diff --git a/Modules/ImageStatistics/mitkImageStatisticsCalculator.cpp b/Modules/ImageStatistics/mitkImageStatisticsCalculator.cpp
index 398dffc557..bc01620bdb 100644
--- a/Modules/ImageStatistics/mitkImageStatisticsCalculator.cpp
+++ b/Modules/ImageStatistics/mitkImageStatisticsCalculator.cpp
@@ -1,517 +1,517 @@
#include <limits>
#include <cmath>
#include <itkImageToHistogramFilter.h>
#include <itkMaskedImageToHistogramFilter.h>
#include <itkHistogram.h>
#include <itkExtractImageFilter.h>
#include <itkImageRegionConstIterator.h>
#include <itkChangeInformationImageFilter.h>
#include <itkMinimumMaximumImageCalculator.h>
#include <itkMinimumMaximumImageFilter.h>
#include <itkMaskImageFilter.h>
#include <itkExceptionObject.h>
#include <mitkImageStatisticsCalculator.h>
#include <mitkImage.h>
#include <mitkHistogramStatisticsCalculator.h>
#include <mitkImageAccessByItk.h>
#include <mitkImageToItk.h>
#include <mitkExtendedStatisticsImageFilter.h>
#include <mitkExtendedLabelStatisticsImageFilter.h>
#include <mitkImageTimeSelector.h>
#include <mitkMinMaxImageFilterWithIndex.h>
#include <mitkMinMaxLabelmageFilterWithIndex.h>
#include <mitkitkMaskImageFilter.h>
#include <mitkImageCast.h>
#include <mitkMaskUtilities.h>
#include "itkImageFileWriter.h"
namespace mitk
{
void ImageStatisticsCalculator::SetInputImage(mitk::Image::Pointer image)
{
if (image != m_Image)
{
m_Image = image;
m_StatisticsByTimeStep.resize(m_Image->GetTimeSteps());
m_StatisticsUpdateTimePerTimeStep.resize(m_Image->GetTimeSteps());
std::fill(m_StatisticsUpdateTimePerTimeStep.begin(), m_StatisticsUpdateTimePerTimeStep.end(), 0);
this->Modified();
}
}
void ImageStatisticsCalculator::SetMask(mitk::MaskGenerator::Pointer mask)
{
if (mask != m_MaskGenerator)
{
m_MaskGenerator = mask;
this->Modified();
}
}
void ImageStatisticsCalculator::SetSecondaryMask(mitk::MaskGenerator::Pointer mask)
{
if (mask != m_SecondaryMaskGenerator)
{
m_SecondaryMaskGenerator = mask;
this->Modified();
}
}
void ImageStatisticsCalculator::SetNBinsForHistogramStatistics(unsigned int nBins)
{
if (nBins != m_nBinsForHistogramStatistics)
{
m_nBinsForHistogramStatistics = nBins;
this->Modified();
this->m_UseBinSizeOverNBins = false;
}
if (m_UseBinSizeOverNBins)
{
this->Modified();
this->m_UseBinSizeOverNBins = false;
}
}
unsigned int ImageStatisticsCalculator::GetNBinsForHistogramStatistics() const
{
return m_nBinsForHistogramStatistics;
}
void ImageStatisticsCalculator::SetBinSizeForHistogramStatistics(double binSize)
{
if (binSize != m_binSizeForHistogramStatistics)
{
m_binSizeForHistogramStatistics = binSize;
this->Modified();
this->m_UseBinSizeOverNBins = true;
}
if (!m_UseBinSizeOverNBins)
{
this->Modified();
this->m_UseBinSizeOverNBins = true;
}
}
double ImageStatisticsCalculator::GetBinSizeForHistogramStatistics() const
{
return m_binSizeForHistogramStatistics;
}
mitk::StatisticsContainer::Pointer ImageStatisticsCalculator::GetStatistics(unsigned int timeStep, unsigned int label)
{
if (timeStep >= m_StatisticsByTimeStep.size())
{
mitkThrow() << "invalid timeStep in ImageStatisticsCalculator_v2::GetStatistics";
}
if (m_Image.IsNull())
{
mitkThrow() << "no image";
}
if (!m_Image->IsInitialized())
{
mitkThrow() << "Image not initialized!";
}
if (IsUpdateRequired(timeStep))
{
if (m_MaskGenerator.IsNotNull())
{
m_MaskGenerator->SetTimeStep(timeStep);
m_InternalMask = m_MaskGenerator->GetMask();
if (m_MaskGenerator->GetReferenceImage().IsNotNull())
{
m_InternalImageForStatistics = m_MaskGenerator->GetReferenceImage();
}
else
{
m_InternalImageForStatistics = m_Image;
}
}
else
{
m_InternalImageForStatistics = m_Image;
}
if (m_SecondaryMaskGenerator.IsNotNull())
{
