/*========================================================================= * * Copyright NumFOCUS * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * https://www.apache.org/licenses/LICENSE-2.0.txt * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * *=========================================================================*/ #ifndef itkMattesMutualInformationImageToImageMetricv4_h #define itkMattesMutualInformationImageToImageMetricv4_h #include "itkImageToImageMetricv4.h" #include "itkMattesMutualInformationImageToImageMetricv4GetValueAndDerivativeThreader.h" #include "itkPoint.h" #include "itkIndex.h" #include "itkBSplineDerivativeKernelFunction.h" #include "itkArray2D.h" #include "itkThreadedIndexedContainerPartitioner.h" #include namespace itk { /** \class MattesMutualInformationImageToImageMetricv4 * * \brief Computes the mutual information between two images to be * registered using the method of Mattes et al. * * MattesMutualInformationImageToImageMetric computes the mutual * information between a fixed and moving image to be registered. * * This class is templated over the FixedImage type and the MovingImage * type. * * The calculations are based on the method of Mattes et al [1,2] * where the probability density distribution are estimated using * Parzen histograms. Since the fixed image PDF does not contribute * to the derivatives, it does not need to be smooth. Hence, * a zero order (box car) BSpline kernel is used * for the fixed image intensity PDF. On the other hand, to ensure * smoothness a third order BSpline kernel is used for the * moving image intensity PDF. * * During each call of GetValue(), GetDerivatives(), * GetValueAndDerivatives(), marginal and joint intensity PDF's * values are estimated at discrete position or bins. * The number of bins used can be set via SetNumberOfHistogramBins(). * To handle data with arbitrary magnitude and dynamic range, * the image intensity is scaled such that any contribution to the * histogram will fall into a valid bin. * * Once the PDF's have been constructed, the mutual information * is obtained by double summing over the discrete PDF values. * * \warning Local-support transforms are not yet supported. If used, * an exception is thrown during Initialize(). * * \note The per-iteration post-processing code is not multi-threaded, but could be * readily be made so for a small performance gain. * See GetValueCommonAfterThreadedExecution(), GetValueAndDerivative() * and threader::AfterThreadedExecution(). * * The algorithm and much of the code was copied from the previous * Mattes MI metric, i.e. itkMattesMutualInformationImageToImageMetric. * * See * MattesMutualInformationImageToImageMetricv4GetValueAndDerivativeThreader::ProcessPoint * for portions of the algorithm implementation. * * See ImageToImageMetricv4 for details of common metric operation and options. * * References: * [1] "Nonrigid multimodality image registration" * D. Mattes, D. R. Haynor, H. Vesselle, T. Lewellen and W. Eubank * Medical Imaging 2001: Image Processing, 2001, pp. 1609-1620. * [2] "PET-CT Image Registration in the Chest Using Free-form Deformations" * D. Mattes, D. R. Haynor, H. Vesselle, T. Lewellen and W. Eubank * IEEE Transactions in Medical Imaging. Vol.22, No.1, January 2003. pp.120-128. * [3] "Optimization of Mutual Information for MultiResolution Image * Registration" * P. Thevenaz and M. Unser * IEEE Transactions in Image Processing, 9(12) December 2000. * * \sa itkImageToImageMetricv4 * \ingroup ITKMetricsv4 */ template > class ITK_TEMPLATE_EXPORT MattesMutualInformationImageToImageMetricv4 : public ImageToImageMetricv4 { public: ITK_DISALLOW_COPY_AND_MOVE(MattesMutualInformationImageToImageMetricv4); /** Standard class type aliases. */ using Self = MattesMutualInformationImageToImageMetricv4; using Superclass = ImageToImageMetricv4; using Pointer = SmartPointer; using ConstPointer = SmartPointer; /** Method for creation through the object factory. */ itkNewMacro(Self); /** \see LightObject::GetNameOfClass() */ itkOverrideGetNameOfClassMacro(MattesMutualInformationImageToImageMetricv4); /** Superclass types */ using typename Superclass::MeasureType; using typename Superclass::DerivativeType; using