diff --git a/Modules/Classification/CLMiniApps/CL2Dto3DImage.cpp b/Modules/Classification/CLMiniApps/CL2Dto3DImage.cpp new file mode 100644 index 0000000000..6d49b49a00 --- /dev/null +++ b/Modules/Classification/CLMiniApps/CL2Dto3DImage.cpp @@ -0,0 +1,60 @@ +/*=================================================================== + +The Medical Imaging Interaction Toolkit (MITK) + +Copyright (c) German Cancer Research Center, +Division of Medical and Biological Informatics. +All rights reserved. + +This software is distributed WITHOUT ANY WARRANTY; without +even the implied warranty of MERCHANTABILITY or FITNESS FOR +A PARTICULAR PURPOSE. + +See LICENSE.txt or http://www.mitk.org for details. + +===================================================================*/ + +#include "mitkCommandLineParser.h" +#include +#include "mitkIOUtil.h" + +int main(int argc, char* argv[]) +{ + mitkCommandLineParser parser; + + parser.setTitle("Dicom Loader"); + parser.setCategory("Preprocessing Tools"); + parser.setDescription(""); + parser.setContributor("MBI"); + + parser.setArgumentPrefix("--","-"); + // Add command line argument names + parser.addArgument("help", "h",mitkCommandLineParser::Bool, "Help:", "Show this help text"); + parser.addArgument("input", "i", mitkCommandLineParser::InputDirectory, "Input file:", "Input file",us::Any(),false); + parser.addArgument("output", "o", mitkCommandLineParser::OutputFile, "Output file:", "Output file",us::Any(),false); + + std::map parsedArgs = parser.parseArguments(argc, argv); + + if (parsedArgs.size()==0) + return EXIT_FAILURE; + + // Show a help message + if ( parsedArgs.count("help") || parsedArgs.count("h")) + { + std::cout << parser.helpText(); + return EXIT_SUCCESS; + } + + std::string inputName = us::any_cast(parsedArgs["input"]); + std::string outputName = us::any_cast(parsedArgs["output"]); + + mitk::Image::Pointer image = mitk::IOUtil::LoadImage(inputName); + mitk::Convert2Dto3DImageFilter::Pointer multiFilter2 = mitk::Convert2Dto3DImageFilter::New(); + multiFilter2->SetInput(image); + multiFilter2->Update(); + mitk::Image::Pointer image2 = multiFilter2->GetOutput(); + + mitk::IOUtil::SaveImage(image2, outputName); + + return EXIT_SUCCESS; +} \ No newline at end of file diff --git a/Modules/Classification/CLMiniApps/CLGlobalImageFeatures.cpp b/Modules/Classification/CLMiniApps/CLGlobalImageFeatures.cpp index c888cb42e5..a7c96568ad 100644 --- a/Modules/Classification/CLMiniApps/CLGlobalImageFeatures.cpp +++ b/Modules/Classification/CLMiniApps/CLGlobalImageFeatures.cpp @@ -1,563 +1,593 @@ /*=================================================================== The Medical Imaging Interaction Toolkit (MITK) Copyright (c) German Cancer Research Center, Division of Medical and Biological Informatics. All rights reserved. This software is distributed WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See LICENSE.txt or http://www.mitk.org for details. ===================================================================*/ #ifndef mitkCLPolyToNrrd_cpp #define mitkCLPolyToNrrd_cpp #include "time.h" #include #include #include #include "mitkCommandLineParser.h" #include #include #include #include #include #include #include #include #include +#include + #include #include #include "itkNearestNeighborInterpolateImageFunction.h" #include "itkResampleImageFilter.h" typedef itk::Image< double, 3 > FloatImageType; typedef itk::Image< unsigned char, 3 > MaskImageType; static std::vector splitDouble(std::string str, char delimiter) { std::vector internal; std::stringstream ss(str); // Turn the string into a stream. std::string tok; double val; while (std::getline(ss, tok, delimiter)) { std::stringstream s2(tok); s2 >> val; internal.push_back(val); } return internal; } template void ResampleImage(itk::Image* itkImage, float resolution, mitk::Image::Pointer& newImage) { typedef itk::Image ImageType; typedef itk::ResampleImageFilter ResampleFilterType; typename ResampleFilterType::Pointer resampler = ResampleFilterType::New(); auto spacing = itkImage->GetSpacing(); auto size = itkImage->GetLargestPossibleRegion().GetSize(); for (int i = 0; i < VImageDimension; ++i) { size[i] = size[i] / (1.0*resolution)*(1.0*spacing[i])+1.0; } spacing.Fill(resolution); resampler->SetInput(itkImage); resampler->SetSize(size); resampler->SetOutputSpacing(spacing); resampler->SetOutputOrigin(itkImage->GetOrigin()); resampler->SetOutputDirection(itkImage->GetDirection()); resampler->Update(); newImage->InitializeByItk(resampler->GetOutput()); mitk::GrabItkImageMemory(resampler->GetOutput(), newImage); } template static void CreateNoNaNMask(itk::Image* itkValue, mitk::Image::Pointer mask, mitk::Image::Pointer& newMask) { typedef itk::Image< TPixel, VImageDimension> LFloatImageType; typedef itk::Image< unsigned char, VImageDimension> LMaskImageType; typename LMaskImageType::Pointer itkMask = LMaskImageType::New(); mitk::CastToItkImage(mask, itkMask); typedef itk::ImageDuplicator< LMaskImageType > DuplicatorType; typename DuplicatorType::Pointer duplicator = DuplicatorType::New(); duplicator->SetInputImage(itkMask); duplicator->Update(); auto tmpMask = duplicator->GetOutput(); itk::ImageRegionIterator mask1Iter(itkMask, itkMask->GetLargestPossibleRegion()); itk::ImageRegionIterator mask2Iter(tmpMask, tmpMask->GetLargestPossibleRegion()); itk::ImageRegionIterator imageIter(itkValue, itkValue->GetLargestPossibleRegion()); while (!mask1Iter.IsAtEnd()) { mask2Iter.Set(0); if (mask1Iter.Value() > 0) { // Is not NaN if (imageIter.Value() == imageIter.Value()) { mask2Iter.Set(1); } } ++mask1Iter; ++mask2Iter; ++imageIter; } newMask->InitializeByItk(tmpMask); mitk::GrabItkImageMemory(tmpMask, newMask); } template static void ResampleMask(itk::Image* itkMoving, mitk::Image::Pointer ref, mitk::Image::Pointer& newMask) { typedef itk::Image< TPixel, VImageDimension> LMaskImageType; typedef itk::NearestNeighborInterpolateImageFunction< LMaskImageType> NearestNeighborInterpolateImageFunctionType; typedef itk::ResampleImageFilter ResampleFilterType; typename NearestNeighborInterpolateImageFunctionType::Pointer nn_interpolator = NearestNeighborInterpolateImageFunctionType::New(); typename LMaskImageType::Pointer itkRef = LMaskImageType::New(); mitk::CastToItkImage(ref, itkRef); typename ResampleFilterType::Pointer resampler = ResampleFilterType::New(); resampler->SetInput(itkMoving); resampler->SetReferenceImage(itkRef); resampler->UseReferenceImageOn(); resampler->SetInterpolator(nn_interpolator); resampler->Update(); newMask->InitializeByItk(resampler->GetOutput()); mitk::GrabItkImageMemory(resampler->GetOutput(), newMask); } static void ExtractSlicesFromImages(mitk::Image::Pointer image, mitk::Image::Pointer mask, mitk::Image::Pointer maskNoNaN, int direction, std::vector &imageVector, std::vector &maskVector, std::vector &maskNoNaNVector) { typedef itk::Image< double, 2 > FloatImage2DType; typedef itk::Image< unsigned char, 2 > MaskImage2DType; FloatImageType::Pointer itkFloat = FloatImageType::New(); MaskImageType::Pointer itkMask = MaskImageType::New(); MaskImageType::Pointer itkMaskNoNaN = MaskImageType::New(); mitk::CastToItkImage(mask, itkMask); mitk::CastToItkImage(maskNoNaN, itkMaskNoNaN); mitk::CastToItkImage(image, itkFloat); int idxA, idxB, idxC; switch (direction) { case 0: idxA = 1; idxB = 2; idxC = 0; break; case 1: idxA = 0; idxB = 2; idxC = 1; break; case 2: idxA = 0; idxB = 1; idxC = 2; break; default: idxA = 1; idxB = 2; idxC = 0; break; } auto imageSize = image->GetLargestPossibleRegion().GetSize(); FloatImageType::IndexType index3D; FloatImage2DType::IndexType index2D; FloatImage2DType::SpacingType spacing2D; spacing2D[0] = itkFloat->GetSpacing()[idxA]; spacing2D[1] = itkFloat->GetSpacing()[idxB]; for (int i = 0; i < imageSize[idxC]; ++i) { FloatImage2DType::RegionType region; FloatImage2DType::IndexType start; FloatImage2DType::SizeType size; start[0] = 0; start[1] = 0; size[0] = imageSize[idxA]; size[1] = imageSize[idxB]; region.SetIndex(start); region.SetSize(size); FloatImage2DType::Pointer image2D = FloatImage2DType::New(); image2D->SetRegions(region); image2D->Allocate(); MaskImage2DType::Pointer mask2D = MaskImage2DType::New(); mask2D->SetRegions(region); mask2D->Allocate(); MaskImage2DType::Pointer masnNoNaN2D = MaskImage2DType::New(); masnNoNaN2D->SetRegions(region); masnNoNaN2D->Allocate(); for (int a = 0; a < imageSize[idxA]; ++a) { for (int b = 0; b < imageSize[idxB]; ++b) { index3D[idxA] = a; index3D[idxB] = b; index3D[idxC] = i; index2D[0] = a; index2D[1] = b; image2D->SetPixel(index2D, itkFloat->GetPixel(index3D)); mask2D->SetPixel(index2D, itkMask->GetPixel(index3D)); masnNoNaN2D->SetPixel(index2D, itkMaskNoNaN->GetPixel(index3D)); } } image2D->SetSpacing(spacing2D); mask2D->SetSpacing(spacing2D); masnNoNaN2D->SetSpacing(spacing2D); mitk::Image::Pointer tmpFloatImage = mitk::Image::New(); tmpFloatImage->InitializeByItk(image2D.GetPointer()); mitk::GrabItkImageMemory(image2D, tmpFloatImage); mitk::Image::Pointer tmpMaskImage = mitk::Image::New(); tmpMaskImage->InitializeByItk(mask2D.GetPointer()); mitk::GrabItkImageMemory(mask2D, tmpMaskImage); mitk::Image::Pointer tmpMaskNoNaNImage = mitk::Image::New(); tmpMaskNoNaNImage->InitializeByItk(masnNoNaN2D.GetPointer()); mitk::GrabItkImageMemory(masnNoNaN2D, tmpMaskNoNaNImage); imageVector.push_back(tmpFloatImage); maskVector.push_back(tmpMaskImage); maskNoNaNVector.push_back(tmpMaskNoNaNImage); } } int main(int argc, char* argv[]) { mitk::GIFFirstOrderStatistics::Pointer firstOrderCalculator = mitk::GIFFirstOrderStatistics::New(); mitk::GIFVolumetricStatistics::Pointer volCalculator = mitk::GIFVolumetricStatistics::New(); mitk::GIFCooccurenceMatrix::Pointer coocCalculator = mitk::GIFCooccurenceMatrix::New(); mitk::GIFCooccurenceMatrix2::Pointer cooc2Calculator = mitk::GIFCooccurenceMatrix2::New(); mitk::GIFNeighbouringGreyLevelDependenceFeature::Pointer ngldCalculator = mitk::GIFNeighbouringGreyLevelDependenceFeature::New(); mitk::GIFGrayLevelRunLength::Pointer rlCalculator = mitk::GIFGrayLevelRunLength::New(); mitkCommandLineParser parser; parser.setArgumentPrefix("--", "-"); // required params