void LoadFiles(int argc, char* argv[]) {
itkcmds::itkImageIO<ImageType> io;
itkcmds::itkImageIO<LabelType> io2;
_src = io.ReadImageT(argv[1]);
_dst = io.ReadImageT(argv[2]);
_dstLabel = io2.ReadImageT(argv[3]);
_transformOut = argv[4];
_resampledOut = argv[5];
cout << "Transform Output: " << _transformOut << endl;
cout << "Resampled Output: " << _resampledOut << endl;
ImageType::SizeType szDst = _dst->GetBufferedRegion().GetSize();
itk::ContinuousIndex<double,3> szIdx;
for (int i = 0; i < 3; i++) {
szIdx[i] = szDst[i] / 2.0;
}
_dst->TransformContinuousIndexToPhysicalPoint(szIdx, _dstCenter);
itk::ImageRegionConstIteratorWithIndex<LabelType> labelIter(_dstLabel, _dstLabel->GetBufferedRegion());
for (labelIter.GoToBegin(); !labelIter.IsAtEnd(); ++labelIter) {
LabelType::PixelType label = labelIter.Get();
if (label > 0) {
_labelIndexes.push_back(labelIter.GetIndex());
}
}
_centerOfRotation.SetSize(ImageType::ImageDimension);
for (int i = 0; i < 3; i++) {
_centerOfRotation[i] = _dstCenter[i];
}
}
FuzzyCMeans::FuzzyCMeans(QVector<ClusterPoint2D> &points, QVector<ClusterCentroid2D> &clusters, float fuzzy, imageType::Pointer myImage, int numCluster)
{
this->Eps = std::pow(10, -5);
this->isConverged=false;
this->Points = points;
this->Clusters = clusters;
this->myImageHeight = myImage->GetBufferedRegion().GetSize()[0];
this->myImageWidth = myImage->GetBufferedRegion().GetSize()[1];
this->myImage = myImage;
U= boost::numeric::ublas::matrix<double>(Points.size(),Clusters.size());
this->Fuzzyness = fuzzy;
double diff;
imageType::SizeType size;
size[0]=myImageWidth; // x axis
size[1]=myImageHeight; // y
imageType::RegionType region;
region.SetSize(size);
imageType::Pointer image = imageType::New();
image->SetRegions(region);
image->Allocate(); // immagine create
// Iterate through all points to create initial U matrix
for (int i = 0; i < Points.size()-1; i++)
{
ClusterPoint2D p = Points.at(i);
double sum = 0.0;
for (int j = 0; j < Clusters.size()-1; j++)
{
ClusterCentroid2D c = Clusters.at(j);
diff = std::sqrt(std::pow(CalculateEuclideanDistance(p, c), 2.0));
//U(i, j) = (diff == 0) ? Eps : diff;
if (diff==0){
U(i,j)=Eps;
}
else{
U(i,j)=diff;
}
sum += U(i, j);
}
}
this->RecalculateClusterMembershipValues();
}
static void createPhantomParticles(ImageType::Pointer edgeImg) {
PointVectorType phantoms;
itk::ImageRegionConstIteratorWithIndex<ImageType> iter(edgeImg, edgeImg->GetBufferedRegion());
for (iter.GoToBegin(); !iter.IsAtEnd(); ++iter) {
if (iter.Get() > 0) {
phantoms.push_back(iter.GetIndex()[0]);
phantoms.push_back(iter.GetIndex()[1]);
}
}
g_phantomParticles.push_back(phantoms);
}
void LoadFiles(int argc, char* argv[]) {
itkcmds::itkImageIO<ImageType> io;
itkcmds::itkImageIO<LabelType> io2;
_src = io.ReadImageT(argv[2]);
_dst = io.ReadImageT(argv[3]);
_dstLabel = io2.ReadImageT(argv[4]);
_transformOut = argv[5];
_resampledOut = argv[6];
cout << "Transform Output: " << _transformOut << endl;
cout << "Resampled Output: " << _resampledOut << endl;
ImageType::SizeType szDst = _dst->GetBufferedRegion().GetSize();
itk::ContinuousIndex<double,3> szIdx;
for (int i = 0; i < 3; i++) {
szIdx[i] = szDst[i] / 2.0;
}
_dst->TransformContinuousIndexToPhysicalPoint(szIdx, _dstCenter);
}
void loadMask(QString f) {
itkcmds::itkImageIO<ImageType> io;
ImageType::Pointer img = io.ReadImageT(f.toAscii().data());
g_boundaryMapList.push_back(img);
ScalarToRGBFilter::Pointer rgbFilter1 = ScalarToRGBFilter::New();
rgbFilter1->SetInput(img);
rgbFilter1->SetAlphaValue(128);
rgbFilter1->Update();
BitmapType::Pointer maskBitmap = rgbFilter1->GetOutput();
g_maskBitmapList.push_back(maskBitmap);
DistanceMapFilter::Pointer distmapFilter = DistanceMapFilter::New();
distmapFilter->SetInput(img);
distmapFilter->Update();
ImageType::Pointer distImg = distmapFilter->GetOutput();
g_distanceMapList.push_back(distmapFilter->GetOutput());
DistanceVectorImageType::Pointer distVector = distmapFilter->GetVectorDistanceMap();
g_distanceVectorList.push_back(distVector);
ScalarToRGBFilter::Pointer rgbFilter = ScalarToRGBFilter::New();
rgbFilter->SetInput(distImg);
rgbFilter->SetNumberOfThreads(8);
rgbFilter->Update();
BitmapType::Pointer rgbImage = rgbFilter->GetOutput();
g_distanceMapBitmapList.push_back(rgbImage);
EdgeDetectionFilterType::Pointer edgeFilter = EdgeDetectionFilterType::New();
edgeFilter->SetInput(img);
edgeFilter->Update();
ImageType::Pointer edgeImg = edgeFilter->GetOutput();
PointVectorType phantoms;
itk::ImageRegionConstIteratorWithIndex<ImageType> iter(edgeImg, edgeImg->GetBufferedRegion());
for (iter.GoToBegin(); !iter.IsAtEnd(); ++iter) {
