Error in branchless distance driven

[JeffFessler]

Thanks for providing branchless distance driven code!
The version posted works for "almost parallel beam" geometry
but if we change the geometry parameters to be more like a typical fan-beam geometry
then the code has errors in a while loop.

OpenCodes/Tomography/Computed Tomography (CT)/DistanceDrive/2D/BranchlessDistanceDriven.m

Line 45 in e4f0dcb

geo.DSO = 99700; % Distance from source to iso-center

Specifically in this while loop

OpenCodes/Tomography/Computed Tomography (CT)/DistanceDrive/2D/BranchlessDistanceDriven.m

Line 230 in e4f0dcb

while(detm(detInd+c2)<rowm(end))

we get indexing past the end of one of the arrays. Any ideas how to fix?

[rodrigovimieiro]

Hello @JeffFessler,

Thanks for posting this issue.

Do you have a code or geometry to reproduce this error?

I have tested the Branchless Distance Driven code for cone-beam geometry, but not for fan-beam. This is the implementation for cone-beam, which works fine.

[JeffFessler]

Thanks for asking.
All one has to do is change geo.DSO = 99700 to geo.DSO = 70000 so that it is more magnification and it causes an Index error.

I had to change the suffix from .m to .txt to upload the exact file.

ddbbug.txt

[rodrigovimieiro]

I guess it is related with the crazy geometry I have created. When you changed the geo.DSO, for certain angles, the object was being projected outside the detector, causing the error. I am not familiar with CT geometries, so years ago I created that one (as you mentioned "almost parallel beam") just for testing.

Now, I have modified to a geometry which I think might be more realistic and maybe close to the fan-beam. I am still not sure.

At least it gives some results:

This is the modified file:

ddbbug_edited.txt

See if it works for you.

[JeffFessler]

That CT detector geometry is quite realistic. However, there are two problems now.

If you look at the sinogram, you can see that there are horizontal bands every 45 or 90 degrees. It should not look like that - so there must be some bug(s) in the code.

And in practice it is usually the case that parts of the image (especially the corners) will project outside of the detector and we have to handle that carefully in the code.

If you are still working with this code and plan to fix it, please let us know.
We were converting this code to Julia and discovered the issue.
JuliaImageRecon/Sinograms.jl#9
If you fix it in Matlab, then we will translate that fix to Julia.
If you are not working with the 2D code anymore then I guess we will go back to the original papers and develop Julia code from there. Thanks.

(cc'ing @SoniaMinseoKim)

[rodrigovimieiro]

I see,

I think it is somehow related to the size of the detector mapped on the X axis, see:

So, in #L231 it makes the sonogram darker at certain angles:

OpenCodes/Tomography/Computed Tomography (CT)/DistanceDrive/2D/BranchlessDistanceDriven.m

Line 231 in 16d062d

sinoTmp(detInd) = sinoTmp(detInd)+(Pdk(detInd+c2)-Pdk(detInd))/deltaDetm;

There is also the fact that the mapped detector size changes depending on the detector element position. I would need to go back to the paper to see how to fix it.

Unfortunately, I am not working with 2D anymore. I will post this as a known issue. Let me know if you guys find out.

[rodrigovimieiro]

Hello @JeffFessler ,

Were you able to figure out the problem?

Thanks!

[JeffFessler]

we ended up doing a pretty big rewrite:
https://github.com/JuliaImageRecon/Sinograms.jl/blob/main/src/sys2/bdd_2d.jl
it works, but still needs further refinement:
https://juliaimagerecon.github.io/Sinograms.jl/dev/generated/examples/08-bdd2d