parser.addArgument("image", "i", mitkCommandLineParser::InputImage, "Input Image", "Path to the input image file", us::Any(), false); parser.addArgument("mask", "m", mitkCommandLineParser::InputImage, "Input Mask", "Path to the mask Image that specifies the area over for the statistic (Values = 1)", us::Any(), false); parser.addArgument("output", "o", mitkCommandLineParser::OutputFile, "Output text file", "Path to output file. The output statistic is appended to this file.", us::Any(), false); parser.addArgument("--","-", mitkCommandLineParser::String, "---", "---", us::Any(),true); firstOrderCalculator->AddArguments(parser); // Does not support single direction volCalculator->AddArguments(parser); // Does not support single direction coocCalculator->AddArguments(parser); // cooc2Calculator->AddArguments(parser); // Needs parameter fixing ngldCalculator->AddArguments(parser); // Needs parameter fixing rlCalculator->AddArguments(parser); // Does not support single direction parser.addArgument("--", "-", mitkCommandLineParser::String, "---", "---", us::Any(), true); //parser.addArgument("cooccurence", "cooc", mitkCommandLineParser::String, "Use Co-occurence matrix", "calculates Co-occurence based features", us::Any()); //parser.addArgument("cooccurence2", "cooc2", mitkCommandLineParser::String, "Use Co-occurence matrix", "calculates Co-occurence based features (new implementation)", us::Any()); //parser.addArgument("ngldm", "ngldm", mitkCommandLineParser::String, "Calculate Neighbouring grey level dependence based features", "Calculate Neighbouring grey level dependence based features", us::Any()); //parser.addArgument("volume","vol",mitkCommandLineParser::String, "Use Volume-Statistic", "calculates volume based features",us::Any()); //parser.addArgument("run-length","rl",mitkCommandLineParser::String, "Use Co-occurence matrix", "calculates Co-occurence based features",us::Any()); //parser.addArgument("first-order","fo",mitkCommandLineParser::String, "Use First Order Features", "calculates First order based features",us::Any()); parser.addArgument("header","head",mitkCommandLineParser::String,"Add Header (Labels) to output","",us::Any()); parser.addArgument("description","d",mitkCommandLineParser::String,"Text","Description that is added to the output",us::Any()); parser.addArgument("same-space", "sp", mitkCommandLineParser::String, "Bool", "Set the spacing of all images to equal. Otherwise an error will be thrown. ", us::Any()); parser.addArgument("resample-mask", "rm", mitkCommandLineParser::Bool, "Bool", "Resamples the mask to the resolution of the input image ", us::Any()); parser.addArgument("save-resample-mask", "srm", mitkCommandLineParser::String, "String", "If specified the resampled mask is saved to this path (if -rm is 1)", us::Any()); parser.addArgument("fixed-isotropic", "fi", mitkCommandLineParser::Float, "Float", "Input image resampled to fixed isotropic resolution given in mm. Should be used with resample-mask ", us::Any()); parser.addArgument("direction", "dir", mitkCommandLineParser::String, "Int", "Allows to specify the direction for Cooc and RL. 0: All directions, 1: Only single direction (Test purpose), 2,3,4... Without dimension 0,1,2... ", us::Any()); parser.addArgument("slice-wise", "slice", mitkCommandLineParser::String, "Int", "Allows to specify if the image is processed slice-wise (number giving direction) ", us::Any()); parser.addArgument("minimum-intensity", "minimum", mitkCommandLineParser::String, "Float", "Minimum intensity. If set, it is overwritten by more specific intensity minima", us::Any()); parser.addArgument("maximum-intensity", "maximum", mitkCommandLineParser::String, "Float", "Maximum intensity. If set, it is overwritten by more specific intensity maxima", us::Any()); parser.addArgument("bins", "bins", mitkCommandLineParser::String, "Int", "Number of bins if bins are used. If set, it is overwritten by more specific bin count", us::Any()); // Miniapp Infos parser.setCategory("Classification Tools"); parser.setTitle("Global Image Feature calculator"); parser.setDescription("Calculates different global statistics for a given segmentation / image combination"); parser.setContributor("MBI"); std::map parsedArgs = parser.parseArguments(argc, argv); if (parsedArgs.size()==0) { return EXIT_FAILURE; } if ( parsedArgs.count("help") || parsedArgs.count("h")) { return EXIT_SUCCESS; } MITK_INFO << "Version: "<< 1.8; //bool useCooc = parsedArgs.count("cooccurence"); bool resampleMask = false; if (parsedArgs.count("resample-mask")) { resampleMask = us::any_cast(parsedArgs["resample-mask"]); } - mitk::Image::Pointer image = mitk::IOUtil::LoadImage(parsedArgs["image"].ToString()); - mitk::Image::Pointer mask = mitk::IOUtil::LoadImage(parsedArgs["mask"].ToString()); + mitk::Image::Pointer image; + mitk::Image::Pointer mask; + + mitk::Image::Pointer tmpImage = mitk::IOUtil::LoadImage(parsedArgs["image"].ToString()); + mitk::Image::Pointer tmpMask = mitk::IOUtil::LoadImage(parsedArgs["mask"].ToString()); + image = tmpImage; + mask = tmpMask; + + if ((image->GetDimension() != mask->GetDimension())) + { + MITK_INFO << "Dimension of image does not match. "; + MITK_INFO << "Correct one image, may affect the result"; + if (image->GetDimension() == 2) + { + mitk::Convert2Dto3DImageFilter::Pointer multiFilter2 = mitk::Convert2Dto3DImageFilter::New(); + multiFilter2->SetInput(tmpImage); + multiFilter2->Update(); + image = multiFilter2->GetOutput(); + } + if (mask->GetDimension() == 2) + { + mitk::Convert2Dto3DImageFilter::Pointer multiFilter3 = mitk::Convert2Dto3DImageFilter::New(); + multiFilter3->SetInput(tmpMask); + multiFilter3->Update(); + mask = multiFilter3->GetOutput(); + } + } + if (parsedArgs.count("fixed-isotropic")) { mitk::Image::Pointer newImage = mitk::Image::New(); float resolution = us::any_cast(parsedArgs["fixed-isotropic"]); AccessByItk_2(image, ResampleImage, resolution, newImage); image = newImage; } if (resampleMask) { mitk::Image::Pointer newMaskImage = mitk::Image::New(); AccessByItk_2(mask, ResampleMask, image, newMaskImage); mask = newMaskImage; if (parsedArgs.count("save-resample-mask")) { mitk::IOUtil::SaveImage(mask, parsedArgs["save-resample-mask"].ToString()); } } bool fixDifferentSpaces = parsedArgs.count("same-space"); if ( ! mitk::Equal(mask->GetGeometry(0)->GetOrigin(), image->GetGeometry(0)->GetOrigin())) { MITK_INFO << "Not equal Origins"; if (fixDifferentSpaces) { MITK_INFO << "Warning!"; MITK_INFO << "The origin of the input image and the mask do not match. They are"; MITK_INFO << "now corrected. Please check to make sure that the images still match"; image->GetGeometry(0)->SetOrigin(mask->GetGeometry(0)->GetOrigin()); } else { return -1; } } if ( ! mitk::Equal(mask->GetGeometry(0)->GetSpacing(), image->GetGeometry(0)->GetSpacing())) { MITK_INFO << "Not equal Sapcings"; if (fixDifferentSpaces) { MITK_INFO << "Warning!"; MITK_INFO << "The spacing of the mask was set to match the spacing of the input image."; MITK_INFO << "This might cause unintended spacing of the mask image"; image->GetGeometry(0)->SetSpacing(mask->GetGeometry(0)->GetSpacing()); } else { MITK_INFO << "The spacing of the mask and the input images is not equal."; MITK_INFO << "Terminating the programm. You may use the '-fi' option"; return -1; } } int direction = 0; if (parsedArgs.count("direction")) { direction = splitDouble(parsedArgs["direction"].ToString(), ';')[0]; } MITK_INFO << "Start creating Mask without NaN"; mitk::Image::Pointer maskNoNaN = mitk::Image::New(); AccessByItk_2(image, CreateNoNaNMask, mask, maskNoNaN); //CreateNoNaNMask(mask, image, maskNoNaN); bool sliceWise = false; int sliceDirection = 0; int currentSlice = 0; bool imageToProcess = true; std::vector floatVector; std::vector maskVector; std::vector maskNoNaNVector; if (parsedArgs.count("slice-wise")) { MITK_INFO << "Enabled slice-wise"; sliceWise = true; sliceDirection = splitDouble(parsedArgs["slice-wise"].ToString(), ';')[0]; + MITK_INFO << sliceDirection; ExtractSlicesFromImages(image, mask, maskNoNaN, sliceDirection, floatVector, maskVector, maskNoNaNVector); + MITK_INFO << "Slice"; } if (parsedArgs.count("minimum-intensity")) { float minimum = splitDouble(parsedArgs["minimum-intensity"].ToString(), ';')[0]; firstOrderCalculator->SetMinimumIntensity(minimum); firstOrderCalculator->SetUseMinimumIntensity(true); volCalculator->SetMinimumIntensity(minimum); volCalculator->SetUseMinimumIntensity(true); coocCalculator->SetMinimumIntensity(minimum); coocCalculator->SetUseMinimumIntensity(true); cooc2Calculator->SetMinimumIntensity(minimum); cooc2Calculator->SetUseMinimumIntensity(true); ngldCalculator->SetMinimumIntensity(minimum); ngldCalculator->SetUseMinimumIntensity(true); rlCalculator->SetMinimumIntensity(minimum); rlCalculator->SetUseMinimumIntensity(true); } if (parsedArgs.count("maximum-intensity")) { float minimum = splitDouble(parsedArgs["maximum-intensity"].ToString(), ';')[0]; firstOrderCalculator->SetMaximumIntensity(minimum); firstOrderCalculator->SetUseMaximumIntensity(true); volCalculator->SetMaximumIntensity(minimum); volCalculator->SetUseMaximumIntensity(true); coocCalculator->SetMaximumIntensity(minimum); coocCalculator->SetUseMaximumIntensity(true); cooc2Calculator->SetMaximumIntensity(minimum); cooc2Calculator->SetUseMaximumIntensity(true); ngldCalculator->SetMaximumIntensity(minimum); ngldCalculator->SetUseMaximumIntensity(true); rlCalculator->SetMaximumIntensity(minimum); rlCalculator->SetUseMaximumIntensity(true); } if (parsedArgs.count("bins")) { int minimum = splitDouble(parsedArgs["bins"].ToString(), ';')[0]; firstOrderCalculator->SetBins(minimum); volCalculator->SetBins(minimum); coocCalculator->SetBins(minimum); cooc2Calculator->SetBins(minimum); ngldCalculator->SetBins(minimum); rlCalculator->SetBins(minimum); } firstOrderCalculator->SetParameter(parsedArgs); firstOrderCalculator->SetDirection(direction); volCalculator->SetParameter(parsedArgs); volCalculator->SetDirection(direction); coocCalculator->SetParameter(parsedArgs); coocCalculator->SetDirection(direction); cooc2Calculator->SetParameter(parsedArgs); cooc2Calculator->SetDirection(direction); ngldCalculator->SetParameter(parsedArgs); ngldCalculator->SetDirection(direction); rlCalculator->SetParameter(parsedArgs); rlCalculator->SetDirection(direction); std::ofstream