if (iter.Get() > 0) {
phantoms.push_back(iter.GetIndex()[0]);
phantoms.push_back(iter.GetIndex()[1]);
}
}
g_phantomParticles.push_back(phantoms);}
virtual ReadResult readImage(const std::string& file, const osgDB::ReaderWriter::Options* options) const
{
std::string ext = osgDB::getLowerCaseFileExtension(file);
if (!acceptsExtension(ext)) return ReadResult::FILE_NOT_HANDLED;
std::string fileName = osgDB::findDataFile( file, options );
if (fileName.empty()) return ReadResult::FILE_NOT_FOUND;
notice()<<"Reading DICOM file "<<fileName<<std::endl;
typedef unsigned short PixelType;
const unsigned int Dimension = 3;
typedef itk::Image< PixelType, Dimension > ImageType;
typedef itk::ImageFileReader< ImageType > ReaderType;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( fileName.c_str() );
typedef itk::GDCMImageIO ImageIOType;
ImageIOType::Pointer gdcmImageIO = ImageIOType::New();
reader->SetImageIO( gdcmImageIO );
try
{
reader->Update();
}
catch (itk::ExceptionObject & e)
{
std::cerr << "exception in file reader " << std::endl;
std::cerr << e.GetDescription() << std::endl;
std::cerr << e.GetLocation() << std::endl;
return ReadResult::ERROR_IN_READING_FILE;
}
ImageType::Pointer inputImage = reader->GetOutput();
ImageType::RegionType region = inputImage->GetBufferedRegion();
ImageType::SizeType size = region.GetSize();
ImageType::IndexType start = region.GetIndex();
//inputImage->GetSpacing();
//inputImage->GetOrigin();
unsigned int width = size[0];
unsigned int height = size[1];
unsigned int depth = size[2];
osg::RefMatrix* matrix = new osg::RefMatrix;
notice()<<"width = "<<width<<" height = "<<height<<" depth = "<<depth<<std::endl;
for(unsigned int i=0; i<Dimension; ++i)
{
(*matrix)(i,i) = inputImage->GetSpacing()[i];
(*matrix)(3,i) = inputImage->GetOrigin()[i];
}
osg::Image* image = new osg::Image;
image->allocateImage(width, height, depth, GL_LUMINANCE, GL_UNSIGNED_BYTE, 1);
unsigned char* data = image->data();
typedef itk::ImageRegionConstIterator< ImageType > IteratorType;
IteratorType it(inputImage, region);
it.GoToBegin();
while (!it.IsAtEnd())
{
*data = it.Get();
++data;
++it;
}
image->setUserData(matrix);
matrix->preMult(osg::Matrix::scale(double(image->s()), double(image->t()), double(image->r())));
return image;
}
int main(int argc, char *argv[])
{
//check that the user specified the right number of command line arguments
if(argc < 3)
{
cerr << argv[0] << " <input file> <output file>" << endl;
cerr << argv[0] << " <raw input file> <sizeX> <sizeY> <sizeZ> <output file>"
<< endl;
return EXIT_FAILURE;
}
//check if the input image is .raw
bool rawInput = false;
string inputFileName = argv[1];
const char *outputFileName;
//double space[3]={1,1,1};
if(inputFileName.rfind(".raw") != string::npos)
{
//if so, the user is also required to pass in image dimensions
if(argc < 6)
{
cerr << "Usage: <raw input file> <sizeX> <sizeY> <sizeZ> <output file>" << endl;
return EXIT_FAILURE;
}
rawInput = true;
sizeX = atoi(argv[2]);
sizeY = atoi(argv[3]);
sizeZ = atoi(argv[4]);
outputFileName = argv[5];
}
else
{
outputFileName = argv[2];
}
FILE *infile = 0;
FILE *outfile;
int i,j,k;
int ii, jj, kk;
DATATYPEOUT *volout;
long idx;
int sizeExpand = 0;
DATATYPEOUT blockMax;
int timesDilate;
int border;
unsigned short *GraphcutResults;
cout << "Volume Processing..." << endl;
//make sure we can write to the output file
if((outfile=fopen(outputFileName, "wb")) == NULL)
{
cerr << "Output file open error!" << endl;
return EXIT_FAILURE;
}
typedef float PixelType;
const unsigned int Dimension = 3;
typedef itk::Image< PixelType, Dimension > ImageType;
ImageType::Pointer inputImage;
if(rawInput)
{
if((infile=fopen(inputFileName.c_str(), "rb"))==NULL)
{
cout << "Input file open error!" << endl;
return EXIT_FAILURE;
}
volin = (DATATYPEIN*)malloc(sizeX*sizeY*(sizeZ+sizeExpand*2)*sizeof(DATATYPEIN));
if (fread(volin, sizeof(DATATYPEIN), sizeX*sizeY*sizeZ, infile) < (unsigned int)(sizeX*sizeY*sizeZ))
{
cerr << "File size is not the same as volume size" << endl;
return EXIT_FAILURE;
}
inputImage = ImageType::New();
ImageType::PointType origin;
origin[0] = 0.;
origin[1] = 0.;
origin[2] = 0.;
inputImage->SetOrigin( origin );
ImageType::IndexType start;
start[0] = 0;
start[1] = 0;
start[2] = 0;
ImageType::SizeType size;
size[0] = sizeX;
size[1] = sizeY;
size[2] = sizeZ;
ImageType::RegionType region;
region.SetSize( size );
region.SetIndex( start );
inputImage->SetRegions( region );
inputImage->Allocate();
inputImage->FillBuffer(0);
inputImage->Update();
typedef itk::ImageRegionIteratorWithIndex< ImageType > IteratorType;
