Slope and intercept when reading Nifti images

Elli · August 5, 2019, 2:41pm

Dear all,

I have a very basic question: I am reading in nifti images using:

ImageType::Pointer readImage(string imageName) {
typename ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName(imageName);
try {
reader->Update();
}
catch (itk::ExceptionObject &excp) {
std::cerr << excp << std::endl;
}
ImageType::Pointer image = reader->GetOutput();
return reader->GetOutput();
}

Am I correct that the ITK image reader is applying the slope and intercept automatically? Or do I have to manually multiply by the slope and add the intercept to get the correct values?

I am asking because I have some images where the results are reasonable, but I have one image where the image values are not making sense after reading it. When I apply then slope and intercept, the values are making sense… (this file can be downloaded here: https://wetransfer.com/downloads/3aa534c3139492a4687df079bd1bde0120190805143844/4ff1ff5b3944d7bfa2b2c40c798ae40820190805143844/b82e99)

Thanks a lot for your answer.

Elli

matt.mccormick · August 9, 2019, 3:46am

Hi Elli,

Yes, rescale / slope and intercept is applied with NIFTI files per:

github.com

InsightSoftwareConsortium/ITK/blob/f162460efc347025d063634153b4dce4db6c3937/Modules/IO/NIFTI/src/itkNiftiImageIO.cxx#L714-L725


// If the scl_slope field is nonzero, then rescale each voxel value in the
// dataset.
// Complete description of can be found in nifti1.h under "DATA SCALING"
if ( this->MustRescale() )
  {
  switch ( this->m_ComponentType )
    {
    case CHAR:
      RescaleFunction(static_cast< char * >( buffer ),
                      this->m_RescaleSlope,
                      this->m_RescaleIntercept, numElts);
      break;

HTH,
Matt

Elli · August 10, 2019, 9:41am

Dear Matt,

thanks for your answer…

But why is it not working for this specific image ? (you can find it here:
https://we.tl/t-ZKbnxnTWwr

)

All other softwares I used for comparison (Matlab, nibabel in python, etc) can read the image correctly. But when I read it with ITK, i get as maximum intensity 32767 as if the image would not have been scaled. When I scale it then manually by mutliplying with slope, I get the correct values. How can that be?

Thanks a lot for your help,

Elli

dzenanz · August 13, 2019, 3:55pm

When I try to download the offending image, I get “Transfer deleted”. Has the the link expired? Can you refresh it?

Elli · August 13, 2019, 4:15pm

Sure, sorry. Here is a new link: https://wetransfer.com/downloads/603bc96030ba5dfea33032524743e07b20190813161307/2586576f53a7554bfd45b001747119a320190813161307/e6f5d9

thanks in advance for looking into this!

dzenanz · August 14, 2019, 2:30pm

The slope in this image is ignored because both SForm and QForm codes are zero:

github.com

InsightSoftwareConsortium/ITK/blob/f162460efc347025d063634153b4dce4db6c3937/Modules/IO/NIFTI/src/itkNiftiImageIO.cxx#L1210-L1224


if ( this->m_NiftiImage->qform_code == 0
     && this->m_NiftiImage->sform_code == 0 )
  {
  this->m_RescaleSlope = 1;
  this->m_RescaleIntercept = 0;
  }
else
  {
  this->m_RescaleSlope = this->m_NiftiImage->scl_slope;
  if ( std::abs( this->m_RescaleSlope ) < NumericTraits<double>::epsilon() )
    {
    this->m_RescaleSlope = 1;
    }
  this->m_RescaleIntercept = this->m_NiftiImage->scl_inter;
  }

Elli · August 15, 2019, 11:49am

Hey Dzenan,

thanks! But officially that is not correct, right? Because I read that the scaling with slope and intercept should always be applied also when sform and qform are 0…

Anyhow, I can make a workaround in my code. So thanks a lot for the information!

Regards

Elli

dzenanz · August 15, 2019, 12:45pm

Where is the official answer to this question? Also, you could dig a little through git history of that file to see when changes were made, and maybe also why.

Elli · August 15, 2019, 2:39pm

Thanks, thats a good idea.

Well, officially is a bit difficult to define. We just had a big discussion in our group whats the right way to handle nifti images… And what we found online in the header descriptions was that scaling should always be applied. Anyhow, I will check the history! Thank you!