m_SecondaryMaskGenerator->SetTimeStep(timeStep);
m_SecondaryMask = m_SecondaryMaskGenerator->GetMask();
}
ImageTimeSelector::Pointer imgTimeSel = ImageTimeSelector::New();
imgTimeSel->SetInput(m_InternalImageForStatistics);
imgTimeSel->SetTimeNr(timeStep);
imgTimeSel->UpdateLargestPossibleRegion();
m_ImageTimeSlice = imgTimeSel->GetOutput();
// Calculate statistics with/without mask
if (m_MaskGenerator.IsNull() && m_SecondaryMaskGenerator.IsNull())
{
// 1) calculate statistics unmasked:
AccessByItk_1(m_ImageTimeSlice, InternalCalculateStatisticsUnmasked, timeStep)
}
else
{
// 2) calculate statistics masked
AccessByItk_1(m_ImageTimeSlice, InternalCalculateStatisticsMasked, timeStep)
}
//this->Modified();
}
m_StatisticsUpdateTimePerTimeStep[timeStep] = m_StatisticsByTimeStep[timeStep][m_StatisticsByTimeStep[timeStep].size()-1]->GetMTime();
for (auto it = m_StatisticsByTimeStep[timeStep].begin(); it != m_StatisticsByTimeStep[timeStep].end(); ++it)
{
StatisticsContainer::Pointer statCont = *it;
if (statCont->GetLabel() == label)
{
- return statCont->Clone();
+ return statCont;
}
}
// these lines will ony be executed if the requested label could not be found!
MITK_WARN << "Invalid label: " << label << " in time step: " << timeStep;
return StatisticsContainer::New();
}
template < typename TPixel, unsigned int VImageDimension > void ImageStatisticsCalculator::InternalCalculateStatisticsUnmasked(
typename itk::Image< TPixel, VImageDimension >* image, unsigned int timeStep)
{
typedef typename itk::Image< TPixel, VImageDimension > ImageType;
typedef typename itk::ExtendedStatisticsImageFilter<ImageType> ImageStatisticsFilterType;
typedef typename itk::MinMaxImageFilterWithIndex<ImageType> MinMaxFilterType;
StatisticsContainer::Pointer statisticsResult = StatisticsContainer::New();
typename ImageStatisticsFilterType::Pointer statisticsFilter = ImageStatisticsFilterType::New();
statisticsFilter->SetInput(image);
statisticsFilter->SetCoordinateTolerance(0.001);
statisticsFilter->SetDirectionTolerance(0.001);
// TODO: this is single threaded. Implement our own image filter that does this multi threaded
// typename itk::MinimumMaximumImageCalculator<ImageType>::Pointer imgMinMaxFilter = itk::MinimumMaximumImageCalculator<ImageType>::New();
// imgMinMaxFilter->SetImage(image);
// imgMinMaxFilter->Compute();
vnl_vector<int> minIndex, maxIndex;
typename MinMaxFilterType::Pointer minMaxFilter = MinMaxFilterType::New();
minMaxFilter->SetInput(image);
minMaxFilter->UpdateLargestPossibleRegion();
typename ImageType::PixelType minval = minMaxFilter->GetMin();
typename ImageType::PixelType maxval = minMaxFilter->GetMax();
typename ImageType::IndexType tmpMinIndex = minMaxFilter->GetMinIndex();
typename ImageType::IndexType tmpMaxIndex = minMaxFilter->GetMaxIndex();
// typename ImageType::IndexType tmpMinIndex = imgMinMaxFilter->GetIndexOfMinimum();
// typename ImageType::IndexType tmpMaxIndex = imgMinMaxFilter->GetIndexOfMaximum();
minIndex.set_size(tmpMaxIndex.GetIndexDimension());
maxIndex.set_size(tmpMaxIndex.GetIndexDimension());
for (unsigned int i=0; i < tmpMaxIndex.GetIndexDimension(); i++)
{
minIndex[i] = tmpMinIndex[i];
maxIndex[i] = tmpMaxIndex[i];
}
statisticsResult->SetMinIndex(minIndex);
statisticsResult->SetMaxIndex(maxIndex);
//convert m_binSize in m_nBins if necessary
unsigned int nBinsForHistogram;
if (m_UseBinSizeOverNBins)
{
nBinsForHistogram = std::max(static_cast<double>(std::ceil(maxval - minval)) / m_binSizeForHistogramStatistics, 10.); // do not allow less than 10 bins
}
else
{
nBinsForHistogram = m_nBinsForHistogramStatistics;
}
statisticsFilter->SetHistogramParameters(nBinsForHistogram, minval, maxval);
try
{