IteratorType iterator1(inputImage,inputImage->GetRequestedRegion());
int lng = sizeX*sizeY*sizeZ;
for(int i=0; i<lng; i++)
{
iterator1.Set((float)volin[i]);
++iterator1;
}
} // end if(rawInput)
else
{
typedef itk::ImageFileReader< ImageType > ReaderType;
ReaderType::Pointer reader = ReaderType::New();
inputImage = reader->GetOutput();
reader->SetFileName( inputFileName.c_str() );
try
{
reader->Update();
}
catch( itk::ExceptionObject & err )
{
cerr << "ExceptionObject caught!" << endl;
cerr << err << endl;
return EXIT_FAILURE;
}
} // end of itk image read
// ---------------Linear Mapping --------------------//
typedef itk::RescaleIntensityImageFilter<
ImageType, ImageType> RescaleFilterType;
RescaleFilterType::Pointer rescaleFilter = RescaleFilterType::New();
rescaleFilter->SetInput(inputImage);
rescaleFilter->SetOutputMinimum( 0 );
rescaleFilter->SetOutputMaximum( 255 );
try
{
rescaleFilter->Update();
}
catch( itk::ExceptionObject & err )
{
cerr << "ExceptionObject caught!" << endl;
cerr << err << endl;
return EXIT_FAILURE;
}
inputImage = rescaleFilter->GetOutput();
cout << "The Linear Mapping is done\n";
# if Curvature_Anistropic_Diffusion
{
cout << "The Curvature Diffusion is doing\n";
typedef itk::CurvatureAnisotropicDiffusionImageFilter<
ImageType, ImageType > MCD_FilterType;
MCD_FilterType::Pointer MCDFilter = MCD_FilterType::New();
//Initialnization, using the paper's optimal parameters
const unsigned int numberOfIterations = 5;
const double timeStep = 0.0425;
const double conductance = 3;
MCDFilter->SetNumberOfIterations(numberOfIterations);
MCDFilter->SetTimeStep( timeStep );
MCDFilter->SetConductanceParameter( conductance );
MCDFilter->SetInput(inputImage);
try
{
MCDFilter->Update();
}
catch( itk::ExceptionObject & err )
{
cerr << "ExceptionObject caught!" << endl;
cerr << err << endl;
return EXIT_FAILURE;
}
inputImage=MCDFilter->GetOutput();
cout << "The Curvature Diffusion is done\n";
}
#endif
ImageType::RegionType region = inputImage->GetBufferedRegion();
ImageType::SizeType size = region.GetSize();
cout << "input image size: " << size << endl;
sizeX = size[0];
sizeY = size[1];
sizeZ = size[2];
itk::ImageRegionIterator< ImageType >
itr( inputImage, inputImage->GetBufferedRegion() );
itr.GoToBegin();
idx = 0;
volin = (DATATYPEIN*)malloc(sizeX*sizeY*(sizeZ+sizeExpand*2)*sizeof(DATATYPEIN));
while( ! itr.IsAtEnd() )
{
volin[idx] = itr.Get();
++itr;
++idx;
}
//allocate memory for the output image
volout = (DATATYPEOUT*)malloc(sizeX*sizeY*(sizeZ+sizeExpand*2)*sizeof(DATATYPEOUT));
// one pre-processing scheme
GraphcutResults = (unsigned short*)malloc(sizeX*sizeY*(sizeZ+sizeExpand*2)*sizeof(unsigned short));
Neuron_Binarization_3D(volin,GraphcutResults,sizeX,sizeY,sizeZ,0,1);
for (k=0; k<(sizeZ+sizeExpand*2); k++)
for (j=0; j<sizeY; j++)
for (i=0; i<sizeX; i++) {
volout[k *sizeX*sizeY + j *sizeX + i] = 0;
} //initial to zeros
std::cout << "Do you think we need the distance transform to make the centerline of image become bright with higher intensity?";
char tmpAnswer = 'n';
//tmpAnswer = getchar();
if (tmpAnswer =='Y' || tmpAnswer =='y')
{
unsigned char *GraphcutResultsTmp;
GraphcutResultsTmp = (unsigned char*)malloc(sizeX*sizeY*(sizeZ+sizeExpand*2)*sizeof(unsigned char));
for (k=0; k<(sizeZ+sizeExpand*2); k++)
for (j=0; j<sizeY; j++)
for (i=0; i<sizeX; i++)
{
GraphcutResultsTmp[k *sizeX*sizeY + j *sizeX + i] = (unsigned char) GraphcutResults[k *sizeX*sizeY + j *sizeX + i];
} //initial to zeros
distTransform(GraphcutResultsTmp,sizeX,sizeY,sizeZ);
for (k=0; k<sizeZ; k++)
{
// threshold first
for (j=0; j<sizeY; j++)
{
for (i=0; i<sizeX; i++)
{
idx = k *sizeX*sizeY + j *sizeX + i;
volin[idx] = GraphcutResultsTmp [idx];
} // end for
} // end for
} // end for
free(GraphcutResultsTmp);
GraphcutResultsTmp=NULL;
} // end if
else {
for (k=0; k<sizeZ; k++)
{
// threshold first
for (j=0; j<sizeY; j++)
{
for (i=0; i<sizeX; i++)
{
idx = k *sizeX*sizeY + j *sizeX + i;
if (GraphcutResults [idx] == 0)
{
volin[idx] = 0;
}
}
}
}
} // end else
free(GraphcutResults);
GraphcutResults=NULL;
// the secpnd Pre-Processing method, corresponding to the old version MDL
/*
double threshold;
// by xiao liang, using 3 sigma theory to estimate threshold;
double meanValue =0.0, VarianceValue = 0.0;
// threshold first
for (k=0; k<sizeZ; k++)
{
for (j=0; j<sizeY; j++)
{
for (i=0; i<sizeX; i++)
{
idx = k *sizeX*sizeY + j *sizeX + i;
meanValue += volin[idx];
}
}
}
meanValue = meanValue/(double)(sizeX*sizeY*sizeZ);
// threshold first
for (k=0; k<sizeZ; k++)
{
for (j=0; j<sizeY; j++)
{
for (i=0; i<sizeX; i++)
{
idx = k *sizeX*sizeY + j *sizeX + i;
