Limits...
An Effective CUDA Parallelization of Projection in Iterative Tomography Reconstruction.

Xie L, Hu Y, Yan B, Wang L, Yang B, Liu W, Zhang L, Luo L, Shu H, Chen Y - PLoS ONE (2015)

Bottom Line: In this paper, a strategy of Fixed Sampling Number Projection (FSNP) is proposed to ensure the operation synchronization in the ray-driven projection with Graphical Processing Unit (GPU).We validate the performance of this FSNP approach using both simulated and real cone-beam CT data.Experimental results show that compare to the conventional approach, the proposed FSNP method together with texture fetching is 10~16 times faster than the conventional approach based on global memory, and thus leads to more efficient iterative algorithm in CT reconstruction.

View Article: PubMed Central - PubMed

Affiliation: Oral Hospital of Jiangsu Province, Affiliated to Nanjing Medical University, Jiangsu, China.

ABSTRACT
Projection and back-projection are the most computationally intensive parts in Computed Tomography (CT) reconstruction, and are essential to acceleration of CT reconstruction algorithms. Compared to back-projection, parallelization efficiency in projection is highly limited by racing condition and thread unsynchronization. In this paper, a strategy of Fixed Sampling Number Projection (FSNP) is proposed to ensure the operation synchronization in the ray-driven projection with Graphical Processing Unit (GPU). Texture fetching is also used utilized to further accelerate the interpolations in both projection and back-projection. We validate the performance of this FSNP approach using both simulated and real cone-beam CT data. Experimental results show that compare to the conventional approach, the proposed FSNP method together with texture fetching is 10~16 times faster than the conventional approach based on global memory, and thus leads to more efficient iterative algorithm in CT reconstruction.

Show MeSH

Related in: MedlinePlus

Profiles of the reconstructed images in Fig 5.The profile is marked in blue line in Fig 5(A), 5(B), 5(C) and 5(D) are the zoomed regions in marked as “zoom 1”, “zoom 2” and “zoom 3” respectively.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4664243&req=5

pone.0142184.g006: Profiles of the reconstructed images in Fig 5.The profile is marked in blue line in Fig 5(A), 5(B), 5(C) and 5(D) are the zoomed regions in marked as “zoom 1”, “zoom 2” and “zoom 3” respectively.

Mentions: We then evaluated the performance of FSNP in iterative reconstruction algorithm. The OSEM (Ordered Subsets Expectation-Maximization) algorithm with 30 subsets and 100 iterations was chosen as the iterative algorithm. Results in Figs 5 and 6 show that the linear interpolation based FSNP together with voxel-driven back-projector can provide results similar to the matched distance-driven pair, both visually and quantitatively. Although the measurements are assumed to be independent form each other in ideal projection model, they are in fact somehow correlated, due to the imperfect collimation and scatter effect. The blob-based kernel takes such correlation into consideration, and thus leads to a more realistic model in characterizing the residual between observed measurements and image projections, especially for differences near the edges. As a result, we may see that different from the results in Fig 4(F), the reconstructed images in Fig 5(D) indicate that the blob based FSNP also leads to a good preservation of edges in addition to artifact suppression. Fig 6 plots the line profile (the vertical blue line in the left image in Fig 5(A)) of the reconstructions in Fig 5, from which we can see that the reconstruction with blob based FSNP results in a smoother profile with a better match of the reference profile than others (zoom 1); the distance driven projection provides blurred edges (zoom 2); and the blob based FSNP has the best performance in recovering peak value (with the best matched profile with phantom data in zoom 3).


An Effective CUDA Parallelization of Projection in Iterative Tomography Reconstruction.

Xie L, Hu Y, Yan B, Wang L, Yang B, Liu W, Zhang L, Luo L, Shu H, Chen Y - PLoS ONE (2015)

Profiles of the reconstructed images in Fig 5.The profile is marked in blue line in Fig 5(A), 5(B), 5(C) and 5(D) are the zoomed regions in marked as “zoom 1”, “zoom 2” and “zoom 3” respectively.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4664243&req=5

pone.0142184.g006: Profiles of the reconstructed images in Fig 5.The profile is marked in blue line in Fig 5(A), 5(B), 5(C) and 5(D) are the zoomed regions in marked as “zoom 1”, “zoom 2” and “zoom 3” respectively.
Mentions: We then evaluated the performance of FSNP in iterative reconstruction algorithm. The OSEM (Ordered Subsets Expectation-Maximization) algorithm with 30 subsets and 100 iterations was chosen as the iterative algorithm. Results in Figs 5 and 6 show that the linear interpolation based FSNP together with voxel-driven back-projector can provide results similar to the matched distance-driven pair, both visually and quantitatively. Although the measurements are assumed to be independent form each other in ideal projection model, they are in fact somehow correlated, due to the imperfect collimation and scatter effect. The blob-based kernel takes such correlation into consideration, and thus leads to a more realistic model in characterizing the residual between observed measurements and image projections, especially for differences near the edges. As a result, we may see that different from the results in Fig 4(F), the reconstructed images in Fig 5(D) indicate that the blob based FSNP also leads to a good preservation of edges in addition to artifact suppression. Fig 6 plots the line profile (the vertical blue line in the left image in Fig 5(A)) of the reconstructions in Fig 5, from which we can see that the reconstruction with blob based FSNP results in a smoother profile with a better match of the reference profile than others (zoom 1); the distance driven projection provides blurred edges (zoom 2); and the blob based FSNP has the best performance in recovering peak value (with the best matched profile with phantom data in zoom 3).

Bottom Line: In this paper, a strategy of Fixed Sampling Number Projection (FSNP) is proposed to ensure the operation synchronization in the ray-driven projection with Graphical Processing Unit (GPU).We validate the performance of this FSNP approach using both simulated and real cone-beam CT data.Experimental results show that compare to the conventional approach, the proposed FSNP method together with texture fetching is 10~16 times faster than the conventional approach based on global memory, and thus leads to more efficient iterative algorithm in CT reconstruction.

View Article: PubMed Central - PubMed

Affiliation: Oral Hospital of Jiangsu Province, Affiliated to Nanjing Medical University, Jiangsu, China.

ABSTRACT
Projection and back-projection are the most computationally intensive parts in Computed Tomography (CT) reconstruction, and are essential to acceleration of CT reconstruction algorithms. Compared to back-projection, parallelization efficiency in projection is highly limited by racing condition and thread unsynchronization. In this paper, a strategy of Fixed Sampling Number Projection (FSNP) is proposed to ensure the operation synchronization in the ray-driven projection with Graphical Processing Unit (GPU). Texture fetching is also used utilized to further accelerate the interpolations in both projection and back-projection. We validate the performance of this FSNP approach using both simulated and real cone-beam CT data. Experimental results show that compare to the conventional approach, the proposed FSNP method together with texture fetching is 10~16 times faster than the conventional approach based on global memory, and thus leads to more efficient iterative algorithm in CT reconstruction.

Show MeSH
Related in: MedlinePlus