151706061 / ivantk Goto Github PK

The Image-based Vascular Analysis Toolkit (IvanTk)
==================================================

1.Introduction and Scope
  ----------------------

The Image-based Vascular Analysis Library (IvanTk) is a set of cross-platform C++
software libraries for 2D/3D vascular analysis in medical imaging developed using
concepts of generic programming. It implements many of the vascular detection, ex-
traction and modelling methods described in the PhD thesis 

"Medical Image Analysis for the Detection Extraction and Modeling of Vascular Images"
by Ivan Macia Oliver, supervised by Prof. Manuel Graña Romay
Computational Intelligence Group, University of the Basque Country (2012)
http://www.ehu.es/ccwintco/uploads/e/ef/Thesis_Ivan_Macia_Final.pdf

The great number of existing approaches to vascular analysis, its increasing com-
plexity and the absence of a unified framework makes difficult to compare the per-
formance of the different methods in a meaningful and consistent way. Moreover,
the implementation of such methods in an object-oriented design allows reusability
of its constituent building-blocks, provides a better understanding of its algorithmic
mechanisms, reveals important implementation details and improves reproducibility of
experiments.

The main motivation behind the development of IvanTk was the lack of dedicated
libraries for image-based analysis of vascular structures. Some of the existing open 
source medical image analysis libraries, such as the Insight Toolkit (ITK), implement 
a few popular vascular detection methods. IvanTk may be considered an extension of ITK 
dedicated specifically to vascular analysis, since it reuses and extends many of the 
concepts and classes of this library. The Vascular Modeling Toolkit (VMTK) 
(http://www.vmtk.org) is an open-source library dedicated to vascular analysis but it 
is more oriented to computational geometry of vessels, generating detailed patient-specific 
models that can be used for structural or computational fluid dynamics studies. 
It provides a set of vascular segmentation methods, mainly based on level-set approaches,
but it lacks some algorithms for image-based analysis of blood vessels, specially for 
small vessels. At the same time of the development of this software, a similar toolkit 
called TubeTk (http://www.tubetk.org) was developed, for the segmentation, registration
and analysis of tubes and surfaces of images. The toolkit does not provide any 
visualization or GUI functionalities, although it is compatible with libraries such as
the Visualization Toolkit (http://www.vtk.org) (see example on the toolkit) and with
different C++ GUI frameworks.

This software pretends to be complementary and if possible compatible to the aforementioned 
softwares. Currently, its main focus is on vascular detection, extraction and modelling, 
trying to establish a framework in which different algorithm implementations can be 
compared consistently. Moreover, it incorporates object oriented paradigms modeling 
extraction schemes that allow to replace components in the extraction chain. Currently 
this is being developed specifically for vessel tracking algorithms. Finally, it incorporates 
a set of tools to develop synthetic tubular volumetric models that may be useful in 
different experiments. Most of the current implementations of vascular detectors are local 
point-based approaches, many of them kernel-based, that may not be effective computationally 
for exploring large volumes. Instead they are designed for local extraction schemes (such 
as vessel tracking) and for testing a sparse number of candidate points,  with the purpose 
of comparing accuracy or robustness of different approaches. In the future it is expected 
to incorporate both global approaches and faster implementations, such as GPU-based kernel 
implementations.

Regarding the scope of the library, it currently provides the following features:
• A repository of vascular detection and extraction algorithms under a unified framework: 
currently most of the methods are focused on centerline-based detection and extraction 
approaches, with focus on medialness functions and vascular tracking algorithms.
• An implementation of the Vessel Knowledge Representation model: it allows the 
representation of vascular structures in a way that can be easily used to store the results 
of the vascular extraction algorithms so this information can be readily used by the clinical 
applications (see Chapter 3 of the thesis).
• An implementation of the Generalized Vascular Tracking (GVT) Framework: it models vascular 
tracking procedures in a set of different individual stages and components in which existing 
methods can be accommodated in a flexible manner. Its object-oriented architecture allows 
testing the influence of the individual components. Moreover, complex extraction schemes 
can be generated by creating new components or extending existing ones, and by combining 
different parts of existing methods in innovative ways.
• A means of generating synthetic models of ideal shapes as volumetric images, such as the 
cylinder, toroid or helix models described in Appendix A of the thesis, with different 
cross-sections, shape and image parameters and noise levels.
• A means of reproducing the experiments of the thesis, since they are incorporated as unit 
tests and examples with configurable parameters.

2. Requirements
   ------------
   
The toolkit has been tested mainly on Windows with Visual Studio 2010 both in 32 and 64 bits. 
However it uses a CMake-based (http://www.cmake.org) build system and the code is multiplatform. 
It should work on most Linux and Mac platforms.
   
Requirements:

- CMake 2.8 or higher (http://www.cmake.org)
- A modern compatible compiler (the toolkit makes heavy use of C++ templates)
- Insight Toolkit (4.0 or higher recommended) (http://www.itk.org)
- Optional: Visualization Toolkit (5.0 or higher) (http://www.vtk.org)
   

3. Links
   -----
   
- Project page: https://sourceforge.net/projects/ivantk/
- Source code: https://github.com/imacia/ivan
- Git repository: https://github.com/imacia/ivan.git
- Mailing list URL: https://lists.sourceforge.net/lists/listinfo/ivantk-users
- Mailing list: [email protected]
   

4. Installation
   ------------
   
1) Download as zip file or clone repository from https://github.com/imacia/ivan.git
2) The source directory for CMake should be the root directory of the toolkit.
3) If not found automatically set the ITK_DIR CMake variable to point to your ITK binary dir.
4) Optionally set the VTK_DIR to point to your VTK binary directory.



5. Documentation
   -------------
   
Reference documentation for the toolkit can be generated using Doxygen (http://www.doxygen.org).
In this case, it is also recommended to install GraphViz (http://www.graphviz.org) in order to
generate the class diagrams. 

The documentation can be easily generated via CMake:
1) Set DOXYGEN_EXECUTABLE to doxygen.exe
2) Set DOXYGEN_DOT_EXECUTABLE to dot.exe from Graphviz
3) Click Configure and then Generate
4) A new project COMPILE_DOXYGEN is created. Build that project.
5) The html documentation should be generated in IVAN_BINARY_DIR/Doxygen/html/index.html where
   IVAN_BINARY_DIR is the binary dir of your choice in CMake.
   
   
6. Known Issues and Future Work
   ----------------------------
   
I am not totally happy with the Modeling part. Currently this part has been completely refactored
to store vessel information as a graph using Boost Graph Library (http://www.boost.org/libs/graph)
which I use in other propietary applications. However this migration process is not 
straightforward.

Gaussian kernel implementations are currently slow. Especially Hessian implementation can be 
further optimized. A GPU implementation could really speed up things.

Currently there are no detectors for bifurcation on vessel tracking. There are interesting
features but still working on increasing utility and robustness.