VarianceValue += (volin[idx]-meanValue)*(volin[idx]-meanValue);
}
}
}
VarianceValue = VarianceValue/(double)(sizeX*sizeY*sizeZ);
VarianceValue = sqrt(VarianceValue);
double m_threshold=OtsuThreshold (sizeX,sizeY,sizeZ);
if (m_threshold > 7 || m_threshold < 0)
{
threshold =(meanValue-VarianceValue/30);
}
else
{
threshold = m_threshold;
}
//threshold =7;
cout << "OTSU optimal threshold " << threshold << endl;
for (k=0; k<(sizeZ+sizeExpand*2); k++)
for (j=0; j<sizeY; j++)
for (i=0; i<sizeX; i++) {
volout[k *sizeX*sizeY + j *sizeX + i] = 0;
} //initial to zeros
for (k=0; k<sizeZ; k++)
{
// threshold first
for (j=0; j<sizeY; j++)
{
for (i=0; i<sizeX; i++)
{
idx = k *sizeX*sizeY + j *sizeX + i;
if (volin[idx] < threshold)
{
volin[idx] = 0;
}
}
}
}
*/
// Method 2: Dilation of the object
timesDilate = 1;
border = 3;
while (timesDilate >0 )
{
for (k=border; k<sizeZ-border; k++)
{
for (j=border; j<sizeY-border; j++)
{
for (i=border; i<sizeX-border; i++)
{
blockMax = volin[k *sizeX*sizeY + j *sizeX + i];
for (kk=-1; kk<=1; kk++)
{
for (jj=-1; jj<=1; jj++)
{
for (ii=-1; ii<=1; ii++)
{
if(volin[(k+kk)*sizeX*sizeY + (j+jj)*sizeX + (i+ii)] > blockMax)
{
blockMax = volin[(k+kk)*sizeX*sizeY + (j+jj)*sizeX + (i+ii)];
}
}
}
}
// Keep the peak of the original intensity
if (blockMax == volin[k *sizeX*sizeY + j *sizeX + i] && blockMax != 0)
{
blockMax = blockMax + 1;
//if (blockMax > 255) blockMax = 255;
}
volout[k *sizeX*sizeY + j *sizeX + i] = blockMax;
}
}
}
// copy volout back to volin for the next dilation
for (k=0; k<sizeZ; k++)
{
for (j=0; j<sizeY; j++)
{
for (i=0; i<sizeX; i++)
{
volin[k *sizeX*sizeY + j *sizeX + i] = volout[k *sizeX*sizeY + j *sizeX + i];
}
}
}
timesDilate--;
}
//----- Image write
/*
typedef itk::ImageRegionIteratorWithIndex< ImageType > IteratorType;
IteratorType iterator1(inputImage,inputImage->GetRequestedRegion());
int lng = sizeX*sizeY*sizeZ;
for(int i=0; i<lng; i++)
{
iterator1.Set((float)volin[i]);
++iterator1;
}
*/
//write the output image and free memory
fwrite(volout, sizeX*sizeY*sizeZ, sizeof(DATATYPEOUT), outfile);
FILE *mhdfile;
if((mhdfile=fopen("volume_Processed.mhd","w"))==NULL)
{
cerr << "output file open error!" << endl;
return -1;
}
fprintf (mhdfile,"ObjectType = Image\n");
fprintf (mhdfile,"NDims = 3\n");
fprintf (mhdfile,"BinaryData = True\n");
fprintf (mhdfile,"BinaryDataByteOrderMSB = False\n");
fprintf (mhdfile,"CompressedData = False\n");
fprintf (mhdfile,"TransformMatrix = 1 0 0 0 1 0 0 0 1\n");
fprintf (mhdfile,"Offset = 0 0 0\n");
fprintf (mhdfile,"CenterOfRotation = 0 0 0\n");
fprintf (mhdfile,"AnatomicalOrientation = RAI\n");
fprintf (mhdfile,"ElementSpacing = 1 1 1\n");
fprintf (mhdfile,"DimSize = %d %d %d\n",sizeX,sizeY,sizeZ);
fprintf (mhdfile,"ElementType = MET_UCHAR\n");
fprintf (mhdfile,"ElementDataFile = volume_Processed.raw\n");
fclose(mhdfile);
if (rawInput)
fclose(infile);
fclose(outfile);
free(volin); // by xiao
free(volout); // by xiao
volin=NULL;
volout=NULL;
//cout << "Done" << endl;
return 0;
}
void RunRegistration() {
_transform = TransformType::New();
OptiReporter::Pointer optiReporter = OptiReporter::New();
Metric::Pointer metric = Metric::New();
metric->SetFixedImage(_dst);
bool useIndexes = false;
if (useIndexes) {
_centerOfRotation.SetSize(ImageType::ImageDimension);
for (int i = 0; i < ImageType::ImageDimension; i++) {
_centerOfRotation[i] = i;
}
itk::ImageRegionConstIteratorWithIndex<LabelType> labelIter(_dstLabel, _dstLabel->GetBufferedRegion());
int nPixels = 0;
for (labelIter.GoToBegin(); !labelIter.IsAtEnd(); ++labelIter) {
LabelType::PixelType label = labelIter.Get();
if (label > 0) {
_labelIndexes.push_back(labelIter.GetIndex());
for (int i = 0; i < ImageType::ImageDimension; i++) {
_centerOfRotation[i] += labelIter.GetIndex()[i];
}
nPixels ++;
}
}
for (int i = 0; i < ImageType::ImageDimension; i++) {
_centerOfRotation[i] /= nPixels;
}
metric->SetFixedImageIndexes(_labelIndexes);
_transform->SetFixedParameters(_centerOfRotation);
} else {
metric->SetFixedImageRegion(_dst->GetBufferedRegion());
metric->SetUseAllPixels(true);
_centerOfRotation.SetSize(ImageType::ImageDimension);
for (int i = 0; i < 3; i++) {
_centerOfRotation[i] = _dstCenter[i];
}
_transform->SetFixedParameters(_centerOfRotation);
}
cout << "Fixed Parameters: " << _centerOfRotation << endl;
metric->SetMovingImage(_src);
metric->SetInterpolator(Interpolator::New());
metric->SetTransform(_transform);
metric->Initialize();
Optimizer::Pointer opti = Optimizer::New();
opti->SetCostFunction(metric);
Optimizer::ScalesType scales;
scales.SetSize(TransformType::ParametersDimension);
scales.Fill(1);
