RealImage::Pointer bsplineRegistration(RealImage::Pointer srcImg, RealImage::Pointer dstImg) {
const unsigned int SpaceDimension = ImageDimension;
const unsigned int SplineOrder = 3;
typedef double CoordinateRepType;
typedef itk::BSplineTransform<CoordinateRepType, SpaceDimension, SplineOrder> TransformType;
typedef itk::LBFGSOptimizer OptimizerType;
typedef itk::MeanSquaresImageToImageMetric<ImageType, ImageType> MetricType;
typedef itk::LinearInterpolateImageFunction<ImageType, double> InterpolatorType;
typedef itk::ImageRegistrationMethod<ImageType, ImageType> RegistrationType;
MetricType::Pointer metric = MetricType::New();
OptimizerType::Pointer optimizer = OptimizerType::New();
InterpolatorType::Pointer interpolator = InterpolatorType::New();
RegistrationType::Pointer registration = RegistrationType::New();
// The old registration framework has problems with multi-threading
// For now, we set the number of threads to 1
// registration->SetNumberOfThreads(1);
registration->SetMetric( metric );
registration->SetOptimizer( optimizer );
registration->SetInterpolator( interpolator );
TransformType::Pointer transform = TransformType::New();
registration->SetTransform( transform );
// Setup the registration
registration->SetFixedImage( dstImg );
registration->SetMovingImage( srcImg);
ImageType::RegionType fixedRegion = srcImg->GetBufferedRegion();
registration->SetFixedImageRegion( fixedRegion );
// Here we define the parameters of the BSplineDeformableTransform grid. We
// arbitrarily decide to use a grid with $5 \times 5$ nodes within the image.
// The reader should note that the BSpline computation requires a
// finite support region ( 1 grid node at the lower borders and 2
// grid nodes at upper borders). Therefore in this example, we set
// the grid size to be $8 \times 8$ and place the grid origin such that
// grid node (1,1) coincides with the first pixel in the fixed image.
TransformType::PhysicalDimensionsType fixedPhysicalDimensions;
TransformType::MeshSizeType meshSize;
for (unsigned int i=0; i < ImageDimension; i++) {
fixedPhysicalDimensions[i] = dstImg->GetSpacing()[i] *
static_cast<double>(dstImg->GetLargestPossibleRegion().GetSize()[i] - 1 );
meshSize[i] = dstImg->GetLargestPossibleRegion().GetSize()[i] / 8 - SplineOrder;
}
// unsigned int numberOfGridNodesInOneDimension = 15;
// meshSize.Fill( numberOfGridNodesInOneDimension - SplineOrder );
transform->SetTransformDomainOrigin( dstImg->GetOrigin() );
transform->SetTransformDomainPhysicalDimensions( fixedPhysicalDimensions );
transform->SetTransformDomainMeshSize( meshSize );
transform->SetTransformDomainDirection( dstImg->GetDirection() );
typedef TransformType::ParametersType ParametersType;
const unsigned int numberOfParameters = transform->GetNumberOfParameters();
ParametersType parameters( numberOfParameters );
parameters.Fill( 0.0 );
transform->SetParameters( parameters );
// We now pass the parameters of the current transform as the initial
// parameters to be used when the registration process starts.
registration->SetInitialTransformParameters( transform->GetParameters() );
std::cout << "Intial Parameters = " << std::endl;
std::cout << transform->GetParameters() << std::endl;
// Next we set the parameters of the LBFGS Optimizer.
