Closed yunsuper123 closed 4 months ago
Hi @yunsuper123 ,
I don't see anything inherently wrong in the results you are showing, why are you worried?
You are reconstructing an image that is quite much bigger than the detector you are simulating (512 vs 128), so it will definitely not be very sharp. Also, you are only simulating 72 projections, which is quite low. The only thing I would say is that SIRT is quite slow converging as an algorithm, I would suggest more like 200 iterations than 20. Other algorithms like OS-SART or CLGS would produce results faster, within 20~50 iterations. I also recommend when exploring iterative algorithms to always compare against an FDK reconstruction, not only the ground truth, so you can see how the iterative algorithms perform compared to a standard clinical reconstruction.
On second look: are you worried about the craneo-caudal shape of the result image? That fully depends on your visualization right? I don't exactly remember how mha
files work, but should't it have some information on the slice thickness? Likely this is not being saved, and therefore when opened by whatever software you are using for visualization, it assumes that pixels are the same size in all directions, which are not.
Expected Behavior
After implementing SIRT, I expect the CT reconstruction volume to be the same as the ground truth CT.
Actual Behavior
I slightly adjusted demo 1,2,3 to create the geometry and data of a CT .mha file, then implemented SIRT to reconstruct CT volume with demo 7. Although I tried to make sure all the geometry variables were right, I still didn't get the right result shape. The first image is the ground truth CT shown in 3D slicer, and the second is my reconstruction. It's quite possible that I might be overlooking something, so I was wondering if you could shed some light on the potential issues of this inconsistency. Thank you in advance for your assistance and for the amazing contributions you've made to our community.
Code to reproduce the problem (If applicable)
Specifications