if (_method == "affine") {
cout << "apply affine scaling ..." << endl;
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
if (i == j) {
scales[3*i+j] = 160;
} else {
scales[3*i+j] = 30;
}
}
}
scales[9] = scales[10] = scales[11] = 0.1;
} else if (_method == "scale") {
scales[0] = scales[1] = scales[2] = 30;
scales[3] = scales[4] = scales[5] = .5;
scales[6] = scales[7] = scales[8] = 100;
} else if (_method == "similar") {
scales[0] = scales[1] = scales[2] = 10;
scales[3] = scales[4] = scales[5] = 0.5;
scales[6] = 100;
}
opti->SetScales(scales);
const int maxIters = 100;
#ifdef USE_CG_OPTIMIZER
opti->SetMaximumIteration(maxIters);
opti->SetMaximumLineIteration(10);
opti->SetUseUnitLengthGradient(true);
opti->SetStepLength(1);
opti->SetToFletchReeves();
#endif
#ifdef USE_GD_OPTIMIZER
opti->SetNumberOfIterations(maxIters);
opti->SetMinimumStepLength(1e-4);
opti->SetMaximumStepLength(3);
opti->SetRelaxationFactor(.5);
opti->SetGradientMagnitudeTolerance(1e-4);
#endif
opti->SetInitialPosition(_transform->GetParameters());
opti->AddObserver(itk::StartEvent(), optiReporter);
opti->AddObserver(itk::IterationEvent(), optiReporter);
opti->StartOptimization();
cout << "Current Cost: " << opti->GetValue() << endl;
_transformResult = opti->GetCurrentPosition();
_transform->SetParameters(opti->GetCurrentPosition());
}
//calculate segmentation values
bool CTImageTreeItem::internalGetSegmentationValues( SegmentationValues &values) const {
//get ITK image
ImageType::Pointer image = getITKImage();
if (image.IsNull())
return false;
//get buffered region of the image
ImageType::RegionType ctregion = image->GetBufferedRegion();
//define an iterator for the binary segment
typedef itk::ImageRegionConstIteratorWithIndex< BinaryImageType > BinaryIteratorType;
//get binary segment
BinaryImageTreeItem::ImageType::Pointer segment = values.m_segment->getITKImage();
if (segment.IsNull())
return false;
/* typedef itk::ImageFileWriter< ImageType > WriterType;
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( "test.dcm" );
writer->SetInput( image );
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception catched !" << std::endl;
std::cerr << excep << std::endl;
}
*/
//create a binary iterator for the segment and its buffered region
BinaryIteratorType binIter( segment, segment->GetBufferedRegion() );
ImageType::PointType point;
//The Accumulators Framework is framework for performing incremental calculations
using namespace boost::accumulators;
//http://boost-sandbox.sourceforge.net/libs/accumulators/doc/html/accumulators/user_s_guide.html#accumulators.user_s_guide.the_accumulators_framework
accumulator_set<double,features<tag::count, tag::min, tag::mean, tag::max, tag::variance> > acc;
//check selected accuracy
if (values.m_accuracy == SegmentationValues::SimpleAccuracy) {
ImageType::IndexType index;
//iterate over the pixel of the binary segment
for(binIter.GoToBegin(); !binIter.IsAtEnd(); ++binIter) {
//if actual value = 255
if (binIter.Value() == BinaryPixelOn) {
//transforms the index to a physical point in the binary segment
segment->TransformIndexToPhysicalPoint(binIter.GetIndex(),point);
//transform that point to an index of the CT image
image->TransformPhysicalPointToIndex(point, index);
//check if that index is inside the CT region
if (ctregion.IsInside(index)) {
//get the pixel value at the index
int t = image->GetPixel(index);
//check if pixel value != -2048
if (isRealHUvalue(t)) {
//accumulate pixel value
acc( t );
}
}
}
}
//check selected accuracy
} else if (values.m_accuracy == SegmentationValues::PreventDoubleSamplingAccuracy) {
ImageType::IndexType index;
//definition for a set of indices, which can be compared
typedef std::set< ImageType::IndexType, IndexCompareFunctor > IndexSetType;
IndexSetType indexSet;
//iterate over the pixel of the binary segment
for(binIter.GoToBegin(); !binIter.IsAtEnd(); ++binIter) {
//if actual value = 255
if (binIter.Value() == BinaryPixelOn) {
//transforms the index to a physical point in the binary segment
segment->TransformIndexToPhysicalPoint(binIter.GetIndex(),point);
//transform that point to an index of the CT image
image->TransformPhysicalPointToIndex(point, index);
//check if that index is inside the CT region
if (ctregion.IsInside(index)) {
std::pair<IndexSetType::iterator,IndexSetType::iterator> ret;
//
ret = indexSet.equal_range(index);
//If x does not match any key in the container, the range returned has a length of zero,
//with both iterators pointing to the nearest value greater than x, if any,
//or to set::end if x is greater than all the elements in the container.