optimizer->SetGradientConvergenceTolerance(0.1);
optimizer->SetLineSearchAccuracy(0.09);
optimizer->SetDefaultStepLength(.1);
optimizer->TraceOn();
optimizer->SetMaximumNumberOfFunctionEvaluations(1000);
std::cout << std::endl << "Starting Registration" << std::endl;
try {
registration->Update();
std::cout << "Optimizer stop condition = "
<< registration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
} catch (itk::ExceptionObject & err) {
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return RealImage::Pointer();
}
OptimizerType::ParametersType finalParameters =
registration->GetLastTransformParameters();
std::cout << "Last Transform Parameters" << std::endl;
std::cout << finalParameters << std::endl;
transform->SetParameters( finalParameters );
typedef itk::ResampleImageFilter<ImageType, ImageType> ResampleFilterType;
ResampleFilterType::Pointer resample = ResampleFilterType::New();
resample->SetTransform( transform );
resample->SetInput( srcImg );
resample->SetSize( dstImg->GetLargestPossibleRegion().GetSize() );
resample->SetOutputOrigin( dstImg->GetOrigin() );
resample->SetOutputSpacing( dstImg->GetSpacing() );
resample->SetOutputDirection( dstImg->GetDirection() );
resample->SetDefaultPixelValue( 100 );
resample->Update();
return resample->GetOutput();
}
// perform B-spline registration for 2D image
void runBspline2D(StringVector& args) {
typedef itk::BSplineTransform<double, 2, 3> TransformType;
typedef itk::LBFGSOptimizer OptimizerType;
typedef itk::MeanSquaresImageToImageMetric<RealImage2, RealImage2> MetricType;
typedef itk:: LinearInterpolateImageFunction<RealImage2, double> InterpolatorType;
typedef itk::ImageRegistrationMethod<RealImage2, RealImage2> RegistrationType;
MetricType::Pointer metric = MetricType::New();
OptimizerType::Pointer optimizer = OptimizerType::New();
InterpolatorType::Pointer interpolator = InterpolatorType::New();
RegistrationType::Pointer registration = RegistrationType::New();
// The old registration framework has problems with multi-threading
// For now, we set the number of threads to 1
registration->SetNumberOfThreads(1);
registration->SetMetric( metric );
registration->SetOptimizer( optimizer );
registration->SetInterpolator( interpolator );
TransformType::Pointer transform = TransformType::New();
registration->SetTransform( transform );
ImageIO<RealImage2> io;
// Create the synthetic images
RealImage2::Pointer fixedImage = io.ReadImage(args[0]);
RealImage2::Pointer movingImage = io.ReadImage(args[1]);
// Setup the registration
registration->SetFixedImage( fixedImage );
registration->SetMovingImage( movingImage);
RealImage2::RegionType fixedRegion = fixedImage->GetBufferedRegion();
registration->SetFixedImageRegion( fixedRegion );
TransformType::PhysicalDimensionsType fixedPhysicalDimensions;
TransformType::MeshSizeType meshSize;
for( unsigned int i=0; i < 2; i++ )
{
fixedPhysicalDimensions[i] = fixedImage->GetSpacing()[i] *
static_cast<double>(
fixedImage->GetLargestPossibleRegion().GetSize()[i] - 1 );
}
unsigned int numberOfGridNodesInOneDimension = 18;
meshSize.Fill( numberOfGridNodesInOneDimension - 3 );
transform->SetTransformDomainOrigin( fixedImage->GetOrigin() );
transform->SetTransformDomainPhysicalDimensions( fixedPhysicalDimensions );
transform->SetTransformDomainMeshSize( meshSize );
transform->SetTransformDomainDirection( fixedImage->GetDirection() );
typedef TransformType::ParametersType ParametersType;
const unsigned int numberOfParameters =
transform->GetNumberOfParameters();
ParametersType parameters( numberOfParameters );
parameters.Fill( 0.0 );
transform->SetParameters( parameters );
// We now pass the parameters of the current transform as the initial
// parameters to be used when the registration process starts.
registration->SetInitialTransformParameters( transform->GetParameters() );
std::cout << "Intial Parameters = " << std::endl;
std::cout << transform->GetParameters() << std::endl;
// Next we set the parameters of the LBFGS Optimizer.
