[Insight-users] Image Intensities Changed During DICOM Read-Write-Read in ITK 3.20 vs. ITK 4.4.1

[Constantine Zakkaroff]

Hello ALL,

I'm wondering if anyone can explain what's going on with the changing 
image intensities in a read-write-read round-trip. I think it's a bug, 
but the issues.itk.org is unresponsive at the moment. Another bug? ;)

Anyhow, the strange thing is that the results for the round trip 
read-write-read test are very different for ITK 3.20 and ITK 4.4.1.

I use to StatisticsImageFilter to examine the intensity values. Here's 
what happens, when I simply read a DICOM image, print stats, save the 
image and read it again, followed by printing the stats:

ITK Version: 3.20
Stats for original file:
Min: 0, Max: 2474, Mean: 169.236, Sigma: 256.719, Variance: 65904.7, 
Sum: 1.40371e+007
Stats for saved file:
Min: 0, Max: 2471, Mean: 167.132, Sigma: 256.125, Variance: 65600.1, 
Sum: 1.38626e+007

ITK Version: 4.4
Stats for original file:
Min: 0, Max: 2474, Mean: 169.236, Sigma: 256.719, Variance: 65904.7, 
Sum: 1.40371e+007
Stats for saved file:
Min: 0, Max: 7446, Mean: 509.114, Sigma: 772.59, Variance: 596895, Sum: 
4.22279e+007

In both cases the code is identical. Just compiled against different 
versions of ITK. The images are read and written as itk::Image<unsigned 
short, 3>, with no processing, no changes to any aspect of the images.

So, ITK 3.20 introduces a small change, which I'm pretty sure shouldn't 
be there. But in ITK 4.4.1 the scale of intensities changes three-fold.

I'd not be surprised by a change of this magnitude if I was photocopying 
those images, but not when using digital software and hardware. :^) Is 
there something I'm not understanding here?

None of this happens with read(DICOM)-write(MHD)-read(MHD), where the 
original and round-trip values are identical:
Stats for saved file: file0.mhd
Min: 0, Max: 2474, Mean: 169.236, Sigma: 256.719, Variance: 65904.7, 
Sum: 1.40371e+007

Many thanks,
Constantine