Parallel/GPU level set volume segmentation using OpenCL
C++ C
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.

README.md

OpenCL Level Set Segmentation

Segmented brain from synthetic MR images using the level set method

This is an implementation of level set image segmentation using OpenCL. The speed function is defined as -alpha*(epsilon-|T-intensity|)+(1-alpha)*curvature.

See http://www.thebigblob.com/level-set-segmentation-on-gpus-using-opencl/ for more information on level sets and this implementation.

See LICENSE file for license information.

Dependencies

  • OpenCL (You need an OpenCL implementation installed (AMD, NVIDIA, Intel, Apple...)
  • Submodules SIPL and OpenCLUtilities (no need to install separately, however remember to initialize them "git submodule update" if you are cloning this repo)
  • GTK2, which SIPL use for displaying the results (sudo apt-get install libgtk2.0-dev)

Compiling

Use CMake and the provided CMakeLists.txt file to compile this program. To compile on the run the example on Ubuntu, do the following:

cmake .
make
./levelSetSeg example_data/mr_brain.mhd result.mhd 100 100 100 10 2000 125 40 0.05 125 255

Usage

The arguments for the program are: ./levelSetSeg inputFile.mhd outputFile.mhd seedX seedY seedZ seedRadius iterations threshold epsilon alpha [level window]

inputFile.mhd is the metadata file of the volume you want to process and outputFile.mhd the path to where the result should be stored. seedX, seedY, seedZ defines a spherical seed point with a radius seedRadius. Iterations is the total number of iterations and threshold, epsilon, alpha are parameters for the speed function -alpha*(epsilon-|T-intensity|)+(1-alpha)*curvature. If the level and window arguments are set, the segmentation result will be displayed as an overlay to the input volume.

Run with provided example data: ./levelSetSeg example_data/mr_brain.mhd result.mhd 100 100 100 10 2000 125 40 0.05 125 255