optimizer->SetGradientConvergenceTolerance( 0.005 );
optimizer->SetLineSearchAccuracy( 0.9 );
optimizer->SetDefaultStepLength( .1 );
optimizer->TraceOn();
optimizer->SetMaximumNumberOfFunctionEvaluations( 1000 );
std::cout << std::endl << "Starting Registration" << std::endl;
try
{
registration->Update();
std::cout << "Optimizer stop condition = "
<< registration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return;
}
OptimizerType::ParametersType finalParameters =
registration->GetLastTransformParameters();
std::cout << "Last Transform Parameters" << std::endl;
std::cout << finalParameters << std::endl;
transform->SetParameters( finalParameters );
typedef itk::ResampleImageFilter<RealImage2, RealImage2> ResampleFilterType;
ResampleFilterType::Pointer resample = ResampleFilterType::New();
resample->SetTransform( transform );
resample->SetInput( movingImage );
resample->SetSize( fixedImage->GetLargestPossibleRegion().GetSize() );
resample->SetOutputOrigin( fixedImage->GetOrigin() );
resample->SetOutputSpacing( fixedImage->GetSpacing() );
resample->SetOutputDirection( fixedImage->GetDirection() );
resample->SetDefaultPixelValue( 100 );
resample->Update();
io.WriteImage(args[2], resample->GetOutput());
}
typename TImage::Pointer modelBasedImageToImageRegistration(std::string referenceFilename,
std::string targetFilename,
typename TStatisticalModelType::Pointer model,
std::string outputDfFilename,
unsigned numberOfIterations){
typedef itk::ImageFileReader<TImage> ImageReaderType;
typedef itk::InterpolatingStatisticalDeformationModelTransform<TRepresenter, double, VImageDimension> TransformType;
typedef itk::LBFGSOptimizer OptimizerType;
typedef itk::ImageRegistrationMethod<TImage, TImage> RegistrationFilterType;
typedef itk::WarpImageFilter< TImage, TImage, TVectorImage > WarperType;
typedef itk::LinearInterpolateImageFunction< TImage, double > InterpolatorType;
typename ImageReaderType::Pointer referenceReader = ImageReaderType::New();
referenceReader->SetFileName(referenceFilename.c_str());
referenceReader->Update();
typename TImage::Pointer referenceImage = referenceReader->GetOutput();
referenceImage->Update();
typename ImageReaderType::Pointer targetReader = ImageReaderType::New();
targetReader->SetFileName(targetFilename.c_str());
targetReader->Update();
typename TImage::Pointer targetImage = targetReader->GetOutput();
targetImage->Update();
// do the fitting
typename TransformType::Pointer transform = TransformType::New();
transform->SetStatisticalModel(model);
transform->SetIdentity();
// Setting up the fitting
OptimizerType::Pointer optimizer = OptimizerType::New();
optimizer->MinimizeOn();
optimizer->SetMaximumNumberOfFunctionEvaluations(numberOfIterations);
typedef IterationStatusObserver ObserverType;
ObserverType::Pointer observer = ObserverType::New();
optimizer->AddObserver( itk::IterationEvent(), observer );
typename TMetricType::Pointer metric = TMetricType::New();
typename InterpolatorType::Pointer interpolator = InterpolatorType::New();
typename RegistrationFilterType::Pointer registration = RegistrationFilterType::New();
registration->SetInitialTransformParameters(transform->GetParameters());
registration->SetMetric(metric);
registration->SetOptimizer( optimizer );
registration->SetTransform( transform );
registration->SetInterpolator( interpolator );
registration->SetFixedImage( targetImage );
registration->SetFixedImageRegion(targetImage->GetBufferedRegion() );
registration->SetMovingImage( referenceImage );
try {
std::cout << "Performing registration... " << std::flush;
registration->Update();
std::cout << "[done]" << std::endl;
} catch ( itk::ExceptionObject& o ) {
std::cout << "caught exception " << o << std::endl;
}
typename TVectorImage::Pointer df = model->DrawSample(transform->GetCoefficients());
// write deformation field
if(outputDfFilename.size()>0){
typename itk::ImageFileWriter<TVectorImage>::Pointer df_writer = itk::ImageFileWriter<TVectorImage>::New();
df_writer->SetFileName(outputDfFilename);
df_writer->SetInput(df);
df_writer->Update();
}
// warp reference
std::cout << "Warping reference... " << std::flush;
typename WarperType::Pointer warper = WarperType::New();
warper->SetInput(referenceImage );
warper->SetInterpolator( interpolator );
warper->SetOutputSpacing( targetImage->GetSpacing() );
warper->SetOutputOrigin( targetImage->GetOrigin() );
warper->SetOutputDirection( targetImage->GetDirection() );
warper->SetDisplacementField( df );
warper->Update();
std::cout << "[done]" << std::endl;
return warper->GetOutput();
}