statisticsFilter->Update();
}
catch (const itk::ExceptionObject& e)
{
mitkThrow() << "Image statistics calculation failed due to following ITK Exception: \n " << e.what();
}
auto voxelVolume = GetVoxelVolume<TPixel, VImageDimension>(image);
// no mask, therefore just one label = the whole image
m_StatisticsByTimeStep[timeStep].resize(1);
statisticsResult->SetLabel(1);
statisticsResult->SetN(image->GetLargestPossibleRegion().GetNumberOfPixels());
statisticsResult->SetVolume(statisticsResult->GetN()*voxelVolume);
statisticsResult->SetMean(statisticsFilter->GetMean());
statisticsResult->SetMin(statisticsFilter->GetMinimum());
statisticsResult->SetMax(statisticsFilter->GetMaximum());
statisticsResult->SetStd(statisticsFilter->GetSigma());
statisticsResult->SetSkewness(statisticsFilter->GetSkewness());
statisticsResult->SetKurtosis(statisticsFilter->GetKurtosis());
statisticsResult->SetRMS(std::sqrt(std::pow(statisticsFilter->GetMean(), 2.) + statisticsFilter->GetVariance())); // variance = sigma^2
statisticsResult->SetMPP(statisticsFilter->GetMPP());
statisticsResult->SetEntropy(statisticsFilter->GetEntropy());
statisticsResult->SetMedian(statisticsFilter->GetMedian());
statisticsResult->SetUniformity(statisticsFilter->GetUniformity());
statisticsResult->SetUPP(statisticsFilter->GetUPP());
statisticsResult->SetHistogram(statisticsFilter->GetHistogram());
m_StatisticsByTimeStep[timeStep][0] = statisticsResult;
}
template < typename TPixel, unsigned int VImageDimension >
double ImageStatisticsCalculator::GetVoxelVolume(typename itk::Image<TPixel, VImageDimension>* image) const
{
auto spacing = image->GetSpacing();
double voxelVolume = 1.;
for (unsigned int i = 0; i < image->GetImageDimension(); i++) {
voxelVolume *= spacing[i];
}
return voxelVolume;
}
template < typename TPixel, unsigned int VImageDimension > void ImageStatisticsCalculator::InternalCalculateStatisticsMasked(
typename itk::Image< TPixel, VImageDimension >* image,
unsigned int timeStep)
{
typedef itk::Image< TPixel, VImageDimension > ImageType;
typedef itk::Image< MaskPixelType, VImageDimension > MaskType;
typedef typename MaskType::PixelType LabelPixelType;
typedef itk::ExtendedLabelStatisticsImageFilter< ImageType, MaskType > ImageStatisticsFilterType;
typedef MaskUtilities< TPixel, VImageDimension > MaskUtilType;
typedef typename itk::MinMaxLabelImageFilterWithIndex<ImageType, MaskType> MinMaxLabelFilterType;
typedef typename ImageType::PixelType InputImgPixelType;
// workaround: if m_SecondaryMaskGenerator ist not null but m_MaskGenerator is! (this is the case if we request a 'ignore zuero valued pixels'
// mask in the gui but do not define a primary mask)
bool swapMasks = false;
if (m_SecondaryMask.IsNotNull() && m_InternalMask.IsNull())
{
m_InternalMask = m_SecondaryMask;
m_SecondaryMask = nullptr;
swapMasks = true;
}
// maskImage has to have the same dimension as image
typename MaskType::Pointer maskImage = MaskType::New();
try {
// try to access the pixel values directly (no copying or casting). Only works if mask pixels are of pixelType unsigned short
maskImage = ImageToItkImage< MaskPixelType, VImageDimension >(m_InternalMask);
}
catch (const itk::ExceptionObject &)
{
// if the pixel type of the mask is not short, then we have to make a copy of m_InternalMask (and cast the values)
CastToItkImage(m_InternalMask, maskImage);
}
// if we have a secondary mask (say a ignoreZeroPixelMask) we need to combine the masks (corresponds to AND)
if (m_SecondaryMask.IsNotNull())
{
// dirty workaround for a bug when pf mask + any other mask is used in conjunction. We need a proper fix for this (Fabian Isensee is responsible and probably working on it!)
