Some Questions About FDK Reconstructed Images

[JinZhaoMu]

Dear Author,
Greetings. I have obtained reconstructed images using the FDK algorithm and have one questions to consult you: is the circular ring in the background the maximum reconstruction region of the cone beam, and can this ring be removed?

[AnderBiguri]

Comercial software often just cut/crops the circle. We have a function for that, cropCBCT, but its basically a circular binary mask.

For the comercial software doing better in corner slices: In microCT TIGRE genrally performs the same as the comercial software, but in medical CBCT machines there is a significant amount of pre and post processing done to the images, sometimes at the cost of resolution. This often improves visually the quality of the images, and may include propietary AI denoising algorithms

[JinZhaoMu]

Thank you very much for your reply. Actually, I am using TIGRE for geometric correction. The reason for the poor reconstruction effect between the upper and lower layers should be that the in-plane angle of the detector has not been calculated successfully. By the way, what does the sign of geo.rotDetector represent? Is clockwise positive and counterclockwise negative when viewed from the detector?

[AnderBiguri]

@JinZhaoMu You can have a look at the diagram in the TIGRE v3 paper, that should hopefully clarify all doubts about geometry. Let me know if it doesn't

[JinZhaoMu]

Okay, I have clarified several geometric parameters in Python that are prone to being mixed up in terms of order: the order of geo.nVoxel, geo.sVoxel and geo.dVoxel is ZYX, and the order of geo.rotDetector is roll, pitch, yaw. I have now applied the cropCBCT function, and the slice images look normal. However, the 3D rendered object is always surrounded by a cylindrical surface (in isosurface rendering), which puzzles me and I have no idea how to remove it.

[AnderBiguri]

@JinZhaoMu that will happen, it's the FOV of CBCT. Just zero it with a mask, it's what comercial software does.

[JinZhaoMu]

Ah, I see. The density of the object I scanned is too low and too close to air, so it's no wonder this happened. Thank you so much.

[sanchalitorpe-source]

Thank you for the clarification.

I confirmed that the cylindrical surface visible in 3D isosurface rendering corresponds to the CBCT field-of-view, not a reconstruction artifact. For low-density objects close to air, voxels outside the valid FOV are non-zero and can be picked up by low isovalue thresholds.

Commercial software appears to explicitly zero voxels outside the CBCT FOV. Applying a cylindrical FOV mask to the reconstructed volume (not just visualization cropping) removes this surface completely and matches commercial behavior.

I have opened a PR adding an optional utility function to apply this CBCT FOV mask in post-processing, so users can easily remove the cylindrical background when needed, without changing reconstruction defaults.

Thanks again for the helpful discussion.

[AnderBiguri]

Thanks @sanchalitorpe-source ! This function exists in MATLAB but never got around to make the python version, really appreaciated.

[sanchalitorpe-source]

Thanks! Happy to see it helps. Hopefully this makes post-processing in Python as convenient as in MATLAB.