output(parsedArgs["output"].ToString(), std::ios::app); mitk::Image::Pointer cImage = image; mitk::Image::Pointer cMask = mask; mitk::Image::Pointer cMaskNoNaN = maskNoNaN; while (imageToProcess) { if (sliceWise) { cImage = floatVector[currentSlice]; cMask = maskVector[currentSlice]; cMaskNoNaN = maskNoNaNVector[currentSlice]; imageToProcess = (floatVector.size()-1 > (currentSlice)) ? true : false ; } else { imageToProcess = false; } mitk::AbstractGlobalImageFeature::FeatureListType stats; firstOrderCalculator->CalculateFeaturesUsingParameters(cImage, cMask, cMaskNoNaN, stats); volCalculator->CalculateFeaturesUsingParameters(cImage, cMask, cMaskNoNaN, stats); coocCalculator->CalculateFeaturesUsingParameters(cImage, cMask, cMaskNoNaN, stats); cooc2Calculator->CalculateFeaturesUsingParameters(cImage, cMask, cMaskNoNaN, stats); ngldCalculator->CalculateFeaturesUsingParameters(cImage, cMask, cMaskNoNaN, stats); rlCalculator->CalculateFeaturesUsingParameters(cImage, cMask, cMaskNoNaN, stats); for (std::size_t i = 0; i < stats.size(); ++i) { std::cout << stats[i].first << " - " << stats[i].second << std::endl; } if (parsedArgs.count("header") && (currentSlice == 0)) { if (parsedArgs.count("description")) { output << "Description" << ";"; } if (sliceWise) { output << "SliceNumber" << ";"; } for (std::size_t i = 0; i < stats.size(); ++i) { output << stats[i].first << ";"; } output << "EndOfMeasurement;"; output << std::endl; } if (parsedArgs.count("description")) { output << parsedArgs["description"].ToString() << ";"; } if (sliceWise) { output << currentSlice << ";"; } for (std::size_t i = 0; i < stats.size(); ++i) { output << stats[i].second << ";"; } output << "EndOfMeasurement;"; output << std::endl; ++currentSlice; } if (sliceWise) { output << "EndOfFile" << std::endl; } output.close(); return 0; } #endif diff --git a/Modules/Classification/CLMiniApps/CMakeLists.txt b/Modules/Classification/CLMiniApps/CMakeLists.txt index 5c2ff09056..1f7ae5cae7 100644 --- a/Modules/Classification/CLMiniApps/CMakeLists.txt +++ b/Modules/Classification/CLMiniApps/CMakeLists.txt @@ -1,136 +1,137 @@ option(BUILD_ClassificationMiniApps "Build commandline tools for classification" OFF) if(BUILD_ClassificationMiniApps OR MITK_BUILD_ALL_APPS) include_directories( ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR} ) # list of miniapps # if an app requires additional dependencies # they are added after a "^^" and separated by "_" set( classificationminiapps RandomForestTraining^^MitkCLVigraRandomForest NativeHeadCTSegmentation^^MitkCLVigraRandomForest ManualSegmentationEvaluation^^MitkCLVigraRandomForest CLGlobalImageFeatures^^MitkCore_MitkCLUtilities CLMRNormalization^^MitkCore_MitkCLUtilities_MitkCLMRUtilities CLStaple^^MitkCore_MitkCLUtilities CLVoxelFeatures^^MitkCore_MitkCLUtilities CLDicom2Nrrd^^MitkCore CLPolyToNrrd^^MitkCore CLImageTypeConverter^^MitkCore CLResampleImageToReference^^MitkCore CLRandomSampling^^MitkCore_MitkCLUtilities CLRemoveEmptyVoxels^^MitkCore CLN4^^MitkCore CLMultiForestPrediction^^MitkDataCollection_MitkCLVigraRandomForest CLNrrdToPoly^^MitkCore + CL2Dto3DImage^^MitkCore # RandomForestPrediction^^MitkCLVigraRandomForest ) foreach(classificationminiapps ${classificationminiapps}) # extract mini app name and dependencies string(REPLACE "^^" "\\;" miniapp_info ${classificationminiapps}) set(miniapp_info_list ${miniapp_info}) list(GET miniapp_info_list 0 appname) list(GET miniapp_info_list 1 raw_dependencies) string(REPLACE "_" "\\;" dependencies "${raw_dependencies}") set(dependencies_list ${dependencies}) mitk_create_executable(${appname} DEPENDS MitkCore MitkCLCore MitkCommandLine ${dependencies_list} PACKAGE_DEPENDS ITK Qt5|Core Vigra CPP_FILES ${appname}.cpp ) # CPP_FILES ${appname}.cpp mitkCommandLineParser.cpp if(EXECUTABLE_IS_ENABLED) # On Linux, create a shell script to start a relocatable application if(UNIX AND NOT APPLE) install(PROGRAMS "${MITK_SOURCE_DIR}/CMake/RunInstalledApp.sh" DESTINATION "." RENAME ${EXECUTABLE_TARGET}.sh) endif() get_target_property(_is_bundle ${EXECUTABLE_TARGET} MACOSX_BUNDLE) if(APPLE) if(_is_bundle) set(_target_locations ${EXECUTABLE_TARGET}.app) set(${_target_locations}_qt_plugins_install_dir ${EXECUTABLE_TARGET}.app/Contents/MacOS) set(_bundle_dest_dir ${EXECUTABLE_TARGET}.app/Contents/MacOS) set(_qt_plugins_for_current_bundle ${EXECUTABLE_TARGET}.app/Contents/MacOS) set(_qt_conf_install_dirs ${EXECUTABLE_TARGET}.app/Contents/Resources) install(TARGETS ${EXECUTABLE_TARGET} BUNDLE DESTINATION . ) else() if(NOT MACOSX_BUNDLE_NAMES) set(_qt_conf_install_dirs bin) set(_target_locations bin/${EXECUTABLE_TARGET}) set(${_target_locations}_qt_plugins_install_dir bin) install(TARGETS ${EXECUTABLE_TARGET} RUNTIME DESTINATION bin) else() foreach(bundle_name ${MACOSX_BUNDLE_NAMES}) list(APPEND _qt_conf_install_dirs ${bundle_name}.app/Contents/Resources) set(_current_target_location ${bundle_name}.app/Contents/MacOS/${EXECUTABLE_TARGET}) list(APPEND _target_locations ${_current_target_location}) set(${_current_target_location}_qt_plugins_install_dir ${bundle_name}.app/Contents/MacOS) message( " set(${_current_target_location}_qt_plugins_install_dir ${bundle_name}.app/Contents/MacOS) ") install(TARGETS ${EXECUTABLE_TARGET} RUNTIME DESTINATION ${bundle_name}.app/Contents/MacOS/) endforeach() endif() endif() else() set(_target_locations bin/${EXECUTABLE_TARGET}${CMAKE_EXECUTABLE_SUFFIX}) set(${_target_locations}_qt_plugins_install_dir bin) set(_qt_conf_install_dirs bin) install(TARGETS ${EXECUTABLE_TARGET} RUNTIME DESTINATION bin) endif() endif() endforeach() # This mini app does not depend on mitkDiffusionImaging at all #mitk_create_executable(CLGlobalImageFeatures # DEPENDS MitkCore MitkCLUtilities # CPP_FILES CLGlobalImageFeatures.cpp mitkCommandLineParser.cpp #) mitk_create_executable(CLSimpleVoxelClassification DEPENDS MitkCore MitkCLCore MitkDataCollection MitkCLVigraRandomForest MitkCommandLine CPP_FILES CLSimpleVoxelClassification.cpp ) # This mini app does not depend on mitkDiffusionImaging at all mitk_create_executable(CLVoxelClassification DEPENDS MitkCore MitkCLCore MitkDataCollection MitkCLImportanceWeighting MitkCLVigraRandomForest CPP_FILES CLVoxelClassification.cpp ) #mitk_create_executable(CLBrainMask # DEPENDS MitkCore MitkCLUtilities # CPP_FILES CLBrainMask.cpp mitkCommandLineParser.cpp #) #mitk_create_executable(CLImageConverter # DEPENDS MitkCore # CPP_FILES CLImageConverter.cpp mitkCommandLineParser.cpp #) #mitk_create_executable(CLSurWeighting # DEPENDS MitkCore MitkCLUtilities MitkDataCollection #MitkCLImportanceWeighting # CPP_FILES CLSurWeighting.cpp mitkCommandLineParser.cpp #) #mitk_create_executable(CLImageCropper # DEPENDS MitkCore MitkCLUtilities MitkAlgorithmsExt # CPP_FILES CLImageCropper.cpp mitkCommandLineParser.cpp #) # On Linux, create a shell script to start a relocatable application if(UNIX AND NOT APPLE) install(PROGRAMS "${MITK_SOURCE_DIR}/CMake/RunInstalledApp.sh" DESTINATION "." RENAME ${EXECUTABLE_TARGET}.sh) endif() if(EXECUTABLE_IS_ENABLED) MITK_INSTALL_TARGETS(EXECUTABLES ${EXECUTABLE_TARGET}) endif() endif() diff --git a/Modules/Classification/CLUtilities/include/mitkGIFCooccurenceMatrix2.h b/Modules/Classification/CLUtilities/include/mitkGIFCooccurenceMatrix2.h index 322631b709..2665167adc 100644 --- a/Modules/Classification/CLUtilities/include/mitkGIFCooccurenceMatrix2.h +++ b/Modules/Classification/CLUtilities/include/mitkGIFCooccurenceMatrix2.h @@ -1,147 +1,147 @@ #ifndef mitkGIFCooccurenceMatrix2_h #define mitkGIFCooccurenceMatrix2_h #include #include #include -#include +#include namespace mitk { struct CoocurenceMatrixHolder { public: CoocurenceMatrixHolder(double min, double max, int number); int IntensityToIndex(double intensity); double IndexToMinIntensity(int index); double IndexToMeanIntensity(int index); double IndexToMaxIntensity(int index); double m_MinimumRange; double m_MaximumRange; double m_Stepsize; int m_NumberOfBins; Eigen::MatrixXd m_Matrix; }; struct CoocurenceMatrixFeatures { CoocurenceMatrixFeatures(): JointMaximum(0), JointAverage(0), JointVariance(0), JointEntropy(0), RowMaximum(0), RowAverage(0), RowVariance(0), RowEntropy(0), FirstRowColumnEntropy(0), SecondRowColumnEntropy(0), DifferenceAverage(0), DifferenceVariance(0), DifferenceEntropy(0), SumAverage(0), SumVariance(0), SumEntropy(0), AngularSecondMoment(0), Contrast(0), Dissimilarity(0), InverseDifference(0), InverseDifferenceNormalised(0), InverseDifferenceMoment(0), InverseDifferenceMomentNormalised(0), InverseVariance(0), Correlation(0), Autocorrelation(0), ClusterTendency(0), ClusterShade(0), ClusterProminence(0), FirstMeasureOfInformationCorrelation(0), SecondMeasureOfInformationCorrelation(0) { } public: double JointMaximum; double JointAverage; double JointVariance; double JointEntropy; double RowMaximum; double RowAverage; double RowVariance; double RowEntropy; double FirstRowColumnEntropy; double SecondRowColumnEntropy; double DifferenceAverage; double DifferenceVariance; double DifferenceEntropy; double SumAverage; double SumVariance; double SumEntropy; double AngularSecondMoment; double Contrast; double Dissimilarity; double InverseDifference; double InverseDifferenceNormalised; double InverseDifferenceMoment; double InverseDifferenceMomentNormalised; double InverseVariance; double Correlation; double Autocorrelation; double ClusterTendency; double ClusterShade; double ClusterProminence; double FirstMeasureOfInformationCorrelation; double SecondMeasureOfInformationCorrelation; }; class MITKCLUTILITIES_EXPORT GIFCooccurenceMatrix2 : public AbstractGlobalImageFeature { public: mitkClassMacro(GIFCooccurenceMatrix2, AbstractGlobalImageFeature); itkFactorylessNewMacro(Self) itkCloneMacro(Self) GIFCooccurenceMatrix2(); /** * \brief Calculates the Cooccurence-Matrix based features for this class. */ virtual FeatureListType CalculateFeatures(const Image::Pointer & image, const Image::Pointer &feature) override; /** * \brief Returns a list of the