[ITK-users] Applying a B-Spline transform to a 3D mesh
Carlos Henrique Villa Pinto
chvillap at gmail.com
Wed Sep 23 09:38:27 EDT 2015
Hi everyone,
I have a 3D MR brain image and a set of 3D meshes of different brain
structures aligned to it. I'm trying to write a very simple program in
which I create a B-Spline transform (with random parameters) and apply such
transform to the image and to the meshes in order to deform them all
together.
Am I wrong to assume that, if I apply exactly the same transform to an
image and to a mesh which is perfectly aligned to a specific brain
structure in such image, the resulting deformed mesh should be aligned to
the same brain structure in the deformed image? Because that's not what
it's happening in practice. The transform deforms the meshes to some
extent, but the deformed meshes do not match the brain structures in the
deformed image, even though the applied transform is the same for all. So I
guess I must be doing something wrong...
The source code is below. I'm using ITK 4.8, and the MR image (NIfTI) and
meshes (VTK) used in my tests come from the NAC Brain Atlas (
https://www.slicer.org/publications/item/view/2037).
Any insight about what is possibly causing these results would be
appreciated. Thanks in advance.
#include <iostream>
#include <vector>
#include <cstdlib>
#include <cstring>
#include <cmath>
#include <ctime>
#include <itkImage.h>
#include <itkImageFileReader.h>
#include <itkImageFileWriter.h>
#include <itkMesh.h>
#include <itkMeshFileReader.h>
#include <itkMeshFileWriter.h>
#include <itkBSplineTransform.h>
#include <itkResampleImageFilter.h>
#include <itkTransformMeshFilter.h>
#include <itkLinearInterpolateImageFunction.h>
int main(int argc, char const *argv[])
{
if (argc < 3)
{
std::cerr << "Usage: " << argv[0]
<< " gridNodesInOneDimension inputImage [inputMesh1, ...]"
<< std::endl;
return EXIT_FAILURE;
}
srand(time(NULL));
const int splineOrder = 3;
const int dimensions = 3;
typedef itk::Image<double, dimensions> ImageType;
typedef itk::ImageFileReader<ImageType> ImageReaderType;
typedef itk::ImageFileWriter<ImageType> ImageWriterType;
typedef itk::Mesh<double, dimensions> MeshType;
typedef itk::MeshFileReader<MeshType> MeshReaderType;
typedef itk::MeshFileWriter<MeshType> MeshWriterType;
typedef itk::BSplineTransform<double, dimensions, splineOrder>
BSplineTransformType;
typedef itk::ResampleImageFilter<ImageType, ImageType, double>
ResampleImageFilterType;
typedef itk::TransformMeshFilter<MeshType, MeshType,
BSplineTransformType>
TransformMeshFilterType;
typedef itk::LinearInterpolateImageFunction<ImageType, double>
LinearInterpolatorType;
//
============================================================================================
// Reading the input image
//
============================================================================================
ImageReaderType::Pointer imageReader = ImageReaderType::New();
imageReader->SetFileName(argv[2]);
imageReader->Update();
ImageType::Pointer inputImage = imageReader->GetOutput();
//
============================================================================================
// Reading the input meshes
//
============================================================================================
std::vector<MeshType::Pointer> inputMeshVector;
for (int i = 3; i < argc; ++i)
{
MeshReaderType::Pointer meshReader = MeshReaderType::New();
meshReader->SetFileName(argv[i]);
meshReader->Update();
MeshType::Pointer inputMesh = meshReader->GetOutput();
inputMeshVector.push_back(inputMesh);
}
//
============================================================================================
// Generating the parameters of the B-Spline transform
//
============================================================================================
const int gridNodesInOneDimension = atoi(argv[1]);
BSplineTransformType::Pointer bsplineTransform =
BSplineTransformType::New();
BSplineTransformType::PhysicalDimensionsType physicalDimensions;
BSplineTransformType::MeshSizeType meshSize;
