Thin Vessels Segmentation

Modules

• Core

  Including general data stuctures for 3D volume, data I/O, data visualization, simple 3D image processing and etc.

• Vesselness

  This is a C++ implementation of Vesselness Measure for 3D volume based on the following paper Frangi.

  • Frangi, Alejandro F., et al. "Multiscale vessel enhancement filtering." Medical Image Computing and Computer-Assisted Interventation—MICCAI’98. Springer Berlin Heidelberg, 1998. 130-137.

  Two sample code is provided in main.cpp:

  • Computing vesselness measure
  • Extracting the vessel centrelines with non-maximum suppression

  Sample Usage:

  • Setting parameters:

    [sigma_from, sigma_to]: the potential size rang of the vessels

    sigma_step: precision of computation

    For other parameters alpha, beta, gamma, please refer to Frangi's papaer or this blog or Frangi's paper.

    float sigma_from = 1.0f;  
    float sigma_to = 8.10f;   
    float sigma_step = 0.5f;  
    float alpha = 1.0e-1f;	
    float beta  = 5.0e0f;     
    float gamma = 3.5e5f; 
    

  • Laoding data

    Data3D<short> im_short;       
    bool flag = im_short.load( "dataname.data" );         
    if(!flag) return 0;       
    

  • Compute Vesselness

    Data3D<Vesselness_Sig> vn_sig;        
    VesselDetector::compute_vesselness( im_short, vn_sig,         
            sigma_from, sigma_to, sigma_step,        
            alpha, beta, gamma );
    

  • Saving Data

    vn_sig.save( "dataname.vn_sig" );
    

  • If you want to visulize the data using maximum-intensity projection

    viewer.addObject( vn_sig,  GLViewer::Volumn::MIP );
    viewer.addDiretionObject( vn_sig );
    viewer.go(600, 400, 2);
    

• ModelFitting (Levenberg Marquardt algorithm)

  Fitting geometric models (lines) to the 3D vessel data using Levenberg-Marquardt algorithm.

  Using Levenberg Marquardt algorithm for energy minimization.

  Energy contains two parts:

  • Data cost Distance to the center of a line model
  • Pair-wise smooth cost Complicated. Please refer to this paper for more details

  Levenberg Marquardt algorithm requires to compute the Jacobin matrix for both the data cost and the smooth cost. This computation is very time-consuming and the computation has been highly paralleled in this version.

• SparseMatrix

  Sparse matrix representation, contains all necessary matrix manipulations such as:

  • addtion
  • subtraction
  • multiplication
  • transpose

• SparseMatrixCV

  OpenCV warper for SparseMatrix. For matrix multiplication between sparse matrix and dense matrix. For example,

  template <class _Tp, int m, int n>    
  SparseMatrixCV operator*( const cv::Matx<_Tp,m,n>& vec, const SparseMatrixCV& sm );     
  

  Mat_<double> operator*( const SparseMatrixCV& sm, const Mat_<double>& sm );    
  

• Vesselness-cuda

  CUDA version of the vesselness measure (under development)

• RingsReduction

  Rings reduction of CT images (under development)

• EigenDecomp

  Eigenvalue decomposition of a 3 by 3 symmetric matrix

Requirements

• Linux

  1. X Window System (X11, X, and sometimes informally X-Windows), which is a windowing system for bitmap displays

     sudo apt-get install libxmu-dev
     sudo apt-get install libxi-dev
     

  2. OpenGL

     sudo apt-get install mesa-common-dev

  3. glew

     sudo apt-get install libglew-dev

  4. freeglut

     sudo apt-get install freeglut3-dev

  5. OpenCV 2.4.9

     Download OpenCV. Generate makefile with cmake, build and install the liabrary.

     cmake ./
     make
     make install
     

     One more thing. Make sure you have added the OpenCV lib direcory to PATH by the following.

     export LD_LIBRARY_PATH=/usr/local/lib/

     /usr/local/lib is where the OpenCV libs are located. You could double check by

     pkg-config --libs opencv
     

  6. Code::Blocks

• Windows

  All the code was written in C++. So it shouldn't be hard to modify and adapt to Windows. As a matter of fact, earlier version of the development was under Windows.

---

Some of the results were eventually published in this paper Thin Structure Estimation with Curvature Regularization. Please check it out :)