names of all features that are calculated from this class */ virtual FeatureNameListType GetFeatureNames() override; virtual void CalculateFeaturesUsingParameters(const Image::Pointer & feature, const Image::Pointer &mask, const Image::Pointer &maskNoNAN, FeatureListType &featureList); virtual void AddArguments(mitkCommandLineParser &parser); itkGetConstMacro(Range,double); itkSetMacro(Range, double); itkGetConstMacro(Bins, int); itkSetMacro(Bins, int); struct GIFCooccurenceMatrix2Configuration { double range; unsigned int direction; double MinimumIntensity; bool UseMinimumIntensity; double MaximumIntensity; bool UseMaximumIntensity; int Bins; }; private: double m_Range; int m_Bins; }; } #endif //mitkGIFCooccurenceMatrix2_h diff --git a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFCooccurenceMatrix2.cpp b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFCooccurenceMatrix2.cpp index 59e2cc5d05..7752f3e433 100644 --- a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFCooccurenceMatrix2.cpp +++ b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFCooccurenceMatrix2.cpp @@ -1,606 +1,604 @@ #include // MITK #include #include #include // ITK #include #include #include #include // STL #include static void MatrixFeaturesTo(mitk::CoocurenceMatrixFeatures features, std::string prefix, mitk::GIFCooccurenceMatrix2::FeatureListType &featureList); static void CalculateMeanAndStdDevFeatures(std::vector featureList, mitk::CoocurenceMatrixFeatures &meanFeature, mitk::CoocurenceMatrixFeatures &stdFeature); static void NormalizeMatrixFeature(mitk::CoocurenceMatrixFeatures &features, std::size_t number); mitk::CoocurenceMatrixHolder::CoocurenceMatrixHolder(double min, double max, int number) : m_MinimumRange(min), m_MaximumRange(max), m_NumberOfBins(number) { m_Matrix.resize(number, number); m_Matrix.fill(0); m_Stepsize = (max - min) / (number); } int mitk::CoocurenceMatrixHolder::IntensityToIndex(double intensity) { return std::floor((intensity - m_MinimumRange) / m_Stepsize); } double mitk::CoocurenceMatrixHolder::IndexToMinIntensity(int index) { return m_MinimumRange + index * m_Stepsize; } double mitk::CoocurenceMatrixHolder::IndexToMeanIntensity(int index) { return m_MinimumRange + (index+0.5) * m_Stepsize; } double mitk::CoocurenceMatrixHolder::IndexToMaxIntensity(int index) { return m_MinimumRange + (index + 1) * m_Stepsize; } template void CalculateCoOcMatrix(itk::Image* itkImage, itk::Image* mask, itk::Offset offset, int range, mitk::CoocurenceMatrixHolder &holder) { typedef itk::Image ImageType; typedef itk::Image MaskImageType; typedef itk::ShapedNeighborhoodIterator ShapeIterType; typedef itk::ShapedNeighborhoodIterator ShapeMaskIterType; typedef itk::ImageRegionConstIterator ConstIterType; typedef itk::ImageRegionConstIterator ConstMaskIterType; itk::Size radius; radius.Fill(range+1); ShapeIterType imageOffsetIter(radius, itkImage, itkImage->GetLargestPossibleRegion()); ShapeMaskIterType maskOffsetIter(radius, mask, mask->GetLargestPossibleRegion()); imageOffsetIter.ActivateOffset(offset); maskOffsetIter.ActivateOffset(offset); ConstIterType imageIter(itkImage, itkImage->GetLargestPossibleRegion()); ConstMaskIterType maskIter(mask, mask->GetLargestPossibleRegion()); // iterator.GetIndex() + ci.GetNeighborhoodOffset() auto region = mask->GetLargestPossibleRegion(); while (!maskIter.IsAtEnd()) { auto ciMask = maskOffsetIter.Begin(); auto ciValue = imageOffsetIter.Begin(); if (maskIter.Value() > 0 && ciMask.Get() > 0 && imageIter.Get() == imageIter.Get() && ciValue.Get() == ciValue.Get() && region.IsInside(maskOffsetIter.GetIndex() + ciMask.GetNeighborhoodOffset())) { int i = holder.IntensityToIndex(imageIter.Get()); int j = holder.IntensityToIndex(ciValue.Get()); holder.m_Matrix(i, j) += 1; holder.m_Matrix(j, i) += 1; } ++imageOffsetIter; ++maskOffsetIter; ++imageIter; ++maskIter; } } void CalculateFeatures( mitk::CoocurenceMatrixHolder &holder, mitk::CoocurenceMatrixFeatures & results ) { auto pijMatrix = holder.m_Matrix; auto piMatrix = holder.m_Matrix; auto pjMatrix = holder.m_Matrix; double Ng = holder.m_NumberOfBins; int NgSize = holder.m_NumberOfBins; pijMatrix /= pijMatrix.sum(); piMatrix.rowwise().normalize(); pjMatrix.colwise().normalize(); for (int i = 0; i < holder.m_NumberOfBins; ++i) for (int j = 0; j < holder.m_NumberOfBins; ++j) { if (pijMatrix(i, j) != pijMatrix(i, j)) pijMatrix(i, j) = 0; if (piMatrix(i, j) != piMatrix(i, j)) piMatrix(i, j) = 0; if (pjMatrix(i, j) != pjMatrix(i, j)) pjMatrix(i, j) = 0; } Eigen::VectorXd piVector = pijMatrix.colwise().sum(); Eigen::VectorXd pjVector = pijMatrix.rowwise().sum(); double sigmai = 0;; for (int i = 0; i < holder.m_NumberOfBins; ++i) { double iInt = holder.IndexToMeanIntensity(i); results.RowAverage += iInt * piVector(i); if (piVector(i) > 0) { results.RowEntropy -= piVector(i) * std::log(piVector(i)) / std::log(2); } } for (int i = 0; i < holder.m_NumberOfBins; ++i) { double iInt = holder.IndexToMeanIntensity(i); results.RowVariance += (iInt - results.RowAverage)*(iInt - results.RowAverage) * piVector(i); } results.RowMaximum = piVector.maxCoeff(); sigmai = std::sqrt(results.RowVariance); Eigen::VectorXd pimj(NgSize); pimj.fill(0); Eigen::VectorXd pipj(2*NgSize); pipj.fill(0); results.JointMaximum += pijMatrix.maxCoeff(); for (int i = 0; i < holder.m_NumberOfBins; ++i) { for (int j = 0; j < holder.m_NumberOfBins; ++j) { double iInt = holder.IndexToMeanIntensity(i); double jInt = holder.IndexToMeanIntensity(j); double pij = pijMatrix(i, j); int deltaK = (i - j)>0?(i-j) : (j-i); pimj(deltaK) += pij; pipj(i + j) += pij; results.JointAverage += iInt * pij; if (pij > 0) { results.JointEntropy -= pij * std::log(pij) / std::log(2); results.FirstRowColumnEntropy -= pij * std::log(piVector(i)*pjVector(j)) / std::log(2); } if (piVector(i) > 0 && pjVector(j) > 0 ) { results.SecondRowColumnEntropy -= piVector(i)*pjVector(j) * std::log(piVector(i)*pjVector(j)) / std::log(2); } results.AngularSecondMoment += pij*pij; results.Contrast += (iInt - jInt)* (iInt - jInt) * pij; results.Dissimilarity += std::abs(iInt - jInt) * pij; results.InverseDifference += pij / (1 + (std::abs(iInt - jInt))); results.InverseDifferenceNormalised += pij / (1 + (std::abs(iInt - jInt) / Ng)); results.InverseDifferenceMoment += pij / (1 + (iInt - jInt)*(iInt - jInt)); results.InverseDifferenceMomentNormalised += pij / (1 + (iInt - jInt)*(iInt - jInt)/Ng/Ng); results.Autocorrelation += iInt*jInt * pij; double cluster = (iInt + jInt - 2 * results.RowAverage); results.ClusterTendency += cluster*cluster * pij; results.ClusterShade += cluster*cluster*cluster * pij; results.ClusterProminence += cluster*cluster*cluster*cluster * pij; if (iInt != jInt) { results.InverseVariance += pij / (iInt - jInt) / (iInt - jInt); } } } results.Correlation = 1 / sigmai / sigmai * (-results.RowAverage*results.RowAverage+ results.Autocorrelation); results.FirstMeasureOfInformationCorrelation = (results.JointEntropy - results.FirstRowColumnEntropy) / results.RowEntropy; if (results.JointEntropy < results.SecondRowColumnEntropy) { results.SecondMeasureOfInformationCorrelation = sqrt(1 - exp(-2 * (results.SecondRowColumnEntropy - results.JointEntropy))); } else { results.SecondMeasureOfInformationCorrelation = 0; } for (int i = 0; i < holder.m_NumberOfBins; ++i) { for (int j = 0; j < holder.m_NumberOfBins; ++j) { double iInt = holder.IndexToMeanIntensity(i); //double jInt = holder.IndexToMeanIntensity(j); double pij = pijMatrix(i, j); results.JointVariance += (iInt - results.JointAverage)* (iInt - results.JointAverage)*pij; } } for (int k = 0; k < NgSize; ++k) { results.DifferenceAverage += k* pimj(k); if (pimj(k) > 0) { results.DifferenceEntropy -= pimj(k) * log(pimj(k)) / std::log(2); } } for (int k = 0; k < NgSize; ++k) { results.DifferenceVariance += (results.DifferenceAverage-k)* (results.DifferenceAverage-k)*pimj(k); } for (int k = 0; k <2* NgSize ; ++k) { results.SumAverage += (2+k)* pipj(k); if (pipj(k) > 0) { results.SumEntropy -= pipj(k) * log(pipj(k)) / std::log(2); } } for (int k = 0; k < 2*NgSize; ++k) { results.SumVariance += (2+k - results.SumAverage)* (2+k - results.SumAverage)*pipj(k); } //MITK_INFO << std::endl << holder.m_Matrix; //MITK_INFO << std::endl << pijMatrix; //MITK_INFO << std::endl << piMatrix; //MITK_INFO << std::endl << pjMatrix; //for (int i = 0; i < holder.m_NumberOfBins; ++i) //{ // MITK_INFO << "Bin " << i << " Min: " << holder.IndexToMinIntensity(i) << " Max: " << holder.IndexToMaxIntensity(i); //} //MITK_INFO << pimj; //MITK_INFO << pipj; } template void CalculateCoocurenceFeatures(itk::Image* itkImage, mitk::Image::Pointer mask, mitk::GIFCooccurenceMatrix2::FeatureListType & featureList, mitk::GIFCooccurenceMatrix2::GIFCooccurenceMatrix2Configuration config) { typedef itk::Image ImageType; typedef itk::Image MaskType; typedef itk::MinimumMaximumImageCalculator MinMaxComputerType; typedef itk::Neighborhood NeighborhoodType; typedef itk::Offset OffsetType; /////////////////////////////////////////////////////////////////////////////////////////////// typename MinMaxComputerType::Pointer minMaxComputer = MinMaxComputerType::New(); minMaxComputer->SetImage(itkImage); minMaxComputer->Compute(); double rangeMin = -0.5 + minMaxComputer->GetMinimum(); double rangeMax = 0.5 + minMaxComputer->GetMaximum(); if (config.UseMinimumIntensity) rangeMin = config.MinimumIntensity; if (config.UseMaximumIntensity) rangeMax = config.MaximumIntensity; // Define Range int numberOfBins = config.Bins; typename MaskType::Pointer maskImage = MaskType::New(); mitk::CastToItkImage(mask, maskImage); - //Find possible directions std::vector < itk::Offset > offsetVector; NeighborhoodType hood; hood.SetRadius(1); unsigned int centerIndex = hood.GetCenterNeighborhoodIndex(); OffsetType offset; for (unsigned int d = 0; d < centerIndex; d++) { offset = hood.GetOffset(d); bool useOffset = true; for (int i = 0; i < VImageDimension; ++i) { offset[i] *= config.range; if (config.direction == i + 2 && offset[i] != 0) { useOffset = false; } } if (useOffset) { offsetVector.push_back(offset); } } if (config.direction == 1) { offsetVector.clear(); offset[0] = 0; offset[1] = 