Ball Binary Morphological filter with different spacing?

chlee · July 30, 2022, 1:52am

Hi,

I want to generate a segmented image and want to do a 3D morphological closing operation just before the thresholding event. I already have a binary 3D image. The issue is that my 3D image x,y,z spacing is not same (x,y dimension is 0.41432 but z dimension is 1). Would it be OK to still OK to use a Ball structure with same radius? Or shall I first resample my image to match the z spacing to the x dimension and then treat the operation and then resample back?

Thank you very much in advance for any tips!

dzenanz · July 30, 2022, 2:31pm

You can have a ball structuring element with different radii along different axes, see docs. If you pass a scalar, you will get a “square” neighborhood.

Also consider approximation to morphological closing via dilating and eroding using thresholding of distance field. Sample code:

github.com

KitwareMedical/HASI/blob/v0.2.2/include/itkSegmentBonesInMicroCTFilter.hxx#L82-L112


      
          template <typename TInputImage, typename TOutputImage>
          typename TOutputImage::Pointer
          SegmentBonesInMicroCTFilter<TInputImage, TOutputImage>::SDFDilate(typename TOutputImage::Pointer labelImage,
                                                                            double                         radius)
          {
            typename FloatThresholdType::Pointer sdfTh = FloatThresholdType::New();
            sdfTh->SetInput(this->SDF(labelImage));
            sdfTh->SetUpperThreshold(radius * radius);
            sdfTh->Update();
            return sdfTh->GetOutput();
          }
          
          
template <typename TInputImage, typename TOutputImage>
          typename TOutputImage::Pointer
          SegmentBonesInMicroCTFilter<TInputImage, TOutputImage>::SDFErode(typename TOutputImage::Pointer labelImage,
                                                                           double                         radius)
          {
            // we need an inversion filter because Maurer's filter distances are not symmetrical
            // inside distances start at 0, while outside distances start at single spacing
            using NotType = NotImageFilter<TOutputImage, TOutputImage>;

This file has been truncated. show original

chlee · August 2, 2022, 7:55am

Thank you very much!! This is very helpful!!