This document describes an implementation of Connected Threshold Segmentation to isolate the ventricles using Insight Toolkit www.itk.org.

Connected Threshold Segmentation of Ventricles in the Brain

This document describes the process used in implementing image segmentation through a region growing algorithm with the use of the Insight Toolkit (ITK) imaging program. The code used to implement the region growing algorithm is documented in the ITK Software Guide and makes use of the ITK library. This paper is accompanied with all source code, input data, parameters and output data that were originally used in developing the image segmentation application, except where prior laws or agreements prohibit redistribution of such material. The only data subject to not being redistributable was the input file used for testing which was a brain dataset from Kitware’s MIDAS collection named Normal090-T2.mha. The rest of this scientific publication is written in a manner such that it facilitates reproducibility of the reported results.

Image Segmentation: Ventricle Isolation Using Region Growing

ITK includes a number of iterators including the {\em itkImageLinearConstIteratorWithIndex} and {\em
itkImageLinearIteratorWithIndex} that support visiting voxels along
lines parallel to the image axes. This is particularly useful for
separable filters. This article introduces modified versions that do
not maintain location index so that filters that don't require index
information can perform faster.

Preliminary tests indicate that the iteration speed is improved, but
that filters such as the RecursiveSmoothingGaussianImageFilter do not
change significantly.

ImageLinearIterator - WithIndex removed

This document describes some experiments for segmentation using the Insight Toolkit
ITK \url{www.itk.org}. The code of the experiment is written following the
ITK CodingStyle as described in the directory \code{Insight/Documentation/Style.pdf}.

Segmentation is still an open problem and there are quite a lot of algorithms 
specific to application domains and data being used. In this experiment I've used 
a 3-D MRA scan of Brain retrieved from MIDAS website. This paper is accompanied 
with the source code, input data, parameters and output data that the author used for experimenting with the algorithm.

Experimenting with Segmentation in ITK

This paper describes the use of the Insight Toolkit (www.itk.org) to segment the ventricles in MR image data obtained from the Designed Database of MR Brain Images of Healthy Volunteers hosted at the MIDAS project.

Ventricle Segmentation with Itk's Watershed filter

Parameterization of surfaces, sometimes alled ”‘flattening”’ as it maps a surface embedded in 3D into its intrinsic 2D domain, is a powerfull tool for the analysis of surfaces. For the past years, there has been a growing interest in the Community that even lead to one implementation of one special type of parameterization in ITK [1,5,8,9]. We are providing here a more general framework for parameterization of single connected surfaces of any genus. It is based on a recent addition to ITK: itkQuadEdgeMesh [6] which allows an elegant an optimal implementation of algorithm for geometry and topology processing of discrete 2-manifolds. The 5 algorithms that we implemented map the meshes into a planar domain with fixed boundary leading to more stability and speed than mapping into the spherical domain. Each of them use different kind of parameterization with different properties. The conformal parameterization is usually used as it is intrinsic to the geometry of the mesh and thus allow shape analysis independently of the connectivity. However if flattening the mesh is your only goal, then the simplest parameterization algorithm (graph theory) will give you the best speed. This is the case when you want to further process the mesh in a lower dimension. Using specific solvers, the parameterization of meshes can be done sufficiently fast (couple of seconds) to consider this approach in interactive applications.

Parameterization of discrete surfaces

Morphological area attribute openings and closings\cite{Cheng92,Vincent93a} are connected set
operations that are able to remove parts of an image that meet certain
criteria without changing the rest of the image. This article describes a number of
classes that implement morphological area attribute operations using
the methods described by Meijster and Wilkinson \cite{meijster02}.

Grayscale morphological attribute operations

ITK's flood filling iterator represents a way to visit pixels/voxels within an image having a specific connectivity. The iterator is initialized at known seed points and starting from these, neighbors that are within the desired connectivity are marked as to be visited and processed by the iterator in the future.

The FloodFilledFunctionConditionalConstIterator builds the foundation for the implementation of the NeighborhoodConnectedImageFilter and is currently fixed to investigating 4-neighborhoods in 2D and 6-neighborhoods in 3D. Since many existing applications use region growing algorithms that perform on full neighborhoods (i.e. 8-connected in 2D and 26-connected in 3D) it is desirable to be able to choose at least between these two standard connectivities.

In this document we describe the extension of the existing iterator to be using ShapedNeighborhoodIterators which has already been proposed in the implementation of the FloodFilledFunctionConditionalConstIterator.

Shaped Neighborhood based Flood Filled Conditional Iterators

This document describes the implementation of several features previously developed[2], extending the
2D scale invariant feature transform (SIFT)[4, 5] for images of arbitrary dimensionality, such as 3D
medical image volumes and time series, using ITK1. Specifically, we provide a scale invariant implementation
of a weighted histogram of gradient feature, a rotationally invariant version of the histogram
feature, and a SIFT-like feature, adapted to handle images of arbitrary dimensionality. This paper is accompanied
with the source code, example input data, parameters and output data, allowing reproduction
of the example results in this paper and the results previously reported[2]. Note that usage of SIFT in the
United States is governed by US Patent 6,711,293.

Scale Invariant Feature Transform for n-Dimensional Images (n-SIFT)

Convolution is a basic operation in image analysis and the Insight Toolkit provides a well-framed mechanism for such operations (see Chapter 11 of the ITK Software Guide. The classes associated with this contribution are meant to simplify convolution operations for the user who wishes to convolve a given image with an arbitrary image kernel. This entails the use of the neighborhood iterator coupled with a image kernel neighborhood operator which we provide. A couple of examples are provided to illustrate its use.