Artifact in CBCT Sart reconstruction

[springXIACJ]

Hello, when I use more than 180 projections (e.g., 250 projections) to reconstruct CBCT, severe streak artifacts appear.

When I used fewer projection views such as 100, I did not find this bar artifact.

Geometry

geo.DSD = 1502.19 # Distance Source Detector (mm)
geo.DSO = 1161.91 # Distance Source Origin (mm)
geo.nDetector = np.array([1536, 1536]) # number of pixels (px)
geo.dDetector = np.array([ 0.278, 0.278]) # size of each pixel (mm)
geo.sDetector = geo.nDetector * geo.dDetector # total size of the detector (mm)
geo.nVoxel = np.array([512,512,512]) # number of voxels (vx)
geo.dVoxel = np.array([0.6, 0.6, 0.6]) # size of each voxel
geo.sVoxel =geo.nVoxel * geo.dVoxel # total size of the image (mm)
geo.offOrigin = np.array([0, 0, 0]) # Offset of image from origin (mm)
geo.offDetector = np.array([0, 0]) # Offset of Detector (mm)
geo.accuracy = 0.5 # Variable to define accuracy o
geo.COR = 0
#geo.rotDetector = np.array([0, 0, 0]) # Rotation of the detector, by
geo.mode = "cone" # Or 'parallel'. Geometry type.
lmbda = 1
lambdared = 0.999
initmode = None
verbose = True
qualmeas = ["RMSE", "SSD"]

[AnderBiguri]

Hum interesting. If you try OS-SART, or CGLS, do you also get it?

I suspect this is another nitche case of the function computeV. Try these two algorithms, and lets see what happens. That will help pinpoint the issue and I can provide an alternative.

[springXIACJ]

Hum interesting. If you try OS-SART, or CGLS, do you also get it?

I suspect this is another nitche case of the function computeV. Try these two algorithms, and lets see what happens. That will help pinpoint the issue and I can provide an alternative.
Both OS-SART and CGLS can be reconstructed normally, and the reconstruction results are as follows.

[AnderBiguri]

@springXIACJ seems to work!

If you want to use SART, try chaninging this line

TIGRE/Python/tigre/algorithms/iterative_recon_alg.py

Line 259 in 9975efe

V[i] = Atb(proj_one, geox, self.angleblocks[i], "FDK", gpuids=self.gpuids).mean(

to

V[i] = Atb(proj_one, geo, self.angleblocks[i], "FDK", gpuids=self.gpuids).mean(

[springXIACJ]

@springXIACJ seems to work!

If you want to use SART, try chaninging this line

TIGRE/Python/tigre/algorithms/iterative_recon_alg.py

Line 259 in 9975efe

V[i] = Atb(proj_one, geox, self.angleblocks[i], "FDK", gpuids=self.gpuids).mean(

to

V[i] = Atb(proj_one, geo, self.angleblocks[i], "FDK", gpuids=self.gpuids).mean(

I have changed this line, but the code didn't work.

[AnderBiguri]

My bad, I gave you a bad suggestion. Try instead to comment this lines:

TIGRE/Python/tigre/algorithms/iterative_recon_alg.py

Lines 251 to 255 in 9975efe

	# shrink the volume size to avoid zeros in backprojection
	geox.sVoxel = (
	geox.sVoxel * np.max(geox.sVoxel[1:] / np.linalg.norm(geox.sVoxel[1:])) * 0.9
	)
	geox.dVoxel = geox.sVoxel / geox.nVoxel