Class MultiLabelSTAPLEImageFilter


public class MultiLabelSTAPLEImageFilter extends ImageFilter
This filter performs a pixelwise combination of an arbitrary number of input images, where each of them represents a segmentation of the same scene (i.e., image). The labelings in the images are weighted relative to each other based on their "performance" as estimated by an expectation-maximization algorithm. In the process, a ground truth segmentation is estimated, and the estimated performances of the individual segmentations are relative to this estimated ground truth. The algorithm is based on the binary STAPLE algorithm by Warfield et al. as published originally in S. Warfield, K. Zou, W. Wells, "Validation of image segmentation and expert quality with an expectation-maximization algorithm" in MICCAI 2002: Fifth International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer-Verlag, Heidelberg, Germany, 2002, pp. 298-306 The multi-label algorithm implemented here is described in detail in T. Rohlfing, D. B. Russakoff, and C. R. Maurer, Jr., "Performance- based classifier combination in atlas-based image segmentation using expectation-maximization parameter estimation," IEEE Transactions on Medical Imaging, vol. 23, pp. 983-994, Aug. 2004. INPUTS All input volumes to this filter must be segmentations of an image, that is, they must have discrete pixel values where each value represents a different segmented object. Input volumes must all contain the same size RequestedRegions. Not all input images must contain all possible labels, but all label values must have the same meaning in all images. The filter can optionally be provided with estimates for the a priori class probabilities through the SetPriorProbabilities function. If no estimate is provided, one is automatically generated by analyzing the relative frequencies of the labels in the input images. OUTPUTS The filter produces a single output volume. Each output pixel contains the label that has the highest probability of being the correct label, based on the performance models of the individual segmentations. If the maximum probability is not unique, i.e., if more than one label have a maximum probability, then an "undecided" label is assigned to that output pixel. By default, the label used for undecided pixels is the maximum label value used in the input images plus one. Since it is possible for an image with 8 bit pixel values to use all 256 possible label values, it is permissible to combine 8 bit (i.e., byte) images into a 16 bit (i.e., short) output image. In addition to the combined image, the estimated confusion matrices for each of the input segmentations can be obtained through the GetConfusionMatrix member function. PARAMETERS The label used for "undecided" labels can be set using SetLabelForUndecidedPixels. This functionality can be unset by calling UnsetLabelForUndecidedPixels. A termination threshold for the EM iteration can be defined by calling SetTerminationUpdateThreshold. The iteration terminates once no single parameter of any confusion matrix changes by less than this threshold. Alternatively, a maximum number of iterations can be specified by calling SetMaximumNumberOfIterations. The algorithm may still terminate after a smaller number of iterations if the termination threshold criterion is satisfied. EVENTS This filter invokes IterationEvent() at each iteration of the E-M algorithm. Setting the AbortGenerateData() flag will cause the algorithm to halt after the current iteration and produce results just as if it had converged. The algorithm makes no attempt to report its progress since the number of iterations needed cannot be known in advance. Torsten Rohlfing, SRI International, Neuroscience Program See: itk::simple::MultiLabelSTAPLE for the procedural interface C++ includes: sitkMultiLabelSTAPLEImageFilter.h
  • Constructor Details

    • MultiLabelSTAPLEImageFilter

      protected MultiLabelSTAPLEImageFilter(long cPtr, boolean cMemoryOwn)
    • MultiLabelSTAPLEImageFilter

      public MultiLabelSTAPLEImageFilter()
      itk::simple::MultiLabelSTAPLEImageFilter::MultiLabelSTAPLEImageFilter() Default Constructor that takes no arguments and initializes default parameters
  • Method Details

    • getCPtr

      protected static long getCPtr(MultiLabelSTAPLEImageFilter obj)
    • swigRelease

      protected static long swigRelease(MultiLabelSTAPLEImageFilter obj)
    • finalize

      protected void finalize()
      Overrides:
      finalize in class ImageFilter
    • delete

      public void delete()
      virtual itk::simple::MultiLabelSTAPLEImageFilter::~MultiLabelSTAPLEImageFilter() Destructor
      Overrides:
      delete in class ImageFilter
    • setLabelForUndecidedPixels

      public void setLabelForUndecidedPixels(long LabelForUndecidedPixels)
      Self& itk::simple::MultiLabelSTAPLEImageFilter::SetLabelForUndecidedPixels(uint64_t LabelForUndecidedPixels) Set label value for undecided pixels.
    • getLabelForUndecidedPixels