0; offset[2] = 1; } std::vector resultVector; mitk::CoocurenceMatrixHolder holderOverall(rangeMin, rangeMax, numberOfBins); mitk::CoocurenceMatrixFeatures overallFeature; for (int i = 0; i < offsetVector.size(); ++i) { - if (config.direction) + if (config.direction > 1) { - if ((config.direction > 1) && offsetVector[i][config.direction - 2] != 0) + if (offsetVector[i][config.direction - 2] != 0) { continue; } } offset = offsetVector[i]; - MITK_INFO << offset; mitk::CoocurenceMatrixHolder holder(rangeMin, rangeMax, numberOfBins); mitk::CoocurenceMatrixFeatures coocResults; CalculateCoOcMatrix(itkImage, maskImage, offset, config.range, holder); holderOverall.m_Matrix += holder.m_Matrix; CalculateFeatures(holder, coocResults); resultVector.push_back(coocResults); } CalculateFeatures(holderOverall, overallFeature); //NormalizeMatrixFeature(overallFeature, offsetVector.size()); mitk::CoocurenceMatrixFeatures featureMean; mitk::CoocurenceMatrixFeatures featureStd; CalculateMeanAndStdDevFeatures(resultVector, featureMean, featureStd); std::ostringstream ss; ss << config.range; std::string strRange = ss.str(); MatrixFeaturesTo(overallFeature, "co-occ. 2 (" + strRange + ") overall", featureList); MatrixFeaturesTo(featureMean, "co-occ. 2 (" + strRange + ") mean", featureList); MatrixFeaturesTo(featureStd, "co-occ. 2 (" + strRange + ") std.dev.", featureList); } static void MatrixFeaturesTo(mitk::CoocurenceMatrixFeatures features, std::string prefix, mitk::GIFCooccurenceMatrix2::FeatureListType &featureList) { featureList.push_back(std::make_pair(prefix + " Joint Maximum", features.JointMaximum)); featureList.push_back(std::make_pair(prefix + " Joint Average", features.JointAverage)); featureList.push_back(std::make_pair(prefix + " Joint Variance", features.JointVariance)); featureList.push_back(std::make_pair(prefix + " Joint Entropy", features.JointEntropy)); featureList.push_back(std::make_pair(prefix + " Row Maximum", features.RowMaximum)); featureList.push_back(std::make_pair(prefix + " Row Average", features.RowAverage)); featureList.push_back(std::make_pair(prefix + " Row Variance", features.RowVariance)); featureList.push_back(std::make_pair(prefix + " Row Entropy", features.RowEntropy)); featureList.push_back(std::make_pair(prefix + " First Row-Column Entropy", features.FirstRowColumnEntropy)); featureList.push_back(std::make_pair(prefix + " Second Row-Column Entropy", features.SecondRowColumnEntropy)); featureList.push_back(std::make_pair(prefix + " Difference Average", features.DifferenceAverage)); featureList.push_back(std::make_pair(prefix + " Difference Variance", features.DifferenceVariance)); featureList.push_back(std::make_pair(prefix + " Difference Entropy", features.DifferenceEntropy)); featureList.push_back(std::make_pair(prefix + " Sum Average", features.SumAverage)); featureList.push_back(std::make_pair(prefix + " Sum Variance", features.SumVariance)); featureList.push_back(std::make_pair(prefix + " Sum Entropy", features.SumEntropy)); featureList.push_back(std::make_pair(prefix + " Angular Second Moment", features.AngularSecondMoment)); featureList.push_back(std::make_pair(prefix + " Contrast", features.Contrast)); featureList.push_back(std::make_pair(prefix + " Dissimilarity", features.Dissimilarity)); featureList.push_back(std::make_pair(prefix + " Inverse Difference", features.InverseDifference)); featureList.push_back(std::make_pair(prefix + " Inverse Difference Normalized", features.InverseDifferenceNormalised)); featureList.push_back(std::make_pair(prefix + " Inverse Difference Moment", features.InverseDifferenceMoment)); featureList.push_back(std::make_pair(prefix + " Inverse Difference Moment Normalized", features.InverseDifferenceMomentNormalised)); featureList.push_back(std::make_pair(prefix + " Inverse Variance", features.InverseVariance)); featureList.push_back(std::make_pair(prefix + " Correlation", features.Correlation)); featureList.push_back(std::make_pair(prefix + " Autocorrleation", features.Autocorrelation)); featureList.push_back(std::make_pair(prefix + " Cluster Tendency", features.ClusterTendency)); featureList.push_back(std::make_pair(prefix + " Cluster Shade", features.ClusterShade)); featureList.push_back(std::make_pair(prefix + " Cluster Prominence", features.ClusterProminence)); featureList.push_back(std::make_pair(prefix + " First Measure of Information Correlation", features.FirstMeasureOfInformationCorrelation)); featureList.push_back(std::make_pair(prefix + " Second Measure of Information Correlation", features.SecondMeasureOfInformationCorrelation)); } static void CalculateMeanAndStdDevFeatures(std::vector featureList, mitk::CoocurenceMatrixFeatures &meanFeature, mitk::CoocurenceMatrixFeatures &stdFeature) { #define ADDFEATURE(a) meanFeature.a += featureList[i].a;stdFeature.a += featureList[i].a*featureList[i].a #define CALCVARIANCE(a) stdFeature.a =sqrt(stdFeature.a - meanFeature.a*meanFeature.a) for (std::size_t i = 0; i < featureList.size(); ++i) { ADDFEATURE(JointMaximum); ADDFEATURE(JointAverage); ADDFEATURE(JointVariance); ADDFEATURE(JointEntropy); ADDFEATURE(RowMaximum); ADDFEATURE(RowAverage); ADDFEATURE(RowVariance); ADDFEATURE(RowEntropy); ADDFEATURE(FirstRowColumnEntropy); ADDFEATURE(SecondRowColumnEntropy); ADDFEATURE(DifferenceAverage); ADDFEATURE(DifferenceVariance); ADDFEATURE(DifferenceEntropy); ADDFEATURE(SumAverage); ADDFEATURE(SumVariance); ADDFEATURE(SumEntropy); ADDFEATURE(AngularSecondMoment); ADDFEATURE(Contrast); ADDFEATURE(Dissimilarity); ADDFEATURE(InverseDifference); ADDFEATURE(InverseDifferenceNormalised); ADDFEATURE(InverseDifferenceMoment); ADDFEATURE(InverseDifferenceMomentNormalised); ADDFEATURE(InverseVariance); ADDFEATURE(Correlation); ADDFEATURE(Autocorrelation); ADDFEATURE(ClusterShade); ADDFEATURE(ClusterTendency); ADDFEATURE(ClusterProminence); ADDFEATURE(FirstMeasureOfInformationCorrelation); ADDFEATURE(SecondMeasureOfInformationCorrelation); } NormalizeMatrixFeature(meanFeature, featureList.size()); NormalizeMatrixFeature(stdFeature, featureList.size()); CALCVARIANCE(JointMaximum); CALCVARIANCE(JointAverage); CALCVARIANCE(JointVariance); CALCVARIANCE(JointEntropy); CALCVARIANCE(RowMaximum); CALCVARIANCE(RowAverage); CALCVARIANCE(RowVariance); CALCVARIANCE(RowEntropy); CALCVARIANCE(FirstRowColumnEntropy); CALCVARIANCE(SecondRowColumnEntropy); CALCVARIANCE(DifferenceAverage); CALCVARIANCE(DifferenceVariance); CALCVARIANCE(DifferenceEntropy); CALCVARIANCE(SumAverage); CALCVARIANCE(SumVariance); CALCVARIANCE(SumEntropy); CALCVARIANCE(AngularSecondMoment); CALCVARIANCE(Contrast); CALCVARIANCE(Dissimilarity); CALCVARIANCE(InverseDifference); CALCVARIANCE(InverseDifferenceNormalised); CALCVARIANCE(InverseDifferenceMoment); CALCVARIANCE(InverseDifferenceMomentNormalised); CALCVARIANCE(InverseVariance); CALCVARIANCE(Correlation); CALCVARIANCE(Autocorrelation); CALCVARIANCE(ClusterShade); CALCVARIANCE(ClusterTendency); CALCVARIANCE(ClusterProminence); CALCVARIANCE(FirstMeasureOfInformationCorrelation); CALCVARIANCE(SecondMeasureOfInformationCorrelation); #undef ADDFEATURE #undef CALCVARIANCE } static void NormalizeMatrixFeature(mitk::CoocurenceMatrixFeatures &features, std::size_t number) { features.JointMaximum = features.JointMaximum / number; features.JointAverage = features.JointAverage / number; features.JointVariance = features.JointVariance / number; features.JointEntropy = features.JointEntropy / number; features.RowMaximum = features.RowMaximum / number; features.RowAverage = features.RowAverage / number; features.RowVariance = features.RowVariance / number; features.RowEntropy = features.RowEntropy / number; features.FirstRowColumnEntropy = features.FirstRowColumnEntropy / number; features.SecondRowColumnEntropy = features.SecondRowColumnEntropy / number; features.DifferenceAverage = features.DifferenceAverage / number; features.DifferenceVariance = features.DifferenceVariance / number; features.DifferenceEntropy = features.DifferenceEntropy / number; features.SumAverage = features.SumAverage / number; features.SumVariance = features.SumVariance / number; features.SumEntropy = features.SumEntropy / number; features.AngularSecondMoment = features.AngularSecondMoment / number; features.Contrast = features.Contrast / number; features.Dissimilarity = features.Dissimilarity / number; features.InverseDifference = features.InverseDifference / number; features.InverseDifferenceNormalised = features.InverseDifferenceNormalised / number; features.InverseDifferenceMoment = features.InverseDifferenceMoment / number; features.InverseDifferenceMomentNormalised = features.InverseDifferenceMomentNormalised / number; features.InverseVariance = features.InverseVariance / number; features.Correlation = features.Correlation / number; features.Autocorrelation = features.Autocorrelation / number; features.ClusterShade = features.ClusterShade / number; features.ClusterTendency = features.ClusterTendency / number; features.ClusterProminence = features.ClusterProminence / number; features.FirstMeasureOfInformationCorrelation = features.FirstMeasureOfInformationCorrelation / number; features.SecondMeasureOfInformationCorrelation = features.SecondMeasureOfInformationCorrelation / number; } mitk::GIFCooccurenceMatrix2::GIFCooccurenceMatrix2(): m_Range(1.0), m_Bins(128) { SetShortName("cooc2"); SetLongName("cooccurence2"); } mitk::GIFCooccurenceMatrix2::FeatureListType mitk::GIFCooccurenceMatrix2::CalculateFeatures(const Image::Pointer & image, const Image::Pointer &mask) { FeatureListType featureList; GIFCooccurenceMatrix2Configuration config; config.direction = GetDirection(); config.range = m_Range; config.MinimumIntensity = GetMinimumIntensity(); config.MaximumIntensity = GetMaximumIntensity(); config.UseMinimumIntensity = GetUseMinimumIntensity(); config.UseMaximumIntensity = GetUseMaximumIntensity(); config.Bins = GetBins(); AccessByItk_3(image, CalculateCoocurenceFeatures, mask, featureList,config); return featureList; } mitk::GIFCooccurenceMatrix2::FeatureNameListType mitk::GIFCooccurenceMatrix2::GetFeatureNames() { FeatureNameListType featureList; return featureList; } void