diff --git a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFFirstOrderNumericStatistics.cpp b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFFirstOrderNumericStatistics.cpp index 0c3893a453..e30bb5b422 100644 --- a/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFFirstOrderNumericStatistics.cpp +++ b/Modules/Classification/CLUtilities/src/GlobalImageFeatures/mitkGIFFirstOrderNumericStatistics.cpp @@ -1,435 +1,490 @@ /*=================================================================== 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 // ITK #include #include #include // STL #include struct FirstOrderNumericParameterStruct { mitk::IntensityQuantifier::Pointer quantifier; double MinimumIntensity; double MaximumIntensity; int Bins; std::string prefix; }; template void CalculateFirstOrderStatistics(itk::Image* itkImage, mitk::Image::Pointer mask, mitk::GIFFirstOrderNumericStatistics::FeatureListType & featureList, FirstOrderNumericParameterStruct params) { typedef itk::Image ImageType; typedef itk::Image MaskType; typename MaskType::Pointer maskImage = MaskType::New(); mitk::CastToItkImage(mask, maskImage); // // Calculate the Volume of Voxel (Maximum to the 3th order) // double voxelVolume = 1; for (unsigned int i = 0; i < std::min(3, VImageDimension); ++i) voxelVolume *= itkImage->GetSpacing()[i]; // // Calculate the Hypervolume of Voxel // double voxelSpace = 1; for (unsigned int i = 0; i < VImageDimension; ++i) voxelSpace *= itkImage->GetSpacing()[i]; unsigned int numberOfBins = params.quantifier->GetBins(); std::vector histogram; histogram.resize(numberOfBins, 0); double minimum = std::numeric_limits::max(); double maximum = std::numeric_limits::lowest(); double sum = 0; double sumTwo= 0; double sumThree = 0; unsigned int numberOfVoxels = 0; itk::ImageRegionIterator imageIter(itkImage, itkImage->GetLargestPossibleRegion()); itk::ImageRegionIterator maskIter(maskImage, maskImage->GetLargestPossibleRegion()); while (!imageIter.IsAtEnd()) { if (maskIter.Get() > 0) { double value = imageIter.Get(); minimum = std::min(minimum, value); maximum = std::max(maximum, value); sum += value; sumTwo += value * value; sumThree += value * value*value; histogram[params.quantifier->IntensityToIndex(value)] += 1; ++numberOfVoxels; } ++maskIter; ++imageIter; } // // Histogram based calculations // unsigned int passedValues = 0; - bool found10th = false; - bool found90th = false; - bool foundMedian = false; + double doubleNoOfVoxes = numberOfVoxels; double median = 0; + double lastIntensityWithValues = params.quantifier->IndexToMeanIntensity(0); + + std::vector percentiles; + percentiles.resize(20, 0); + for (std::size_t idx = 0; idx < histogram.size(); ++idx) { + unsigned int actualValues = histogram[idx]; - } + for (std::size_t percentileIdx = 0; percentileIdx < percentiles.size(); ++percentileIdx) + { + double threshold = doubleNoOfVoxes * (percentileIdx + 1) *1.0 / (percentiles.size()); + if ((passedValues < threshold) & ((passedValues + actualValues) >= threshold)) + { + // Lower Bound + if (passedValues == std::floor(threshold)) + { + percentiles[percentileIdx] = 0.5*(lastIntensityWithValues + params.quantifier->IndexToMeanIntensity(idx)); + } + else + { + percentiles[percentileIdx] = params.quantifier->IndexToMeanIntensity(idx); + } + } + } + if ((passedValues < doubleNoOfVoxes * 0.5) & ((passedValues + actualValues) >= doubleNoOfVoxes * 0.5)) + { + // Lower Bound + if (passedValues == std::floor(doubleNoOfVoxes * 0.5)) + { + median = 0.5*(lastIntensityWithValues + params.quantifier->IndexToMeanIntensity(idx)); + } + else + { + median = params.quantifier->IndexToMeanIntensity(idx); + } + } + + if (actualValues > 0) + { + lastIntensityWithValues = params.quantifier->IndexToMeanIntensity(idx); + } + passedValues += actualValues; + } - double mean = sum / numberOfVoxels; - double variance = sumTwo / numberOfVoxels - (mean*mean); - double skewness = (sumThree / numberOfVoxels - 3 * mean * variance - mean * mean * mean) / (std::pow(variance, 3/2 )); + double mean = sum / (numberOfVoxels); + double variance = sumTwo / (numberOfVoxels) - (mean*mean); double energy = sumTwo; double rootMeanSquare = std::sqrt(sumTwo / numberOfVoxels); double sumAbsoluteDistanceToMean = 0; + double sumValueMinusMean = 0; + double sumValueMinusMeanTwo = 0; + double sumValueMinusMeanThree = 0; + double sumValueMinusMeanFour = 0; maskIter.GoToBegin(); imageIter.GoToBegin(); while (!imageIter.IsAtEnd()) { if (maskIter.Get() > 0) { double value = imageIter.Get(); + double valueMinusMean = value - mean; - sumAbsoluteDistanceToMean += std::abs(value - mean); - + sumAbsoluteDistanceToMean += std::abs(valueMinusMean); + sumValueMinusMean += valueMinusMean; + sumValueMinusMeanTwo += valueMinusMean * valueMinusMean; + sumValueMinusMeanThree += valueMinusMean * valueMinusMean * valueMinusMean; + sumValueMinusMeanFour += valueMinusMean * valueMinusMean * valueMinusMean * valueMinusMean; } ++maskIter; ++imageIter; } double meanAbsoluteDeviation = sumAbsoluteDistanceToMean / numberOfVoxels; - + double skewness = sumValueMinusMeanThree / numberOfVoxels / variance / std::sqrt(variance); + double kurtosis = sumValueMinusMeanFour / numberOfVoxels / variance / variance; + double interquantileRange = params.quantifier->IntensityToIndex(percentiles[14]) - params.quantifier->IntensityToIndex(percentiles[4]); featureList.push_back(std::make_pair(params.prefix + "Mean", mean)); featureList.push_back(std::make_pair(params.prefix + "Variance", variance)); featureList.push_back(std::make_pair(params.prefix + "Skewness", skewness)); + featureList.push_back(std::make_pair(params.prefix + "Excess kurtosis", kurtosis-3)); + featureList.push_back(std::make_pair(params.prefix + "Median", median)); featureList.push_back(std::make_pair(params.prefix + "Minimum", minimum)); + featureList.push_back(std::make_pair(params.prefix + "Percentile 10", percentiles[1])); + featureList.push_back(std::make_pair(params.prefix + "Percentile 90", percentiles[17])); featureList.push_back(std::make_pair(params.prefix + "Maximum", maximum)); + featureList.push_back(std::make_pair(params.prefix + "Interquantile range", interquantileRange)); featureList.push_back(std::make_pair(params.prefix + "Range", maximum-minimum)); featureList.push_back(std::make_pair(params.prefix + "Mean absolute deviation", meanAbsoluteDeviation)); featureList.push_back(std::make_pair(params.prefix + "Energy", energy)); featureList.push_back(std::make_pair(params.prefix + "Root mean square", rootMeanSquare)); + featureList.push_back(std::make_pair(params.prefix + "Standard Deviation", std::sqrt(variance))); + featureList.push_back(std::make_pair(params.prefix + "Kurtosis", kurtosis)); + return; /* typedef itk::LabelStatisticsImageFilter FilterType; typedef typename FilterType::HistogramType HistogramType; typedef typename HistogramType::IndexType HIndexType; typedef itk::MinimumMaximumImageCalculator MinMaxComputerType; typename MinMaxComputerType::Pointer minMaxComputer = MinMaxComputerType::New(); minMaxComputer->SetImage(itkImage); minMaxComputer->Compute(); double imageRange = minMaxComputer->GetMaximum() - minMaxComputer->GetMinimum(); typename FilterType::Pointer labelStatisticsImageFilter = FilterType::New(); labelStatisticsImageFilter->SetInput( itkImage ); labelStatisticsImageFilter->SetLabelInput(maskImage); labelStatisticsImageFilter->SetUseHistograms(true); double min = params.MinimumIntensity; double max = params.MaximumIntensity; labelStatisticsImageFilter->SetHistogramParameters(params.Bins, min,max); labelStatisticsImageFilter->Update(); // --------------- Range -------------------- double range = labelStatisticsImageFilter->GetMaximum(1) - labelStatisticsImageFilter->GetMinimum(1); // --------------- Uniformity, Entropy -------------------- double count = labelStatisticsImageFilter->GetCount(1); //double std_dev = labelStatisticsImageFilter->GetSigma(1); double mean = labelStatisticsImageFilter->GetMean(1); double median = labelStatisticsImageFilter->GetMedian(1); auto histogram = labelStatisticsImageFilter->GetHistogram(1); bool histogramIsCalculated = histogram; HIndexType index; index.SetSize(1); double uniformity = 0; double entropy = 0; double squared_sum = 0; double kurtosis = 0; double mean_absolut_deviation = 0; double median_absolut_deviation = 0; double skewness = 0; double sum_prob = 0; double binWidth = 0; double p05th = 0, p10th = 0, p15th = 0, p20th = 0, p25th = 0, p30th = 0, p35th = 0, p40th = 0, p45th = 0, p50th = 0; double p55th = 0, p60th = 0, p65th = 0, p70th = 0, p75th = 0, p80th = 0, p85th = 0, p90th = 0, p95th = 0; double voxelValue = 0; if (histogramIsCalculated) { binWidth = histogram->GetBinMax(0, 0) - histogram->GetBinMin(0, 0); p05th = histogram->Quantile(0, 0.05); p10th = histogram->Quantile(0, 0.10); p15th = histogram->Quantile(0, 0.15); p20th = histogram->Quantile(0, 0.20); p25th = histogram->Quantile(0, 0.25); p30th = histogram->Quantile(0, 0.30); p35th = histogram->Quantile(0, 0.35); p40th = histogram->Quantile(0, 0.40); p45th = histogram->Quantile(0, 0.45); p50th = histogram->Quantile(0, 0.50); p55th = histogram->Quantile(0, 0.55); p60th = histogram->Quantile(0, 0.60); p65th = histogram->Quantile(0, 0.65); p70th = histogram->Quantile(0, 0.70); p75th = histogram->Quantile(0, 0.75); p80th = histogram->Quantile(0, 0.80); p85th = histogram->Quantile(0, 0.85); p90th = histogram->Quantile(0, 0.90); p95th = histogram->Quantile(0, 0.95); } double Log2=log(2); double mode_bin; double mode_value = 0; double variance = 0; if (histogramIsCalculated) { for (int i = 0; i < (int)(histogram->GetSize(0)); ++i) { index[0] = i; double prob = histogram->GetFrequency(index); if (prob < 0.00000001) continue; voxelValue = histogram->GetBinMin(0, i) + binWidth * 0.5; if (prob > mode_value) { mode_value = prob; mode_bin = voxelValue; } sum_prob += prob; squared_sum += prob * voxelValue*voxelValue; prob /= count; mean_absolut_deviation += prob* std::abs(voxelValue - mean); median_absolut_deviation += prob* std::abs(voxelValue - median); variance += prob * (voxelValue - mean) * (voxelValue - mean); kurtosis += prob* (voxelValue - mean) * (voxelValue - mean) * (voxelValue - mean) * (voxelValue - mean); skewness += prob* (voxelValue - mean) * (voxelValue - mean) * (voxelValue - mean); uniformity += prob*prob; if (prob > 0) { entropy += prob * std::log(prob) / Log2; } } } entropy = -entropy; double uncorrected_std_dev = std::sqrt(variance); double rms = std::sqrt(squared_sum / count); kurtosis = kurtosis / (variance * variance); skewness = skewness / (variance * uncorrected_std_dev); double coveredGrayValueRange = range / imageRange; double coefficient_of_variation = (mean == 0) ? 