DerivativeValueType = typename DerivativeType::ValueType; using typename Superclass::FixedImageType; using typename Superclass::FixedImagePointType; using typename Superclass::FixedImageIndexType; using typename Superclass::FixedImagePixelType; using typename Superclass::FixedImageGradientType; using typename Superclass::MovingImagePointType; using typename Superclass::MovingImagePixelType; using typename Superclass::MovingImageGradientType; using typename Superclass::MovingTransformType; using typename Superclass::JacobianType; using VirtualImageType = typename Superclass::VirtualImageType; using typename Superclass::VirtualIndexType; using typename Superclass::VirtualPointType; using typename Superclass::VirtualPointSetType; /** Types inherited from Superclass. */ using typename Superclass::FixedSampledPointSetPointer; /* Image dimension accessors */ static constexpr typename TVirtualImage::ImageDimensionType VirtualImageDimension = TVirtualImage::ImageDimension; static constexpr typename TFixedImage::ImageDimensionType FixedImageDimension = TFixedImage::ImageDimension; static constexpr typename TMovingImage::ImageDimensionType MovingImageDimension = TMovingImage::ImageDimension; /** Number of bins to used in the histogram. Typical value is * 50. The minimum value is 5 due to the padding required by the Parzen * windowing with a cubic-BSpline kernel. Note that even if the metric * is used on binary images, the number of bins should at least be * equal to five. */ itkSetClampMacro(NumberOfHistogramBins, SizeValueType, 5, NumericTraits::max()); itkGetConstReferenceMacro(NumberOfHistogramBins, SizeValueType); void Initialize() override; /** The marginal PDFs are stored as std::vector. */ // NOTE: floating point precision is not as stable. // Double precision proves faster and more robust in real-world testing. using PDFValueType = TInternalComputationValueType; /** Typedef for the joint PDF and PDF derivatives are stored as ITK Images. */ using JointPDFType = Image; using JointPDFDerivativesType = Image; /** * Get the internal JointPDF image that was used in * creating the metric value. */ const typename JointPDFType::Pointer GetJointPDF() const { if (this->m_ThreaderJointPDF.empty()) { return typename JointPDFType::Pointer(nullptr); } return this->m_ThreaderJointPDF[0]; } /** * Get the internal JointPDFDeriviative image that was used in * creating the metric derivative value. * This is only created when a global support transform is used, and * derivatives are requested. */ const typename JointPDFDerivativesType::Pointer GetJointPDFDerivatives() const { return this->m_JointPDFDerivatives; } void FinalizeThread(const ThreadIdType threadId) override; protected: MattesMutualInformationImageToImageMetricv4(); ~MattesMutualInformationImageToImageMetricv4() override = default; friend class MattesMutualInformationImageToImageMetricv4GetValueAndDerivativeThreader< ThreadedImageRegionPartitioner, Superclass, Self>; friend class MattesMutualInformationImageToImageMetricv4GetValueAndDerivativeThreader< ThreadedIndexedContainerPartitioner, Superclass, Self>; using MattesMutualInformationDenseGetValueAndDerivativeThreaderType = MattesMutualInformationImageToImageMetricv4GetValueAndDerivativeThreader< ThreadedImageRegionPartitioner, Superclass, Self>; using MattesMutualInformationSparseGetValueAndDerivativeThreaderType = MattesMutualInformationImageToImageMetricv4GetValueAndDerivativeThreader; void PrintSelf(std::ostream & os, Indent indent) const override; using JointPDFIndexType = typename JointPDFType::IndexType; using JointPDFValueType = typename JointPDFType::PixelType; using JointPDFRegionType = typename JointPDFType::RegionType; using JointPDFSizeType = typename JointPDFType::SizeType; using JointPDFDerivativesIndexType = typename JointPDFDerivativesType::IndexType; using JointPDFDerivativesValueType = typename JointPDFDerivativesType::PixelType; using JointPDFDerivativesRegionType = typename JointPDFDerivativesType::RegionType; using JointPDFDerivativesSizeType = typename JointPDFDerivativesType::SizeType; /** Typedefs for BSpline kernel and derivative functions. */ using CubicBSplineFunctionType = BSplineKernelFunction<3, PDFValueType>; using