mitk::GIFCooccurenceMatrix2::AddArguments(mitkCommandLineParser &parser) { std::string name = GetOptionPrefix(); parser.addArgument(GetLongName(), name, mitkCommandLineParser::String, "Use Co-occurence matrix", "calculates Co-occurence based features (new implementation)", us::Any()); parser.addArgument(name+"::range", name+"::range", mitkCommandLineParser::String, "Cooc 2 Range", "Define the range that is used (Semicolon-separated)", us::Any()); parser.addArgument(name + "::bins", name + "::bins", mitkCommandLineParser::String, "Cooc 2 Number of Bins", "Define the number of bins that is used ", us::Any()); } void mitk::GIFCooccurenceMatrix2::CalculateFeaturesUsingParameters(const Image::Pointer & feature, const Image::Pointer &, const Image::Pointer &maskNoNAN, FeatureListType &featureList) { auto parsedArgs = GetParameter(); std::string name = GetOptionPrefix(); if (parsedArgs.count(GetLongName())) { std::vector ranges; if (parsedArgs.count(name + "::range")) { ranges = SplitDouble(parsedArgs[name + "::range"].ToString(), ';'); } else { ranges.push_back(1); } if (parsedArgs.count(name + "::bins")) { auto bins = SplitDouble(parsedArgs[name + "::bins"].ToString(), ';')[0]; this->SetBins(bins); } for (std::size_t i = 0; i < ranges.size(); ++i) { MITK_INFO << "Start calculating coocurence with range " << ranges[i] << "...."; this->SetRange(ranges[i]); auto localResults = this->CalculateFeatures(feature, maskNoNAN); featureList.insert(featureList.end(), localResults.begin(), localResults.end()); MITK_INFO << "Finished calculating coocurence with range " << ranges[i] << "...."; } } } diff --git a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFNeighbouringGreyLevelDependenceFeatures.cpp b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFNeighbouringGreyLevelDependenceFeatures.cpp index 1c26fe34b6..464734c7d3 100644 --- a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFNeighbouringGreyLevelDependenceFeatures.cpp +++ b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFNeighbouringGreyLevelDependenceFeatures.cpp @@ -1,406 +1,404 @@ #include // MITK #include #include #include // ITK #include #include #include #include // STL #include static void MatrixFeaturesTo(mitk::NGLDMMatrixFeatures features, std::string prefix, mitk::GIFNeighbouringGreyLevelDependenceFeature::FeatureListType &featureList); mitk::NGLDMMatrixHolder::NGLDMMatrixHolder(double min, double max, int number, int depenence) : m_MinimumRange(min), m_MaximumRange(max), m_Stepsize(0), m_NumberOfDependences(depenence), m_NumberOfBins(number), m_NeighbourhoodSize(1), m_NumberOfNeighbourVoxels(0), m_NumberOfDependenceNeighbourVoxels(0), m_NumberOfNeighbourhoods(0), m_NumberOfCompleteNeighbourhoods(0) { m_Matrix.resize(number, depenence); m_Matrix.fill(0); m_Stepsize = (max - min) / (number); } int mitk::NGLDMMatrixHolder::IntensityToIndex(double intensity) { return std::floor((intensity - m_MinimumRange) / m_Stepsize); } double mitk::NGLDMMatrixHolder::IndexToMinIntensity(int index) { return m_MinimumRange + index * m_Stepsize; } double mitk::NGLDMMatrixHolder::IndexToMeanIntensity(int index) { return m_MinimumRange + (index+0.5) * m_Stepsize; } double mitk::NGLDMMatrixHolder::IndexToMaxIntensity(int index) { return m_MinimumRange + (index + 1) * m_Stepsize; } template void CalculateNGLDMMatrix(itk::Image* itkImage, itk::Image* mask, int alpha, int range, int direction, mitk::NGLDMMatrixHolder &holder) { typedef itk::Image ImageType; typedef itk::Image MaskImageType; typedef itk::NeighborhoodIterator ShapeIterType; typedef itk::NeighborhoodIterator ShapeMaskIterType; holder.m_NumberOfCompleteNeighbourhoods = 0; holder.m_NumberOfNeighbourhoods = 0; holder.m_NumberOfNeighbourVoxels = 0; holder.m_NumberOfDependenceNeighbourVoxels = 0; - MITK_INFO << direction; - itk::Size radius; radius.Fill(range); if ((direction > 1) && (direction +2 GetLargestPossibleRegion()); ShapeMaskIterType maskIter(radius, mask, mask->GetLargestPossibleRegion()); auto region = mask->GetLargestPossibleRegion(); auto center = imageIter.Size() / 2; auto iterSize = imageIter.Size(); holder.m_NeighbourhoodSize = iterSize-1; while (!maskIter.IsAtEnd()) { int sameValues = 0; bool completeNeighbourhood = true; int i = holder.IntensityToIndex(imageIter.GetCenterPixel()); if (imageIter.GetCenterPixel() != imageIter.GetCenterPixel()) { ++imageIter; ++maskIter; continue; } for (int position = 0; position < iterSize; ++position) { if ((position == center) || ( ! region.IsInside(maskIter.GetIndex(position)))) { completeNeighbourhood = false; continue; } bool isInBounds; auto jIntensity = imageIter.GetPixel(position, isInBounds); auto jMask = maskIter.GetPixel(position, isInBounds); if (jMask < 0 || (jIntensity != jIntensity) || ( ! isInBounds)) { completeNeighbourhood = false; continue; } int j = holder.IntensityToIndex(jIntensity); holder.m_NumberOfNeighbourVoxels += 1; if (std::abs(i - j) <= alpha) { holder.m_NumberOfDependenceNeighbourVoxels += 1; ++sameValues; } } holder.m_Matrix(i, sameValues) += 1; holder.m_NumberOfNeighbourhoods += 1; if (completeNeighbourhood) { holder.m_NumberOfCompleteNeighbourhoods += 1; } ++imageIter; ++maskIter; } } void LocalCalculateFeatures( mitk::NGLDMMatrixHolder &holder, mitk::NGLDMMatrixFeatures & results ) { auto sijMatrix = holder.m_Matrix; auto piMatrix = holder.m_Matrix; auto pjMatrix = holder.m_Matrix; // double Ng = holder.m_NumberOfBins; // int NgSize = holder.m_NumberOfBins; double Ns = sijMatrix.sum(); piMatrix.rowwise().normalize(); pjMatrix.colwise().normalize(); for (int i = 0; i < holder.m_NumberOfBins; ++i) { double sj = 0; for (int j = 0; j < holder.m_NumberOfDependences; ++j) { double iInt = holder.IndexToMeanIntensity(i); double sij = sijMatrix(i, j); double k = j+1; double pij = sij / Ns; results.LowDependenceEmphasis += sij / k / k; results.HighDependenceEmphasis += sij * k*k; if (iInt != 0) { results.LowGreyLevelCountEmphasis += sij / iInt / iInt; } results.HighGreyLevelCountEmphasis += sij * iInt*iInt; if (iInt != 0) { results.LowDependenceLowGreyLevelEmphasis += sij / k / k / iInt / iInt; } results.LowDependenceHighGreyLevelEmphasis += sij * iInt*iInt / k / k; if (iInt != 0) { results.HighDependenceLowGreyLevelEmphasis += sij *k * k / iInt / iInt; } results.HighDependenceHighGreyLevelEmphasis += sij * k*k*iInt*iInt; results.MeanGreyLevelCount += iInt * pij; results.MeanDependenceCount += k * pij; if (pij > 0) { results.DependenceCountEntropy -= pij * std::log(pij) / std::log(2); } sj += sij; } results.GreyLevelNonUniformity += sj*sj; results.GreyLevelNonUniformityNormalised += sj*sj; } for (int j = 0; j < holder.m_NumberOfDependences; ++j) { double si = 0; for (int i = 0; i < holder.m_NumberOfBins; ++i) { double sij = sijMatrix(i, j); si += sij; } results.DependenceCountNonUniformity += si*si; results.DependenceCountNonUniformityNormalised += si*si; } for (int i = 0; i < holder.m_NumberOfBins; ++i) { for (int j = 0; j < holder.m_NumberOfDependences; ++j) { double iInt = holder.IndexToMeanIntensity(i); double sij = sijMatrix(i, j); double k = j + 1; double pij = sij / Ns; results.GreyLevelVariance += (iInt - results.MeanGreyLevelCount)* (iInt - results.MeanGreyLevelCount) * pij; results.DependenceCountVariance += (k - results.MeanDependenceCount)* (k - results.MeanDependenceCount) * pij; } } results.LowDependenceEmphasis /= Ns; results.HighDependenceEmphasis /= Ns; results.LowGreyLevelCountEmphasis /= Ns; results.HighGreyLevelCountEmphasis /= Ns; results.LowDependenceLowGreyLevelEmphasis /= Ns; results.LowDependenceHighGreyLevelEmphasis /= Ns; results.HighDependenceLowGreyLevelEmphasis /= Ns; results.HighDependenceHighGreyLevelEmphasis /= Ns; results.GreyLevelNonUniformity /= Ns; results.GreyLevelNonUniformityNormalised /= (Ns*Ns); results.DependenceCountNonUniformity /= Ns; results.DependenceCountNonUniformityNormalised /= (Ns*Ns); results.DependenceCountPercentage = 1; results.ExpectedNeighbourhoodSize = holder.m_NeighbourhoodSize; results.AverageNeighbourhoodSize = holder.m_NumberOfNeighbourVoxels / (1.0 * holder.m_NumberOfNeighbourhoods); results.AverageIncompleteNeighbourhoodSize = (holder.m_NumberOfNeighbourVoxels - holder.m_NumberOfCompleteNeighbourhoods* holder.m_NeighbourhoodSize) / (1.0 * (holder.m_NumberOfNeighbourhoods - holder.m_NumberOfCompleteNeighbourhoods)); results.PercentageOfCompleteNeighbourhoods = holder.m_NumberOfCompleteNeighbourhoods / (1.0 * holder.m_NumberOfNeighbourhoods); results.PercentageOfDependenceNeighbours = holder.m_NumberOfDependenceNeighbourVoxels / (1.0 * holder.m_NumberOfNeighbourVoxels); } template void CalculateCoocurenceFeatures(itk::Image* itkImage, mitk::Image::Pointer mask, mitk::GIFNeighbouringGreyLevelDependenceFeature::FeatureListType & featureList, mitk::GIFNeighbouringGreyLevelDependenceFeature::GIFNeighbouringGreyLevelDependenceFeatureConfiguration config) { typedef itk::Image ImageType; typedef itk::Image MaskType; typedef itk::MinimumMaximumImageCalculator MinMaxComputerType; /////////////////////////////////////////////////////////////////////////////////////////////// typename MinMaxComputerType::Pointer minMaxComputer = MinMaxComputerType::New(); minMaxComputer->SetImage(itkImage); minMaxComputer->Compute(); double rangeMin = -0.5 + minMaxComputer->GetMinimum(); double rangeMax = 0.5 + minMaxComputer->GetMaximum(); if (config.UseMinimumIntensity) rangeMin = config.MinimumIntensity; if (config.UseMaximumIntensity) rangeMax = config.MaximumIntensity; // Define Range int numberOfBins = config.Bins; typename MaskType::Pointer maskImage = MaskType::New(); mitk::CastToItkImage(mask, maskImage); std::vector resultVector; mitk::NGLDMMatrixHolder holderOverall(rangeMin, rangeMax, numberOfBins,37); mitk::NGLDMMatrixFeatures overallFeature; CalculateNGLDMMatrix(itkImage, maskImage, config.alpha, config.range, config.direction, holderOverall); LocalCalculateFeatures(holderOverall, overallFeature); std::ostringstream ss; ss << config.range; std::string strRange = ss.str(); MatrixFeaturesTo(overallFeature, "NGLDM (" + strRange + ") overall", featureList); } static void MatrixFeaturesTo(mitk::NGLDMMatrixFeatures features, std::string