if (m_InternalMask->GetDimension() == 2 && (m_SecondaryMask->GetDimension() == 3 || m_SecondaryMask->GetDimension() == 4))
{
mitk::Image::Pointer old_img = m_SecondaryMaskGenerator->GetReferenceImage();
m_SecondaryMaskGenerator->SetInputImage(m_MaskGenerator->GetReferenceImage());
m_SecondaryMask = m_SecondaryMaskGenerator->GetMask();
m_SecondaryMaskGenerator->SetInputImage(old_img);
}
typename MaskType::Pointer secondaryMaskImage = MaskType::New();
secondaryMaskImage = ImageToItkImage< MaskPixelType, VImageDimension >(m_SecondaryMask);
// secondary mask should be a ignore zero value pixel mask derived from image. it has to be cropped to the mask region (which may be planar or simply smaller)
typename MaskUtilities<MaskPixelType, VImageDimension>::Pointer secondaryMaskMaskUtil = MaskUtilities<MaskPixelType, VImageDimension>::New();
secondaryMaskMaskUtil->SetImage(secondaryMaskImage.GetPointer());
secondaryMaskMaskUtil->SetMask(maskImage.GetPointer());
typename MaskType::Pointer adaptedSecondaryMaskImage = secondaryMaskMaskUtil->ExtractMaskImageRegion();
typename itk::MaskImageFilter2<MaskType, MaskType, MaskType>::Pointer maskFilter = itk::MaskImageFilter2<MaskType, MaskType, MaskType>::New();
maskFilter->SetInput1(maskImage);
maskFilter->SetInput2(adaptedSecondaryMaskImage);
maskFilter->SetMaskingValue(1); // all pixels of maskImage where secondaryMaskImage==1 will be kept, all the others are set to 0
maskFilter->UpdateLargestPossibleRegion();
maskImage = maskFilter->GetOutput();
}
typename MaskUtilType::Pointer maskUtil = MaskUtilType::New();
maskUtil->SetImage(image);
maskUtil->SetMask(maskImage.GetPointer());
// if mask is smaller than image, extract the image region where the mask is
typename ImageType::Pointer adaptedImage = ImageType::New();
adaptedImage = maskUtil->ExtractMaskImageRegion(); // this also checks mask sanity
// find min, max, minindex and maxindex
typename MinMaxLabelFilterType::Pointer minMaxFilter = MinMaxLabelFilterType::New();
minMaxFilter->SetInput(adaptedImage);
minMaxFilter->SetLabelInput(maskImage);
minMaxFilter->UpdateLargestPossibleRegion();
// set histogram parameters for each label individually (min/max may be different for each label)
typedef typename std::map<LabelPixelType, InputImgPixelType> MapType;
typedef typename std::pair<LabelPixelType, InputImgPixelType> PairType;
std::vector<LabelPixelType> relevantLabels = minMaxFilter->GetRelevantLabels();
MapType minVals;
MapType maxVals;
std::map<LabelPixelType, unsigned int> nBins;
for (LabelPixelType label:relevantLabels)
{
minVals.insert(PairType(label, minMaxFilter->GetMin(label)));
maxVals.insert(PairType(label, minMaxFilter->GetMax(label)));
unsigned int nBinsForHistogram;
if (m_UseBinSizeOverNBins)
{
nBinsForHistogram = std::max(static_cast<double>(std::ceil(minMaxFilter->GetMax(label) - minMaxFilter->GetMin(label))) / m_binSizeForHistogramStatistics, 10.); // do not allow less than 10 bins
}
else
{
nBinsForHistogram = m_nBinsForHistogramStatistics;
}
nBins.insert(typename std::pair<LabelPixelType, unsigned int>(label, nBinsForHistogram));
}
typename ImageStatisticsFilterType::Pointer imageStatisticsFilter = ImageStatisticsFilterType::New();
imageStatisticsFilter->SetDirectionTolerance(0.001);
imageStatisticsFilter->SetCoordinateTolerance(0.001);
imageStatisticsFilter->SetInput(adaptedImage);
imageStatisticsFilter->SetLabelInput(maskImage);
imageStatisticsFilter->SetHistogramParametersForLabels(nBins, minVals, maxVals);
imageStatisticsFilter->Update();
std::list<int> labels = imageStatisticsFilter->GetRelevantLabels();
auto it = labels.begin();
m_StatisticsByTimeStep[timeStep].resize(0);