for (int i = 0; i < dimensions; ++i)
{
physicalDimensions[i] = inputImage->GetSpacing()[i] *
static_cast<double>(inputImage->GetLargestPossibleRegion().GetSize()[i] -
1);
}
meshSize.Fill(gridNodesInOneDimension - splineOrder);
bsplineTransform->SetTransformDomainOrigin(inputImage->GetOrigin());
bsplineTransform->SetTransformDomainDirection(inputImage->GetDirection());
bsplineTransform->SetTransformDomainPhysicalDimensions(physicalDimensions);
bsplineTransform->SetTransformDomainMeshSize(meshSize);
BSplineTransformType::ParametersType
parameters(bsplineTransform->GetNumberOfParameters());
double minValue = -20.0;
double maxValue = 20.0;
for (int i = 0; i < bsplineTransform->GetNumberOfParameters(); ++i)
{
parameters[i] = (double) rand() / RAND_MAX * (maxValue - minValue)
+ minValue;
std::cout << parameters[i] << " ";
}
std::cout << std::endl;
bsplineTransform->SetParameters(parameters);
//
============================================================================================
// Applying the B-Spline transform to the input image
//
============================================================================================
LinearInterpolatorType::Pointer linearInterpolator =
LinearInterpolatorType::New();
ResampleImageFilterType::Pointer resampleFilter =
ResampleImageFilterType::New();
resampleFilter->SetTransform(bsplineTransform);
resampleFilter->SetInterpolator(linearInterpolator);
resampleFilter->SetInput(inputImage);
resampleFilter->SetSize(inputImage->GetLargestPossibleRegion().GetSize());
resampleFilter->SetOutputStartIndex(inputImage->GetLargestPossibleRegion().GetIndex());
resampleFilter->SetOutputOrigin(inputImage->GetOrigin());
resampleFilter->SetOutputSpacing(inputImage->GetSpacing());
resampleFilter->SetOutputDirection(inputImage->GetDirection());
resampleFilter->Update();
//
============================================================================================
// Applying the B-Spline transform to the meshes
//
============================================================================================
std::vector<MeshType::Pointer> outputMeshVector;
for (int i = 0; i < inputMeshVector.size(); ++i)
{
TransformMeshFilterType::Pointer transformFilter =
TransformMeshFilterType::New();
transformFilter->SetInput(inputMeshVector[i]);
transformFilter->SetTransform(bsplineTransform);
transformFilter->Update();
outputMeshVector.push_back(transformFilter->GetOutput());
}
//
============================================================================================
// Writing the transformed image
//
============================================================================================
char outputImageFileName[256];
int length1 = strlen(argv[2]) - 4;
strncpy(outputImageFileName, argv[2], length1);
outputImageFileName[length1] = '\0';
strcat(outputImageFileName, "_transformed.nii");
ImageWriterType::Pointer imageWriter = ImageWriterType::New();
imageWriter->SetFileName(outputImageFileName);
imageWriter->SetInput(resampleFilter->GetOutput());
imageWriter->Update();
//
============================================================================================
// Writing the transformed meshes
//
============================================================================================
for (int i = 0; i < outputMeshVector.size(); ++i)
{
char outputMeshFilename[256];
int length2 = strlen(argv[i + 3]) - 4;
strncpy(outputMeshFilename, argv[i + 3], length2);
outputMeshFilename[length2] = '\0';
strcat(outputMeshFilename, "_transformed.vtk");
MeshType::Pointer outputMesh = outputMeshVector[i];
MeshWriterType::Pointer meshWriter = MeshWriterType::New();
meshWriter->SetFileName(outputMeshFilename);
meshWriter->SetInput(outputMesh);
meshWriter->Update();
}
return EXIT_SUCCESS;
}
[]s
--
Carlos Henrique Villa Pinto
Graduate Student in Computer Science
Federal University of São Carlos - Brazil
XCS
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-users/attachments/20150923/cdcb448b/attachment.html>
More information about the Insight-users
mailing list