if (ret.first == ret.second) {
indexSet.insert(ret.first, index);
//get the pixel value at the index
int t = image->GetPixel(index);
//check if pixel value != -2048
if (isRealHUvalue(t)) {
//accumulate pixel value
acc( t );
}
}
}
}
}
//check selected accuracy
} else if (values.m_accuracy == SegmentationValues::InterpolatedAccuracy) {
//define an interpolate function
typedef itk::LinearInterpolateImageFunction< CTImageType > InterpolatorType;
InterpolatorType::Pointer interpolator = InterpolatorType::New();
//set input to the interpolator
interpolator->SetInputImage( image );
InterpolatorType::ContinuousIndexType index;
//iterate over the pixel of the binary segment
for(binIter.GoToBegin(); !binIter.IsAtEnd(); ++binIter) {
//if actual value = 255
if (binIter.Value() == BinaryPixelOn) {
//transforms the index to a physical point in the binary segment
segment->TransformIndexToPhysicalPoint(binIter.GetIndex(),point);
//transform that point to an index of the CT image
image->TransformPhysicalPointToContinuousIndex(point, index);
//check if the index is inside the buffer of the interpolator
if (interpolator->IsInsideBuffer(index)) {
//Returns the linearly interpolated image intensity
int t = interpolator->EvaluateAtContinuousIndex(index);
//check if pixel value != -2048
if (isRealHUvalue(t)) {
//accumulate pixel value
acc( t );
}
}
}
}
}
//set sample count, min, mean, max and standard deviation
//from the accumulated intensities
values.m_sampleCount = count( acc );
values.m_min = min( acc );
values.m_mean = mean( acc );
values.m_max = max( acc );
values.m_stddev = std::sqrt( variance( acc ) );
//set the image modification time
values.m_mtime = segment->GetMTime();
const_cast<CTImageTreeItem*>(this)->m_segmentationValueCache[ values.m_segment ] = values;
return values.m_sampleCount > 0;
}
int main( int argc, char* argv[] )
{
if( argc != 3 )
{
std::cerr << "Usage: "<< std::endl;
std::cerr << argv[0];
std::cerr << " <InputFileName> n";
std::cerr << std::endl;
return EXIT_FAILURE;
}
int operations = atoi(argv[2]);
//sscanf(&operations,"%d",argv[2]);
//printf("%d\n", operations);
itk::TimeProbe itkClock;
double t0 = 0.0;
double tf = 0.0;
itk::MultiThreader::SetGlobalDefaultNumberOfThreads(1);
// Loading file
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
reader->Update();
ImageType::Pointer image = reader->GetOutput();
#ifdef GPU
GPUReaderType::Pointer gpureader = GPUReaderType::New();
gpureader->SetFileName( argv[1] );
gpureader->Update();
GPUImageType::Pointer gpuImage = gpureader->GetOutput();
#endif
saveFile((char*) "/tmp/itk_input.dcm", image);
// Allocate output image
ImageType::Pointer output = ImageType::New();
ImageType::RegionType region = image->GetBufferedRegion();
output->SetRegions( region );
output->SetOrigin( image->GetOrigin() );
output->SetSpacing( image->GetSpacing() );
output->Allocate();
// Negative
typedef itk::UnaryFunctorImageFilter<ImageType,ImageType,
Negate<ImageType::PixelType,ImageType::PixelType> > NegateImageFilterType;
NegateImageFilterType::Pointer negateFilter = NegateImageFilterType::New();
negateFilter = NegateImageFilterType::New();
negateFilter->SetInput(image);
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
negateFilter->Modified();
negateFilter->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d negative: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
#endif
// Saving Not result
saveFile((char*) "/tmp/itk_not.dcm", negateFilter->GetOutput());
#ifdef GPU
// GPU Negative
typedef itk::GPUUnaryFunctorImageFilter<ImageType,ImageType,
Negate<ImageType::PixelType,ImageType::PixelType> > GPUNegateImageFilterType;
GPUNegateImageFilterType::Pointer gpuNegateFilter = GPUNegateImageFilterType::New();
gpuNegateFilter->SetInput(gpureader->GetOutput());
gpuNegateFilter->Update();
// Saving Not result
//saveFile("/tmp/itk_gpu_not.dcm", gpuNegateFilter->GetOutput());
#endif
// Common Threshold
int lowerThreshold = 100;
int upperThreshold = 200;
// Threshold
typedef itk::BinaryThresholdImageFilter <ImageType, ImageType>
BinaryThresholdImageFilterType;
BinaryThresholdImageFilterType::Pointer thresholdFilter
= BinaryThresholdImageFilterType::New();
thresholdFilter = BinaryThresholdImageFilterType::New();
thresholdFilter->SetInput(reader->GetOutput());
thresholdFilter->SetLowerThreshold(lowerThreshold);
thresholdFilter->SetUpperThreshold(upperThreshold);
thresholdFilter->SetInsideValue(255);
thresholdFilter->SetOutsideValue(0);
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
thresholdFilter->Modified();
thresholdFilter->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d threshold: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Threshold result
saveFile((char*) "/tmp/itk_thresh.dcm", thresholdFilter->GetOutput());
#endif
#ifdef GPU
// GPU Threshold
typedef itk::GPUBinaryThresholdImageFilter <GPUImageType, GPUImageType>
GPUBinaryThresholdImageFilterType;
GPUBinaryThresholdImageFilterType::Pointer gpuThresholdFilter
= GPUBinaryThresholdImageFilterType::New();
gpuThresholdFilter->SetInput(gpureader->GetOutput());
gpuThresholdFilter->SetLowerThreshold(lowerThreshold);