while(it != labels.end())
{
StatisticsContainer::Pointer statisticsResult = StatisticsContainer::New();
// find min, max, minindex and maxindex
// make sure to only look in the masked region, use a masker for this
vnl_vector<int> minIndex, maxIndex;
mitk::Point3D worldCoordinateMin;
mitk::Point3D worldCoordinateMax;
mitk::Point3D indexCoordinateMin;
mitk::Point3D indexCoordinateMax;
m_InternalImageForStatistics->GetGeometry()->IndexToWorld(minMaxFilter->GetMinIndex(*it), worldCoordinateMin);
m_InternalImageForStatistics->GetGeometry()->IndexToWorld(minMaxFilter->GetMaxIndex(*it), worldCoordinateMax);
m_Image->GetGeometry()->WorldToIndex(worldCoordinateMin, indexCoordinateMin);
m_Image->GetGeometry()->WorldToIndex(worldCoordinateMax, indexCoordinateMax);
minIndex.set_size(3);
maxIndex.set_size(3);
//for (unsigned int i=0; i < tmpMaxIndex.GetIndexDimension(); i++)
for (unsigned int i=0; i < 3; i++)
{
minIndex[i] = indexCoordinateMin[i];
maxIndex[i] = indexCoordinateMax[i];
}
statisticsResult->SetMinIndex(minIndex);
statisticsResult->SetMaxIndex(maxIndex);
assert(std::abs(minMaxFilter->GetMax(*it) - imageStatisticsFilter->GetMaximum(*it)) < mitk::eps);
assert(std::abs(minMaxFilter->GetMin(*it) - imageStatisticsFilter->GetMinimum(*it)) < mitk::eps);
auto voxelVolume = GetVoxelVolume<TPixel, VImageDimension>(image);
statisticsResult->SetN(imageStatisticsFilter->GetSum(*it) / (double) imageStatisticsFilter->GetMean(*it));
statisticsResult->SetVolume(statisticsResult->GetN()*voxelVolume);
statisticsResult->SetMean(imageStatisticsFilter->GetMean(*it));
statisticsResult->SetMin(imageStatisticsFilter->GetMinimum(*it));
statisticsResult->SetMax(imageStatisticsFilter->GetMaximum(*it));
statisticsResult->SetStd(imageStatisticsFilter->GetSigma(*it));
statisticsResult->SetSkewness(imageStatisticsFilter->GetSkewness(*it));
statisticsResult->SetKurtosis(imageStatisticsFilter->GetKurtosis(*it));
statisticsResult->SetRMS(std::sqrt(std::pow(imageStatisticsFilter->GetMean(*it), 2.) + imageStatisticsFilter->GetVariance(*it))); // variance = sigma^2
statisticsResult->SetMPP(imageStatisticsFilter->GetMPP(*it));
statisticsResult->SetLabel(*it);
statisticsResult->SetEntropy(imageStatisticsFilter->GetEntropy(*it));
statisticsResult->SetMedian(imageStatisticsFilter->GetMedian(*it));
statisticsResult->SetUniformity(imageStatisticsFilter->GetUniformity(*it));
statisticsResult->SetUPP(imageStatisticsFilter->GetUPP(*it));
statisticsResult->SetHistogram(imageStatisticsFilter->GetHistogram(*it));
m_StatisticsByTimeStep[timeStep].push_back(statisticsResult);
++it;
}
// swap maskGenerators back
if (swapMasks)
{
m_SecondaryMask = m_InternalMask;
m_InternalMask = nullptr;
}
}
bool ImageStatisticsCalculator::IsUpdateRequired(unsigned int timeStep) const
{
unsigned long thisClassTimeStamp = this->GetMTime();
unsigned long inputImageTimeStamp = m_Image->GetMTime();
unsigned long statisticsTimeStamp = m_StatisticsUpdateTimePerTimeStep[timeStep];
if (thisClassTimeStamp > statisticsTimeStamp) // inputs have changed
{
return true;
}
if (inputImageTimeStamp > statisticsTimeStamp) // image has changed
{
return true;
}
if (m_MaskGenerator.IsNotNull())
{
unsigned long maskGeneratorTimeStamp = m_MaskGenerator->GetMTime();
if (maskGeneratorTimeStamp > statisticsTimeStamp) // there is a mask generator and it has changed
{
return true;
}
}
if (m_SecondaryMaskGenerator.IsNotNull())
{
unsigned long maskGeneratorTimeStamp = m_SecondaryMaskGenerator->GetMTime();
if (maskGeneratorTimeStamp > statisticsTimeStamp) // there is a secondary mask generator and it has changed
{
return true;
}
}
return false;
}
}