will be
{"nodeIDs":[String]}
.
SUPRA_URL/parameters
returns all parameters for one node.
The shape of the object to send with the body is {"nodeID":"ID"}
.
POST REQUESTs
SUPRA_URL/parameters
sets the value of a parameter of a node. The request has to be shaped like below.
{
"nodeID":"id",
"parameterID":"id",
"value":"value"
}
Demo (No US-system required!)
Change to your build directory. If you used the commands above, you can execute
cd $HOME/git/supra/build
Start the SUPRA GUI with a demo config file
src/RestInterface/SUPRA_REST data/configDemo.xml
Additionaly used libraries
See above for most used libraries. This build uses additionally:
- Microsoft C++ Rest SDK >=2.8 - (BSD 3clause)
- Boost (MIT)
Generate a self-building deb source file
Build Requirements:
apt-get install debmake
cd supra
debmake -cc >> copyright
mkdir -p build && cd build
cmake ..
make package_source
The deb file can be found in the 'binpackages' folder.
When installing the deb file in a system the package will try to build with the standard cmake configuration on that system.
Acknowledgement
SUPRA logo by Raphael Kretz.
This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No. 688279.