gpuThresholdFilter->SetUpperThreshold(upperThreshold);
gpuThresholdFilter->SetInsideValue(255);
gpuThresholdFilter->SetOutsideValue(0);
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
gpuThresholdFilter->Modified();
gpuThresholdFilter->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d GPU threshold: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving GPU Threshold result
gpuThresholdFilter->GetOutput()->UpdateBuffers();
saveFile((char*) "/tmp/itk_gpu_thresh.dcm", gpuThresholdFilter->GetOutput());
#endif
// Mean
typedef itk::MeanImageFilter< ImageType, ImageType > MeanFilterType;
MeanFilterType::Pointer meanFilter = MeanFilterType::New();
meanFilter = MeanFilterType::New();
meanFilter->SetInput( image );
meanFilter->SetRadius( 1 );
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
meanFilter->Modified();
meanFilter->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d mean blur: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Convolution result
saveFile((char*) "/tmp/itk_mean3x3.dcm", meanFilter->GetOutput());
#endif
// Binomial Blur (aproximation of gaussian blur)
typedef itk::BinomialBlurImageFilter<ImageType, ImageType> BinomialBlurImageFilterType;
int repetitions = 1;
BinomialBlurImageFilterType::Pointer blurFilter = BinomialBlurImageFilterType::New();
blurFilter = BinomialBlurImageFilterType::New();
blurFilter->SetInput( reader->GetOutput() );
blurFilter->SetRepetitions( repetitions );
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
blurFilter->Modified();
blurFilter->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d blur: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Blur result
saveFile((char*) "/tmp/itk_blur.dcm", blurFilter->GetOutput());
#endif
#ifdef GPU
// GPU Blur
typedef itk::BoxImageFilter< GPUImageType, GPUImageType > BoxImageFilterType;
typedef itk::GPUBoxImageFilter< GPUImageType, GPUImageType, BoxImageFilterType > GPUBoxImageFilterType;
GPUBoxImageFilterType::Pointer GPUBlurFilter = GPUBoxImageFilterType::New();
//ImageType::SizeType indexRadius;
//indexRadius[0] = 2;
//indexRadius[1] = 2;
//indexRadius[2] = 2;
GPUBlurFilter->SetInput(gpureader->GetOutput());
//GPUBlurFilter->SetRadius(indexRadius);
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
GPUBlurFilter->Update();
GPUBlurFilter->Modified();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d gpu blur: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
GPUBlurFilter->GetOutput()->UpdateBuffers();
// Saving GPU Blur result
saveFile((char*) "/tmp/itk_gpu_blur.dcm", GPUBlurFilter->GetOutput());
#endif
//Erosion Common
typedef itk::BinaryBallStructuringElement<
ImageType::PixelType, 3> StructuringElementType;
typedef itk::GrayscaleErodeImageFilter <ImageType, ImageType, StructuringElementType>
GrayscaleErodeImageFilterType;
unsigned int radius;
// Erosion 3x3
StructuringElementType structuringElement3x3;
radius = 1;
structuringElement3x3.SetRadius(radius);
structuringElement3x3.CreateStructuringElement();
GrayscaleErodeImageFilterType::Pointer erodeFilter3x3;
erodeFilter3x3= GrayscaleErodeImageFilterType::New();
erodeFilter3x3->SetInput(reader->GetOutput());
erodeFilter3x3->SetKernel(structuringElement3x3);
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
erodeFilter3x3->Modified();
erodeFilter3x3->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d erosion 3x3: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Erosion result
saveFile((char*) "/tmp/itk_erode3x3.dcm", erodeFilter3x3->GetOutput());
#endif
// Erosion 5x5
StructuringElementType structuringElement5x5;
radius = 2;
structuringElement5x5.SetRadius(radius);
structuringElement5x5.CreateStructuringElement();
GrayscaleErodeImageFilterType::Pointer erodeFilter5x5;
erodeFilter5x5 = GrayscaleErodeImageFilterType::New();
erodeFilter5x5->SetInput(reader->GetOutput());
erodeFilter5x5->SetKernel(structuringElement5x5);
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
erodeFilter5x5->Modified();
erodeFilter5x5->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d erosion 5x5: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Erosion result
saveFile((char*) "/tmp/itk_erode5x5.dcm", erodeFilter5x5->GetOutput());
#endif
// Copy
typedef itk::ImageDuplicator< ImageType > DuplicatorType;
DuplicatorType::Pointer duplicator;
duplicator = DuplicatorType::New();
duplicator->SetInputImage(image);
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
duplicator->Modified();
duplicator->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d copias cpu: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Copy result
saveFile((char*) "/tmp/itk_copy.dcm", duplicator->GetOutput());
#endif
// Convolution common
typedef itk::ConvolutionImageFilter<ImageType> ConvolutionImageFilterType;
ConvolutionImageFilterType::Pointer convolutionFilter;
convolutionFilter = ConvolutionImageFilterType::New();
convolutionFilter->SetInput(reader->GetOutput());
int convWidth;
// Convolution 3x3
ImageType::Pointer kernel3x3 = ImageType::New();
convWidth = 3;
CreateKernel(kernel3x3, convWidth);