0 : std::sqrt(variance) / mean; double quantile_coefficient_of_dispersion = (p75th - p25th) / (p75th + p25th); //Calculate the robust mean absolute deviation //First, set all frequencies to 0 that are <10th or >90th percentile double meanRobust = 0.0; double robustMeanAbsoluteDeviation = 0.0; if (histogramIsCalculated) { for (int i = 0; i < (int)(histogram->GetSize(0)); ++i) { index[0] = i; if (histogram->GetBinMax(0, i) < p10th) { histogram->SetFrequencyOfIndex(index, 0); } else if (histogram->GetBinMin(0, i) > p90th) { histogram->SetFrequencyOfIndex(index, 0); } } //Calculate the mean for (int i = 0; i < (int)(histogram->GetSize(0)); ++i) { index[0] = i; meanRobust += histogram->GetFrequency(index) * 0.5 * (histogram->GetBinMin(0, i) + histogram->GetBinMax(0, i)); } meanRobust = meanRobust / histogram->GetTotalFrequency(); for (int i = 0; i < (int)(histogram->GetSize(0)); ++i) { index[0] = i; robustMeanAbsoluteDeviation += std::abs(histogram->GetFrequency(index) * ((0.5 * (histogram->GetBinMin(0, i) + histogram->GetBinMax(0, i))) - meanRobust )); } robustMeanAbsoluteDeviation = robustMeanAbsoluteDeviation / histogram->GetTotalFrequency(); } featureList.push_back(std::make_pair(params.prefix + "Mean", labelStatisticsImageFilter->GetMean(1))); featureList.push_back(std::make_pair(params.prefix + "Unbiased Variance", labelStatisticsImageFilter->GetVariance(1))); //Siehe Definition von Unbiased Variance estimation. (Wird nicht durch n sondern durch n-1 normalisiert) featureList.push_back(std::make_pair(params.prefix + "Biased Variance", variance)); featureList.push_back(std::make_pair(params.prefix + "Skewness", skewness)); featureList.push_back(std::make_pair(params.prefix + "Kurtosis", kurtosis)); featureList.push_back(std::make_pair(params.prefix + "Median", labelStatisticsImageFilter->GetMedian(1))); featureList.push_back(std::make_pair(params.prefix + "Minimum", labelStatisticsImageFilter->GetMinimum(1))); featureList.push_back(std::make_pair(params.prefix + "Maximum", labelStatisticsImageFilter->GetMaximum(1))); featureList.push_back(std::make_pair(params.prefix + "Range", range)); featureList.push_back(std::make_pair(params.prefix + "Mean Absolute Deviation", mean_absolut_deviation)); featureList.push_back(std::make_pair(params.prefix + "Robust Mean Absolute Deviation", robustMeanAbsoluteDeviation)); featureList.push_back(std::make_pair(params.prefix + "Median Absolute Deviation", median_absolut_deviation)); featureList.push_back(std::make_pair(params.prefix + "Coefficient Of Variation", coefficient_of_variation)); featureList.push_back(std::make_pair(params.prefix + "Quantile Coefficient Of Dispersion", quantile_coefficient_of_dispersion)); featureList.push_back(std::make_pair(params.prefix + "Energy", squared_sum)); featureList.push_back(std::make_pair(params.prefix + "Root Mean Square", rms)); typename HistogramType::MeasurementVectorType mv(1); mv[0] = 0; typename HistogramType::IndexType resultingIndex; histogram->GetIndex(mv, resultingIndex); featureList.push_back(std::make_pair(params.prefix + "Robust Mean", meanRobust)); featureList.push_back(std::make_pair(params.prefix + "Uniformity", uniformity)); featureList.push_back(std::make_pair(params.prefix + "Entropy", entropy)); featureList.push_back(std::make_pair(params.prefix + "Excess Kurtosis", kurtosis - 3)); featureList.push_back(std::make_pair(params.prefix + "Covered Image Intensity Range", coveredGrayValueRange)); featureList.push_back(std::make_pair(params.prefix + "Sum", labelStatisticsImageFilter->GetSum(1))); featureList.push_back(std::make_pair(params.prefix + "Mode", mode_bin)); featureList.push_back(std::make_pair(params.prefix + "Mode Probability", mode_value)); featureList.push_back(std::make_pair(params.prefix + "Unbiased Standard deviation", labelStatisticsImageFilter->GetSigma(1))); featureList.push_back(std::make_pair(params.prefix + "Biased Standard deviation", sqrt(variance))); featureList.push_back(std::make_pair(params.prefix + "Number Of Voxels", labelStatisticsImageFilter->GetCount(1))); featureList.push_back(std::make_pair(params.prefix + "05th Percentile", p05th)); featureList.push_back(std::make_pair(params.prefix + "10th Percentile", p10th)); featureList.push_back(std::make_pair(params.prefix + "15th Percentile", p15th)); featureList.push_back(std::make_pair(params.prefix + "20th Percentile", p20th)); featureList.push_back(std::make_pair(params.prefix + "25th Percentile", p25th)); featureList.push_back(std::make_pair(params.prefix + "30th Percentile", p30th)); featureList.push_back(std::make_pair(params.prefix + "35th Percentile", p35th)); featureList.push_back(std::make_pair(params.prefix + "40th Percentile", p40th)); featureList.push_back(std::make_pair(params.prefix + "45th Percentile", p45th)); featureList.push_back(std::make_pair(params.prefix + "50th Percentile", p50th)); featureList.push_back(std::make_pair(params.prefix + "55th Percentile", p55th)); featureList.push_back(std::make_pair(params.prefix + "60th Percentile", p60th)); featureList.push_back(std::make_pair(params.prefix + "65th Percentile", p65th)); featureList.push_back(std::make_pair(params.prefix + "70th Percentile", p70th)); featureList.push_back(std::make_pair(params.prefix + "75th Percentile", p75th)); featureList.push_back(std::make_pair(params.prefix + "80th Percentile", p80th)); featureList.push_back(std::make_pair(params.prefix + "85th Percentile", p85th)); featureList.push_back(std::make_pair(params.prefix + "90th Percentile", p90th)); featureList.push_back(std::make_pair(params.prefix + "95th Percentile", p95th)); featureList.push_back(std::make_pair(params.prefix + "Interquartile Range", (p75th - p25th))); featureList.push_back(std::make_pair(params.prefix + "Image Dimension", VImageDimension)); featureList.push_back(std::make_pair(params.prefix + "Voxel Space", voxelSpace)); featureList.push_back(std::make_pair(params.prefix + "Voxel Volume", voxelVolume)); */ } mitk::GIFFirstOrderNumericStatistics::GIFFirstOrderNumericStatistics() { SetShortName("fon"); SetLongName("first-order-numeric"); SetFeatureClassName("First Order Numeric"); } mitk::GIFFirstOrderNumericStatistics::FeatureListType mitk::GIFFirstOrderNumericStatistics::CalculateFeatures(const Image::Pointer & image, const Image::Pointer &mask) { InitializeQuantifier(image, mask); FeatureListType featureList; FirstOrderNumericParameterStruct params; params.quantifier = GetQuantifier(); params.MinimumIntensity = GetQuantifier()->GetMinimum(); params.MaximumIntensity = GetQuantifier()->GetMaximum(); params.Bins = GetQuantifier()->GetBins(); params.prefix = FeatureDescriptionPrefix(); AccessByItk_3(image, CalculateFirstOrderStatistics, mask, featureList, params); return featureList; } mitk::GIFFirstOrderNumericStatistics::FeatureNameListType mitk::GIFFirstOrderNumericStatistics::GetFeatureNames() { FeatureNameListType featureList; return featureList; } void mitk::GIFFirstOrderNumericStatistics::AddArguments(mitkCommandLineParser &parser) { std::string name = GetOptionPrefix(); parser.addArgument(GetLongName(), name, mitkCommandLineParser::Bool, "Use First Order Statistic (Numeric)", "calculates First Order Statistic (Numeric)", us::Any()); AddQuantifierArguments(parser); } void mitk::GIFFirstOrderNumericStatistics::CalculateFeaturesUsingParameters(const Image::Pointer & feature, const Image::Pointer &, const Image::Pointer &maskNoNAN, FeatureListType &featureList) { auto parsedArgs = GetParameter(); if (parsedArgs.count(GetLongName())) { InitializeQuantifierFromParameters(feature, maskNoNAN); MITK_INFO << "Start calculating first order features ...."; auto localResults = this->CalculateFeatures(feature, maskNoNAN); featureList.insert(featureList.end(), localResults.begin(), localResults.end()); MITK_INFO << "Finished calculating first order features...."; } } std::string mitk::GIFFirstOrderNumericStatistics::GetCurrentFeatureEncoding() { return QuantifierParameterString(); } \ No newline at end of file