      public long getLabelForUndecidedPixels()
      uint64_t itk::simple::MultiLabelSTAPLEImageFilter::GetLabelForUndecidedPixels() const Get label value used for undecided pixels. After updating the filter, this function returns the actual label value used for undecided pixels in the current output. Note that this value is overwritten when SetLabelForUndecidedPixels is called and the new value only becomes effective upon the next filter update.
    • setTerminationUpdateThreshold

      public void setTerminationUpdateThreshold(float TerminationUpdateThreshold)
      Self& itk::simple::MultiLabelSTAPLEImageFilter::SetTerminationUpdateThreshold(float TerminationUpdateThreshold) Set termination threshold based on confusion matrix parameter updates.
    • getTerminationUpdateThreshold

      public float getTerminationUpdateThreshold()
      float itk::simple::MultiLabelSTAPLEImageFilter::GetTerminationUpdateThreshold() const Set termination threshold based on confusion matrix parameter updates.
    • setMaximumNumberOfIterations

      public void setMaximumNumberOfIterations(long MaximumNumberOfIterations)
      Self& itk::simple::MultiLabelSTAPLEImageFilter::SetMaximumNumberOfIterations(unsigned int MaximumNumberOfIterations) Set maximum number of iterations.
    • getMaximumNumberOfIterations

      public long getMaximumNumberOfIterations()
      unsigned int itk::simple::MultiLabelSTAPLEImageFilter::GetMaximumNumberOfIterations() const Set maximum number of iterations.
    • setPriorProbabilities

      public void setPriorProbabilities(VectorFloat PriorProbabilities)
      Self& itk::simple::MultiLabelSTAPLEImageFilter::SetPriorProbabilities(std::vector< float > PriorProbabilities) Set manual estimates for the a priori class probabilities. The size of the array must be greater than the value of the largest label. The index into the array corresponds to the label value in the segmented image for the class.
    • getPriorProbabilities

      public VectorFloat getPriorProbabilities()
      std::vector<float> itk::simple::MultiLabelSTAPLEImageFilter::GetPriorProbabilities() const Get prior class probabilities. After updating the filter, this function returns the actual prior class probabilities. If these were not previously set by a call to SetPriorProbabilities, then they are estimated from the input segmentations and the result is available through this function.
    • getConfusionMatrix

      public VectorFloat getConfusionMatrix(long input)
      std::vector<float> itk::simple::MultiLabelSTAPLEImageFilter::GetConfusionMatrix(unsigned int input) const Get confusion matrix for the i-th input segmentation. This is an active measurement. It may be accessed while the filter is being executing in command call-backs and can be accessed after execution.
    • getName

      public String getName()
      std::string itk::simple::MultiLabelSTAPLEImageFilter::GetName() const Name of this class
      Overrides:
      getName in class ProcessObject
    • toString

      public String toString()
      std::string itk::simple::MultiLabelSTAPLEImageFilter::ToString() const Print ourselves out
      Overrides:
      toString in class ProcessObject
    • execute

      public Image execute(VectorOfImage images)
      Image itk::simple::MultiLabelSTAPLEImageFilter::Execute(const std::vector< Image > &images) Execute the filter on the input images
    • execute

      public Image execute(Image image1)
      Image itk::simple::MultiLabelSTAPLEImageFilter::Execute(const std::vector< Image > &images) Execute the filter on the input images
    • execute

      public Image execute(Image image1, Image image2)
      Image itk::simple::MultiLabelSTAPLEImageFilter::Execute(const std::vector< Image > &images) Execute the filter on the input images
    • execute

      public Image execute(Image image1, Image image2, Image image3)
      Image itk::simple::MultiLabelSTAPLEImageFilter::Execute(const std::vector< Image > &images) Execute the filter on the input images
    • execute

      public Image execute(Image image1, Image image2, Image image3, Image image4)
      Image itk::simple::MultiLabelSTAPLEImageFilter::Execute(const std::vector< Image > &images) Execute the filter on the input images
    • execute

      public Image execute(Image image1, Image image2, Image image3, Image image4, Image image5)
      Image itk::simple::MultiLabelSTAPLEImageFilter::Execute(const std::vector< Image > &images) Execute the filter on the input images