lts5 / ussr-ius2017

USSR: An UltraSound Sparse Regularization framework

USSR: An UltraSound Sparse Regularization framework

Adrien Besson¹, Dimitris Perdios¹, Florian Martinez¹, Marcel Arditi¹, Yves Wiaux² and Jean-Philippe Thiran^{1, 3}

¹Signal Processing Laboratory (LTS5), Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland

²Institute of Sensors, Signals and Systems, Heriot-Watt University , UK

³Department of Radiology, University Hospital Center (CHUV), Switzerland

Code used to reproduce the results presented in this paper, accepted at the IEEE International Ultrasonics Symposium 2017

Requirements

• Linux (code tested on Linux Mint 18.1)
• Python >= 3.5 (code tested with default Python 3.5.2 and Anaconda 4.2.0)
• pip (with setuptools and wheel)
• MATLAB (code tested on MATLAB R2016b and R2017a)
• git
• (Optional) NVIDIA GPU with Kepler or newer architecture and its driver (version >= 367.48). The CUDA toolkit is not required to run the code.

Installation

1. Clone the repository (--recursive is used to download the PICMUS submodule when cloning the repo)

   git clone --recursive https://github.com/LTS5/USSR-IUS2017.git

2. Enter in the USSR-IUS2017 folder

   cd USSR-IUS2017

3. (Optional) Create a dedicated Python environment

   • Using Anaconda:

     conda create -n ussr_ius2017 python=3.5
     source activate ussr_ius2017

   • Using pyenv:

     pyvenv /path/to/new/virtual/env . /path/to/new/virtual/env/bin/activate

4. Install Python dependencies from python_requirements.txt. Depending on your installation, pip may refer to Python 2 (you can verify with pip -V). In that case, use pip3 instead of pip.

   pip install --upgrade pip
   pip install -r python_requirements.txt

Reproduce the results of the paper

1. Launch the bash file

   • To run the GPU version of the code

     ./ius_results.sh 'GPU'

   • To run the CPU version of the code

     ./ius_results.sh 'CPU'

   The bash file does the following operations:

   • Download the PICMUS datasets (may take some time depending on your Internet connection)
   • Reconstruct the images
   • Compute and export the metrics
   • Generate and export the figures

2. The folder USSR-IUS2017/results contains the metrics reported in Table I of the paper and the B-mode images displayed on Figure 2 of the paper

Remarks

The GPU code uses NVIDIA CUDA and therefore is only compatible with NVIDIA GPUs. It should support every GPU architecture starting from Kepler (compute capability >= 3.0). The code has been tested on the following architectures:

• GPUs:
  • NVIDIA Titan X (cc = 6.1)
  • NVIDIA GeForce GTX 1080 Ti (cc = 6.1)
  • NVIDIA GeForce GT 630 (cc = 3.5)
• CPU:
  • Intel(R) Core(TM) i7-4930K CPU @ 3.40GHz

The CPU version of the code may take a very long time. It took 10 minutes to run 200 iterations of FISTA on the above mentioned CPU.

Developers

• Florian Martinez ([email protected]): CUDA and C/C++
• Dimitris Perdios ([email protected]): Python interface
• Adrien Besson ([email protected]): Matlab interface

License

License for non-commercial use of the software. Please cite the following paper when using the code.

ussr-ius2017's Issues

git clone --recursive clones the latest commit

Should be changed when future commits in PICMUS