prefix, mitk::GIFNeighbouringGreyLevelDependenceFeature::FeatureListType &featureList) { featureList.push_back(std::make_pair(prefix + " Low Dependence Emphasis", features.LowDependenceEmphasis)); featureList.push_back(std::make_pair(prefix + " High Dependence Emphasis", features.HighDependenceEmphasis)); featureList.push_back(std::make_pair(prefix + " Low Grey Level Count Emphasis", features.LowGreyLevelCountEmphasis)); featureList.push_back(std::make_pair(prefix + " High Grey Level Count Emphasis", features.HighGreyLevelCountEmphasis)); featureList.push_back(std::make_pair(prefix + " Low Dependence Low Grey Level Emphasis", features.LowDependenceLowGreyLevelEmphasis)); featureList.push_back(std::make_pair(prefix + " Low Dependence High Grey Level Emphasis", features.LowDependenceHighGreyLevelEmphasis)); featureList.push_back(std::make_pair(prefix + " High Dependence Low Grey Level Emphasis", features.HighDependenceLowGreyLevelEmphasis)); featureList.push_back(std::make_pair(prefix + " High Dependence High Grey Level Emphasis", features.HighDependenceHighGreyLevelEmphasis)); featureList.push_back(std::make_pair(prefix + " Grey Level Non-Uniformity", features.GreyLevelNonUniformity)); featureList.push_back(std::make_pair(prefix + " Grey Level Non-Uniformity Normalised", features.GreyLevelNonUniformityNormalised)); featureList.push_back(std::make_pair(prefix + " Dependence Count Non-Uniformity", features.DependenceCountNonUniformity)); featureList.push_back(std::make_pair(prefix + " Dependence Count Non-Uniformtiy Normalised", features.DependenceCountNonUniformityNormalised)); featureList.push_back(std::make_pair(prefix + " Dependence Count Percentage", features.DependenceCountPercentage)); featureList.push_back(std::make_pair(prefix + " Grey Level Mean", features.MeanGreyLevelCount)); featureList.push_back(std::make_pair(prefix + " Grey Level Variance", features.GreyLevelVariance)); featureList.push_back(std::make_pair(prefix + " Dependence Count Mean", features.MeanDependenceCount)); featureList.push_back(std::make_pair(prefix + " Dependence Count Variance", features.DependenceCountVariance)); featureList.push_back(std::make_pair(prefix + " Dependence Count Entropy", features.DependenceCountEntropy)); featureList.push_back(std::make_pair(prefix + " Expected Neighbourhood Size", features.ExpectedNeighbourhoodSize)); featureList.push_back(std::make_pair(prefix + " Average Neighbourhood Size", features.AverageNeighbourhoodSize)); featureList.push_back(std::make_pair(prefix + " Average Incomplete Neighbourhood Size", features.AverageIncompleteNeighbourhoodSize)); featureList.push_back(std::make_pair(prefix + " Percentage of complete Neighbourhoods", features.PercentageOfCompleteNeighbourhoods)); featureList.push_back(std::make_pair(prefix + " Percentage of Dependence Neighbour Voxels", features.PercentageOfDependenceNeighbours)); } mitk::GIFNeighbouringGreyLevelDependenceFeature::GIFNeighbouringGreyLevelDependenceFeature() : m_Range(1.0), m_Bins(6) { SetShortName("ngldm"); SetLongName("ngldm"); } mitk::GIFNeighbouringGreyLevelDependenceFeature::FeatureListType mitk::GIFNeighbouringGreyLevelDependenceFeature::CalculateFeatures(const Image::Pointer & image, const Image::Pointer &mask) { FeatureListType featureList; GIFNeighbouringGreyLevelDependenceFeatureConfiguration config; config.direction = GetDirection(); config.range = m_Range; config.alpha = 0; config.MinimumIntensity = GetMinimumIntensity(); config.MaximumIntensity = GetMaximumIntensity(); config.UseMinimumIntensity = GetUseMinimumIntensity(); config.UseMaximumIntensity = GetUseMaximumIntensity(); config.Bins = GetBins(); AccessByItk_3(image, CalculateCoocurenceFeatures, mask, featureList,config); return featureList; } mitk::GIFNeighbouringGreyLevelDependenceFeature::FeatureNameListType mitk::GIFNeighbouringGreyLevelDependenceFeature::GetFeatureNames() { FeatureNameListType featureList; featureList.push_back("co-occ. Energy Means"); featureList.push_back("co-occ. Energy Std."); return featureList; } void mitk::GIFNeighbouringGreyLevelDependenceFeature::AddArguments(mitkCommandLineParser &parser) { std::string name = GetOptionPrefix(); parser.addArgument(GetLongName(), name, mitkCommandLineParser::String, "Calculate Neighbouring grey level dependence based features", "Calculate Neighbouring grey level dependence based features", us::Any()); parser.addArgument(name + "::range", name + "::range", mitkCommandLineParser::String, "NGLD Range", "Define the range that is used (Semicolon-separated)", us::Any()); parser.addArgument(name + "::bins", name + "::bins", mitkCommandLineParser::String, "NGLD Number of Bins", "Define the number of bins that is used ", us::Any()); } void mitk::GIFNeighbouringGreyLevelDependenceFeature::CalculateFeaturesUsingParameters(const Image::Pointer & feature, const Image::Pointer &, const Image::Pointer &maskNoNAN, FeatureListType &featureList) { auto parsedArgs = GetParameter(); std::string name = GetOptionPrefix(); if (parsedArgs.count(GetLongName())) { std::vector ranges; if (parsedArgs.count(name + "::range")) { ranges = SplitDouble(parsedArgs[name + "::range"].ToString(), ';'); } else { ranges.push_back(1); } if (parsedArgs.count(name + "::bins")) { auto bins = SplitDouble(parsedArgs[name + "::bins"].ToString(), ';')[0]; this->SetBins(bins); } for (std::size_t i = 0; i < ranges.size(); ++i) { MITK_INFO << "Start calculating coocurence with range " << ranges[i] << "...."; this->SetRange(ranges[i]); auto localResults = this->CalculateFeatures(feature, maskNoNAN); featureList.insert(featureList.end(), localResults.begin(), localResults.end()); MITK_INFO << "Finished calculating coocurence with range " << ranges[i] << "...."; } } } diff --git a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFVolumetricStatistics.cpp b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFVolumetricStatistics.cpp index c9a56a77f3..027758b5db 100644 --- a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFVolumetricStatistics.cpp +++ b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFVolumetricStatistics.cpp @@ -1,426 +1,431 @@ /*=================================================================== The Medical Imaging Interaction Toolkit (MITK) Copyright (c) German Cancer Research Center, Division of Medical and Biological Informatics. All rights reserved. This software is distributed WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See LICENSE.txt or http://www.mitk.org for details. ===================================================================*/ #include // MITK #include #include #include #include #include // ITK #include #include // VTK #include #include #include // STL #include #include // OpenCV #include template void CalculateVolumeStatistic(itk::Image* itkImage, mitk::Image::Pointer mask, mitk::GIFVolumetricStatistics::FeatureListType & featureList) { typedef itk::Image ImageType; typedef itk::Image MaskType; typedef itk::LabelStatisticsImageFilter FilterType; typename MaskType::Pointer maskImage = MaskType::New(); mitk::CastToItkImage(mask, maskImage); typename FilterType::Pointer labelStatisticsImageFilter = FilterType::New(); labelStatisticsImageFilter->SetInput( itkImage ); labelStatisticsImageFilter->SetLabelInput(maskImage); labelStatisticsImageFilter->Update(); double volume = labelStatisticsImageFilter->GetCount(1); double voxelVolume = 1; for (int i = 0; i < (int)(VImageDimension); ++i) { volume *= itkImage->GetSpacing()[i]; voxelVolume *= itkImage->GetSpacing()[i]; } featureList.push_back(std::make_pair("Volumetric Features Volume (pixel based)", volume)); featureList.push_back(std::make_pair("Volumetric Features Voxel Volume", voxelVolume)); } template void CalculateLargestDiameter(itk::Image* mask, mitk::Image::Pointer valueImage, mitk::GIFVolumetricStatistics::FeatureListType & featureList) { typedef itk::Image ValueImageType; typename ValueImageType::Pointer itkValueImage = ValueImageType::New(); mitk::CastToItkImage(valueImage, itkValueImage); typedef itk::Image ImageType; typedef typename ImageType::PointType PointType; typename ImageType::SizeType radius; for (int i=0; i < (int)VImageDimension; ++i) radius[i] = 1; itk::NeighborhoodIterator iterator(radius, mask, mask->GetRequestedRegion()); itk::NeighborhoodIterator valueIter(radius, itkValueImage, itkValueImage->GetRequestedRegion()); std::vector borderPoints; // // Calculate surface in different directions // double surface = 0; std::vector directionSurface; for (int i = 0; i < (int)(iterator.Size()); ++i) { auto offset = iterator.GetOffset(i); double deltaS = 1; int nonZeros = 0; for (int j = 0; j < VImageDimension; ++j) { if (offset[j] != 0 && nonZeros == 0) { for (int k = 0; k < VImageDimension; ++k) { if (k != j) deltaS *= mask->GetSpacing()[k]; } nonZeros++; } else if (offset[j] != 0) { deltaS = 0; } } if (nonZeros < 1) deltaS = 0; directionSurface.push_back(deltaS); } // // Prepare calulation of Centre of mass shift // PointType normalCenter(0); PointType normalCenterUncorrected(0); PointType weightedCenter(0); PointType currentPoint; int numberOfPoints = 0; int numberOfPointsUncorrected = 0; double sumOfPoints = 0; while(!iterator.IsAtEnd()) { if (iterator.GetCenterPixel() == 0) { ++iterator; ++valueIter; continue; } mask->TransformIndexToPhysicalPoint(iterator.GetIndex(), currentPoint); normalCenterUncorrected += currentPoint.GetVectorFromOrigin(); ++numberOfPointsUncorrected; double intensityValue = valueIter.GetCenterPixel(); if (intensityValue == intensityValue) { normalCenter += currentPoint.GetVectorFromOrigin(); weightedCenter += currentPoint.GetVectorFromOrigin() * intensityValue; sumOfPoints += intensityValue; ++numberOfPoints; } bool border = false; for (int i = 0; i < (int)(iterator.Size()); ++i) { if (iterator.GetPixel(i) == 0 || ( ! iterator.IndexInBounds(i))) { border = true; surface += directionSurface[i]; //break; } } if (border) { auto centerIndex = iterator.GetIndex(); PointType centerPoint; mask->TransformIndexToPhysicalPoint(centerIndex, centerPoint ); borderPoints.push_back(centerPoint); } ++iterator; ++valueIter; } auto normalCenterVector = normalCenter.GetVectorFromOrigin() / numberOfPoints; auto normalCenterVectorUncorrected = normalCenter.GetVectorFromOrigin() / numberOfPointsUncorrected; auto weightedCenterVector = weightedCenter.GetVectorFromOrigin() / sumOfPoints; auto differenceOfCentersUncorrected = (normalCenterVectorUncorrected - weightedCenterVector).GetNorm(); auto differenceOfCenters = (normalCenterVector - weightedCenterVector).GetNorm(); double longestDiameter = 0; unsigned long numberOfBorderPoints = borderPoints.size(); for (int i = 0; i < (int)numberOfBorderPoints; ++i) { auto point = borderPoints[i]; for (int j = i; j < (int)numberOfBorderPoints; ++j) { double newDiameter=point.EuclideanDistanceTo(borderPoints[j]); if (newDiameter > longestDiameter) longestDiameter = newDiameter; } } featureList.push_back(std::make_pair("Volumetric Features Maximum 3D diameter", longestDiameter)); featureList.push_back(std::make_pair("Volumetric Features Surface (Voxel based)", surface)); featureList.push_back(std::make_pair("Volumetric Features Centre of mass shift", differenceOfCenters)); featureList.push_back(std::make_pair("Volumetric Features Centre of mass shift (Uncorrected)", differenceOfCentersUncorrected)); } mitk::GIFVolumetricStatistics::GIFVolumetricStatistics() { SetLongName("volume"); SetShortName("vol"); } mitk::GIFVolumetricStatistics::FeatureListType mitk::GIFVolumetricStatistics::CalculateFeatures(const Image::Pointer & image, const Image::Pointer &mask) { FeatureListType featureList; + if (image->GetDimension() < 3) + { + return featureList; + } + AccessByItk_2(image, CalculateVolumeStatistic, mask, featureList); AccessByItk_2(mask, CalculateLargestDiameter, image, featureList); vtkSmartPointer mesher = vtkSmartPointer::New(); vtkSmartPointer stats = vtkSmartPointer::New(); mesher->SetInputData(mask->GetVtkImageData()); stats->SetInputConnection(mesher->GetOutputPort()); stats->Update(); double pi = vnl_math::pi; double meshVolume = stats->GetVolume(); double meshSurf = stats->GetSurfaceArea(); double pixelVolume = featureList[0].second; double pixelSurface = featureList[3].second; MITK_INFO << "Surf: " << pixelSurface << " Vol " << pixelVolume; double compactness1 = pixelVolume / (std::sqrt(pi) * std::pow(meshSurf, 2.0 / 3.0)); double compactness1Pixel = pixelVolume / (std::sqrt(pi) * std::pow(pixelSurface, 2.0 / 3.0)); //This is the definition used by Aertz. However, due to 2/3 this feature is not demensionless. Use compactness3 instead. double compactness2 = 36 * pi*pixelVolume*pixelVolume / meshSurf / meshSurf / meshSurf; double compactness2Pixel = 36 * pi*pixelVolume*pixelVolume / pixelSurface / pixelSurface / pixelSurface; double compactness3 = pixelVolume / (std::sqrt(pi) * std::pow(meshSurf, 3.0 / 2.0)); double compactness3Pixel = pixelVolume / (std::sqrt(pi) * std::pow(pixelSurface, 3.0 / 2.0)); double sphericity = std::pow(pi, 1 / 3.0) *std::pow(6 * pixelVolume, 2.0 / 3.0) / meshSurf; double sphericityPixel = std::pow(pi, 1 / 3.0) *std::pow(6 * pixelVolume, 2.0 / 3.0) / pixelSurface; double surfaceToVolume = meshSurf / pixelVolume; double surfaceToVolumePixel = pixelSurface / pixelVolume; double sphericalDisproportion = meshSurf / 4 / pi / std::pow(3.0 / 4.0 / pi * pixelVolume, 2.0 / 3.0); double sphericalDisproportionPixel = pixelSurface / 4 / pi / std::pow(3.0 / 4.0 / pi * pixelVolume, 2.0 / 3.0); double asphericity = std::pow(1.0/compactness2, (1.0 / 3.0)) - 1; double asphericityPixel = std::pow(1.0/compactness2Pixel, (1.0 / 3.0)) - 1; //Calculate center of mass shift int xx = mask->GetDimensions()[0]; int yy = mask->GetDimensions()[1]; int zz = mask->GetDimensions()[2]; double xd = mask->GetGeometry()->GetSpacing()[0]; double yd = mask->GetGeometry()->GetSpacing()[1]; double zd = mask->GetGeometry()->GetSpacing()[2]; std::vector pointsForPCA; std::vector pointsForPCAUncorrected; for (int x = 0; x < xx; x++) { for (int y = 0; y < yy; y++) { for (int z = 0; z < zz; z++) { itk::Image::IndexType index; index[0] = x; index[1] = y; index[2] = z; mitk::ScalarType pxImage; mitk::ScalarType pxMask; mitkPixelTypeMultiplex5( mitk::FastSinglePixelAccess, image->GetChannelDescriptor().GetPixelType(), image, image->GetVolumeData(), index, pxImage, 0); mitkPixelTypeMultiplex5( mitk::FastSinglePixelAccess, mask->GetChannelDescriptor().GetPixelType(), mask, mask->GetVolumeData(), index, pxMask, 0); //Check if voxel is contained in segmentation if (pxMask > 0) { cv::Point3d tmp; tmp.x = x * xd; tmp.y = y * yd; tmp.z = z * zd; pointsForPCAUncorrected.push_back(tmp); if (pxImage == pxImage) { pointsForPCA.push_back(tmp); } } } } } //Calculate PCA Features int sz = pointsForPCA.size(); cv::Mat data_pts = cv::Mat(sz, 3, CV_64FC1); for (int i = 0; i < data_pts.rows; ++i) { data_pts.at(i, 0) = pointsForPCA[i].x; data_pts.at(i, 1) = pointsForPCA[i].y; data_pts.at(i, 2) = pointsForPCA[i].z; } //Calculate PCA Features int szUC = pointsForPCAUncorrected.size(); cv::Mat data_ptsUC = cv::Mat(szUC, 3, CV_64FC1); for (int i = 0; i < data_ptsUC.rows; ++i) { data_ptsUC.at(i, 0) = pointsForPCAUncorrected[i].x; data_ptsUC.at(i, 1) = pointsForPCAUncorrected[i].y; data_ptsUC.at(i, 2) = pointsForPCAUncorrected[i].z; } //Perform PCA analysis cv::PCA pca_analysis(data_pts, cv::Mat(), CV_PCA_DATA_AS_ROW); cv::PCA pca_analysisUC(data_ptsUC, cv::Mat(), CV_PCA_DATA_AS_ROW); //Store the eigenvalues std::vector eigen_val(3); std::vector eigen_valUC(3); for (int i = 0; i < 3; ++i) { eigen_val[i] = pca_analysis.eigenvalues.at(0, i); eigen_valUC[i] = pca_analysisUC.eigenvalues.at(0, i); } std::sort(eigen_val.begin(), eigen_val.end()); std::sort(eigen_valUC.begin(), eigen_valUC.end()); double major = eigen_val[2]; double minor = eigen_val[1]; double least = eigen_val[0]; double elongation = major == 0 ? 0 : minor/major; double flatness = major == 0 ? 0 : least / major; double majorUC = eigen_valUC[2]; double minorUC = eigen_valUC[1]; double leastUC = eigen_valUC[0]; double elongationUC = majorUC == 0 ? 0 : minorUC / majorUC; double flatnessUC = majorUC == 0 ? 0 : leastUC / majorUC; featureList.push_back(std::make_pair("Volumetric Features Volume (mesh based)",meshVolume)); featureList.push_back(std::make_pair("Volumetric Features Surface area",meshSurf)); featureList.push_back(std::make_pair("Volumetric Features Surface to volume ratio",surfaceToVolume)); featureList.push_back(std::make_pair("Volumetric Features Sphericity",sphericity)); featureList.push_back(std::make_pair("Volumetric Features Asphericity",asphericity)); featureList.push_back(std::make_pair("Volumetric Features Compactness 1",compactness1)); featureList.push_back(std::make_pair("Volumetric Features Compactness 2",compactness2)); featureList.push_back(std::make_pair("Volumetric Features Compactness 3",compactness3)); featureList.push_back(std::make_pair("Volumetric Features Spherical disproportion", sphericalDisproportion)); featureList.push_back(std::make_pair("Volumetric Features Surface to volume ratio (Voxel based)", surfaceToVolumePixel)); featureList.push_back(std::make_pair("Volumetric Features Sphericity (Voxel based)", sphericityPixel)); featureList.push_back(std::make_pair("Volumetric Features Asphericity (Voxel based)", asphericityPixel)); featureList.push_back(std::make_pair("Volumetric Features Compactness 1 (Voxel based)", compactness1Pixel)); featureList.push_back(std::make_pair("Volumetric Features Compactness 2 (Voxel based)", compactness2Pixel)); featureList.push_back(std::make_pair("Volumetric Features Compactness 3 (Voxel based)", compactness3Pixel)); featureList.push_back(std::make_pair("Volumetric Features Spherical disproportion (Voxel based)", sphericalDisproportionPixel)); featureList.push_back(std::make_pair("Volumetric Features PCA Major Axis",major)); featureList.push_back(std::make_pair("Volumetric Features PCA Minor Axis",minor)); featureList.push_back(std::make_pair("Volumetric Features PCA Least Axis",least)); featureList.push_back(std::make_pair("Volumetric Features PCA Elongation",elongation)); featureList.push_back(std::make_pair("Volumetric Features PCA Flatness",flatness)); featureList.push_back(std::make_pair("Volumetric Features PCA Major Axis (Uncorrected)", majorUC)); featureList.push_back(std::make_pair("Volumetric Features PCA Minor Axis (Uncorrected)", minorUC)); featureList.push_back(std::make_pair("Volumetric Features PCA Least Axis (Uncorrected)", leastUC)); featureList.push_back(std::make_pair("Volumetric Features PCA Elongation (Uncorrected)", elongationUC)); featureList.push_back(std::make_pair("Volumetric Features PCA Flatness (Uncorrected)", flatnessUC)); return featureList; } mitk::GIFVolumetricStatistics::FeatureNameListType mitk::GIFVolumetricStatistics::GetFeatureNames() { FeatureNameListType featureList; featureList.push_back("Volumetric Features Compactness 1"); featureList.push_back("Volumetric Features Compactness 2"); featureList.push_back("Volumetric Features Compactness 3"); featureList.push_back("Volumetric Features Sphericity"); featureList.push_back("Volumetric Features Asphericity"); featureList.push_back("Volumetric Features Surface to volume ratio"); featureList.push_back("Volumetric Features Surface area"); featureList.push_back("Volumetric Features Volume (mesh based)"); featureList.push_back("Volumetric Features Volume (pixel based)"); featureList.push_back("Volumetric Features Spherical disproportion"); featureList.push_back("Volumetric Features Maximum 3D diameter"); return featureList; } void mitk::GIFVolumetricStatistics::AddArguments(mitkCommandLineParser &parser) { std::string name = GetOptionPrefix(); parser.addArgument(GetLongName(), name, mitkCommandLineParser::String, "Use Volume-Statistic", "calculates volume based features", us::Any()); } void mitk::GIFVolumetricStatistics::CalculateFeaturesUsingParameters(const Image::Pointer & feature, const Image::Pointer &mask, const Image::Pointer &, FeatureListType &featureList) { auto parsedArgs = GetParameter(); if (parsedArgs.count(GetLongName())) { MITK_INFO << "Start calculating volumetric features ...."; auto localResults = this->CalculateFeatures(feature, mask); featureList.insert(featureList.end(), localResults.begin(), localResults.end()); MITK_INFO << "Finished calculating volumetric features...."; } }