Hi

I have tried to reproduce Chris' results here. When I compare the images with and without tilt correction using SPM Check Reg, the shear appears to be correctly applied to both, but the uncorrected one looks longer anterior to posterior. The screenshot is from the open CQ500 dataset, but my concern is with my local hospital data, which has a similar effect.



I would like to know if there's a fix for this, either to apply to the NIfTI header, or to fix something that is wrong with the DICOM metadata. I have no idea if the issue is with my version of SPM (SPM12 r7771, which may be different to what Chris used in Dec 2018) or something to do with my image metadata.

I can change the voxel scaling in SPM to get the images looking more similar, but not knowing my matrix maths as well as Chris, I would not want to apply this without expert reassurance.

Here's as much info as I can think to include from an example from my local hospital set.

Manufacturer / Model: GE Optima CT660.
DICOM geometry: 512 x 512 x 256 matrix with 0.449219 x 0.449219 x 0.625 mm voxels.
DICOM gantry tilt (0018,1120): 20.5 (dcm2niix estimates the same value from slice vector).

Results of spm_get_space (I tried to use code tags to line up the numbers but it looks like the main page doesn't render them, despite the editor doing so):
Uncorrected
Corrected

Results from spm_imatrix:


Geometry from Check Reg:

Opening in SPM Display gives similar voxel size, orientation, and direction cosine, and says "Warning shears involved" for the uncorrected but not the corrected, which I'd expect.

Each volume that I checked has 8 acquisition numbers, each with 32 slices, which I guess are acquired with a pause between them as you can sometimes see slight discontinuities in the image about 32 slices apart when the patient moves. So maybe this acquisition protocol leads to the unexpected results (it is the same for CQ500 data).

Other than that, I couldn't see the effect in another volume with a smaller gantry tilt (e.g. 4 degrees) but that might be just because it was harder to see.

I'm happy to go digging for more info in the DICOM metadata or SPM if requested, and appreciate any help you can offer. 

Best wishes
Paul Wright
King's College London

Well spotted. I have created a github issue to describe this behavior and my solution:
  https://github.com/rordenlab/dcm2niix/is...
Please test my solution, it simply negates the shear. I suspect there are more complex but generalized matrix multiplication solutions, but they are somewhat complicated by anisotropic voxels. For this reason, I hope my simple solution proves robust. Coming from MR, I have very little exposure to CT exemplars, so please validate my solution.