CubicBSplineDerivativeFunctionType = BSplineDerivativeKernelFunction<3, PDFValueType>; /** Post-processing code common to both GetValue * and GetValueAndDerivative. */ virtual void GetValueCommonAfterThreadedExecution(); OffsetValueType ComputeSingleFixedImageParzenWindowIndex(const FixedImagePixelType & value) const; /** Variables to define the marginal and joint histograms. */ SizeValueType m_NumberOfHistogramBins{ 50 }; PDFValueType m_MovingImageNormalizedMin{}; PDFValueType m_FixedImageNormalizedMin{}; PDFValueType m_FixedImageTrueMin{}; PDFValueType m_FixedImageTrueMax{}; PDFValueType m_MovingImageTrueMin{}; PDFValueType m_MovingImageTrueMax{}; PDFValueType m_FixedImageBinSize{}; PDFValueType m_MovingImageBinSize{}; /** Helper array for storing the values of the JointPDF ratios. */ using PRatioType = PDFValueType; using PRatioArrayType = std::vector; mutable PRatioArrayType m_PRatioArray{}; /** Helper array for storing per-parameter linearized index to * retrieve the pRatio during evaluation with local-support transform. */ mutable std::vector m_JointPdfIndex1DArray{}; /** The moving image marginal PDF. */ mutable std::vector m_MovingImageMarginalPDF{}; mutable std::vector> m_ThreaderFixedImageMarginalPDF{}; /** The joint PDF and PDF derivatives. */ typename std::vector m_ThreaderJointPDF{}; /* \class DerivativeBufferManager * A helper class to manage complexities of minimizing memory * needs for mattes mutual information derivative computations * per thread. * * Thread safety note: * A separate object is used locally per each thread. Only the members * m_ParentJointPDFDerivativesMutexPtr and m_ParentJointPDFDerivatives * are shared between threads and access to m_ParentJointPDFDerivatives * is controlled with the m_ParentJointPDFDerivativesMutexPtr mutex lock. * \ingroup ITKMetricsv4 */ class DerivativeBufferManager { using Self = DerivativeBufferManager; public: /* All these methods are thread safe except ReduceBuffer */ void Initialize(size_t maxBufferLength, const size_t cachedNumberOfLocalParameters, std::mutex * parentDerivativeMutexPtr, typename JointPDFDerivativesType::Pointer parentJointPDFDerivatives); void DoubleBufferSize(); DerivativeBufferManager() : m_MemoryBlock(0) {} ~DerivativeBufferManager() = default; size_t GetCachedNumberOfLocalParameters() const { return this->m_CachedNumberOfLocalParameters; } /** * Attempt to dump the buffer if it is full. * If the attempt to acquire the lock fails, double the buffer size and try again. */ void CheckAndReduceIfNecessary(); /** * Force the buffer to dump by blocking. */ void BlockAndReduce(); // If offset is same as previous offset, then accumulate with previous PDFValueType * GetNextElementAndAddOffset(const OffsetValueType & offset) { m_BufferOffsetContainer[m_CurrentFillSize] = offset; PDFValueType * PDFBufferForWriting = m_BufferPDFValuesContainer[m_CurrentFillSize]; ++m_CurrentFillSize; return PDFBufferForWriting; } /** * Apply the operations stored in the buffer. * This method is not thread safe and requires a lock while threading. */ void ReduceBuffer(); private: // How many AccumulatorElements used size_t m_CurrentFillSize{ 0 }; // Continguous chunk of memory for efficiency std::vector m_MemoryBlock; // The (number of lines in the buffer) * (cells per line) size_t m_MemoryBlockSize; std::vector m_BufferPDFValuesContainer; std::vector m_BufferOffsetContainer; size_t m_CachedNumberOfLocalParameters; size_t m_MaxBufferSize; // Pointer handle to parent version std::mutex * m_ParentJointPDFDerivativesMutexPtr; // Smart pointer handle to parent version typename JointPDFDerivativesType::Pointer m_ParentJointPDFDerivatives; }; std::vector m_ThreaderDerivativeManager{}; std::mutex m_JointPDFDerivativesLock{}; typename JointPDFDerivativesType::Pointer m_JointPDFDerivatives{}; PDFValueType m_JointPDFSum{}; /** Store the per-point local derivative result by parzen window bin. * For local-support transforms only. */ mutable std::vector m_LocalDerivativeByParzenBin{}; private: /** Perform the final step in computing results */ virtual void ComputeResults() const; }; } // end namespace itk #ifndef ITK_MANUAL_INSTANTIATION # include "itkMattesMutualInformationImageToImageMetricv4.hxx" #endif #endif