convolutionFilter->SetKernelImage(kernel3x3);
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
convolutionFilter->Modified();
convolutionFilter->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d convolucoes 3x3 cpu: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Convolution result
saveFile((char*) "/tmp/itk_convolution3x3.dcm", convolutionFilter->GetOutput());
#endif
// Convolution 5x5
ImageType::Pointer kernel5x5 = ImageType::New();
convWidth = 5;
CreateKernel(kernel5x5, convWidth);
convolutionFilter->SetKernelImage(kernel5x5);
#ifndef GPU_only
itkClock.Start();
TimerStart();
for(int n = 0; n < operations; n++)
{
convolutionFilter->Modified();
convolutionFilter->Update();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d convolucoes 5x5 cpu: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
// Saving Convolution result
saveFile((char*) "/tmp/itk_convolution5x5.dcm", convolutionFilter->GetOutput());
#endif
#ifdef GPU
// GPU Mean
typedef itk::GPUMeanImageFilter<GPUImageType, GPUImageType> GPUMeanFilterType;
GPUMeanFilterType::Pointer GPUMean = GPUMeanFilterType::New();
GPUMean->SetInput(gpureader->GetOutput());
GPUMean->SetRadius( 1 );
TimerStart();
for(int n = 0; n < operations; n++)
{
GPUMean->Update();
GPUMean->Modified();
}
itkClock.Stop();
printf("Tempo gasto para fazer %d GPU mean blur: %s\n",operations, getTimeElapsedInSeconds());
tf = itkClock.GetTotal();
std::cout << "My: " << (tf - t0) << std::endl;
t0 = tf;
GPUMean->GetOutput()->UpdateBuffers();
saveFile((char*) "/tmp/itk_gpu_blurmean.dcm", GPUMean->GetOutput());
#endif
// Visualize
/*
QuickView viewer;
viewer.AddImage<ImageType>(
image,true,
itksys::SystemTools::GetFilenameName(argv[1]));
std::stringstream desc;
desc << "ITK QuickView: "
<< argv[1];
viewer.Visualize();
*/
// Saving input image as is
saveFile((char*) "/tmp/itk_input.dcm", image);
return EXIT_SUCCESS;
}
// This function is designed to compute the optimal threshold using OTSU method;
// this algoritm is implemented by xiao liang based on ITK's OTSU algorithm
double VolumeProcess::getXiaoLiangOtsuThreshold(ImageType::Pointer img)
{
if(!img)
return 0;
double threshold = 0;
unsigned char m_min = 255;
unsigned char m_max = 0;
double m_mean = 0.0;
double m_variance = 0.0;
ImageType::RegionType region = img->GetBufferedRegion();
double numPix = region.GetSize(2)*region.GetSize(1)*region.GetSize(0);
//Get min, max, and mean:
itk::ImageRegionIterator< ImageType > itr( img, region );
for(itr.GoToBegin(); !itr.IsAtEnd(); ++itr)
{
double val = itr.Get();
if(val > m_max) m_max = val;
if(val < m_min) m_min = val;
m_mean += itr.Get();
}
m_mean = m_mean/numPix;
if(debug)
std::cerr << "Max = " << (int)m_max << ", Min = " << (int)m_min << std::endl;
//Do a sanity check
if ( m_min >= m_max)
{
threshold=m_min;
return threshold;
}
//Get the variance:
for(itr.GoToBegin(); !itr.IsAtEnd(); ++itr)
{
double val = (double)itr.Get();
m_variance += (val-m_mean)*(val-m_mean);
}
//These were not Xiao Liang's version:
//m_variance = m_variance / numPix;
//m_variance = sqrt(m_variance);
threshold = m_mean - (m_variance/30);
// this step is only initialized a good experimental value for m_Threshold, because the 3D image
// is sparse, there are lots of zero values;
//Create a histogram & init to zero
double relativeFrequency[m_NumberOfHistogramBins];
for ( unsigned char j = 0; j < m_NumberOfHistogramBins; j++ )
{
relativeFrequency[j] = 0.0;
}
double binMultiplier = (double)m_NumberOfHistogramBins/(double)(m_max-m_min);
if(debug)
std::cerr << "binMultiplier = " << binMultiplier << std::endl;
unsigned int binNumber;
for(itr.GoToBegin(); !itr.IsAtEnd(); ++itr)
{
double val = itr.Get();
if ( val == m_min )
{
binNumber = 0;
}
else
{
binNumber = (unsigned int)(((val-m_min)*binMultiplier) - 1);
if ( binNumber == m_NumberOfHistogramBins ) // in case of rounding errors
{
binNumber -= 1;
}
}
relativeFrequency[binNumber] += 1.0;
}
// normalize the frequencies
double totalMean = 0.0;
for ( unsigned char j = 0; j < m_NumberOfHistogramBins; j++ )
{
relativeFrequency[j] /= numPix;
totalMean += (j+1) * relativeFrequency[j];
}
// compute Otsu's threshold by maximizing the between-class variance
double freqLeft = relativeFrequency[0];
double meanLeft = 1.0;
double meanRight = ( totalMean - freqLeft ) / ( 1.0 - freqLeft );
double maxVarBetween = freqLeft * ( 1.0 - freqLeft ) * sqrt( meanLeft - meanRight );
int maxBinNumber = 0;
double freqLeftOld = freqLeft;
double meanLeftOld = meanLeft;
for ( unsigned char j = 1; j < m_NumberOfHistogramBins; j++ )
{
freqLeft += relativeFrequency[j];
meanLeft = ( ((meanLeftOld * freqLeftOld)+(j+1)) * relativeFrequency[j] ) / freqLeft;
if (freqLeft == 1.0)
{
meanRight = 0.0;
}
else
{
meanRight = ( totalMean - meanLeft * freqLeft ) / ( 1.0 - freqLeft );
}
double varBetween = freqLeft * ( 1.0 - freqLeft ) * sqrt( meanLeft - meanRight );
if ( varBetween > maxVarBetween )
{
maxVarBetween = varBetween;
maxBinNumber = j;
}
// cache old values
freqLeftOld = freqLeft;
meanLeftOld = meanLeft;
}
threshold = double( m_min + ( maxBinNumber + 1 ) / binMultiplier );
return threshold;
}
