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

FDK reconstructions of Shepp-Logan phantom.From left to right are cross-section, sagittal view and coronal view, respectively. Row (A): phantom images; Row (B): the reconstruction with ray-driven projection and ray-driven back-projector (RMSE = 7.11%); Row (C) reconstruction with distance-driven projection and distance-driven back-projector (RMSE = 5.06%); Row (D): the reconstruction with linear interpolation based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 5.30%); Row (E): the reconstruction with linear interpolation based FSNP (512 sampling points per ray) and voxel-driven back-projector (RMSE = 5.22%); Row (F): the reconstruction with blob based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 2.98%).
© Copyright Policy
Related In: Results  -  Collection

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

pone.0142184.g004: FDK reconstructions of Shepp-Logan phantom.From left to right are cross-section, sagittal view and coronal view, respectively. Row (A): phantom images; Row (B): the reconstruction with ray-driven projection and ray-driven back-projector (RMSE = 7.11%); Row (C) reconstruction with distance-driven projection and distance-driven back-projector (RMSE = 5.06%); Row (D): the reconstruction with linear interpolation based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 5.30%); Row (E): the reconstruction with linear interpolation based FSNP (512 sampling points per ray) and voxel-driven back-projector (RMSE = 5.22%); Row (F): the reconstruction with blob based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 2.98%).

Mentions: We performed FDK reconstruction with Ram-Lak filter. 3-D Shepp-Logan phantom was used with the system configuration in Table 2 (low resolution mode), and the phantom data was projected into 360 angles. The results are shown in Fig 4. We do not provide the result for the voxel-driven projections because the grid artifacts in such projection mode often results in severe artifacts in FDK reconstruction [26]. With respect to the phantom images in Fig 4(A), obvious streak artifacts can be observed in Fig 4(B), 4(C), 4(D) and 4(E); the combination of linear interpolation based FSNP and voxel-driven back-projection operator can provide better results than the classic ray-driven projector and back-projector pair. We can also see in Fig 4(F) that the blob based FSNP leads to the reconstructions with effective artifact suppression and the lowest MSE values, but at the cost of blurred edges. This is because the simulated projections generated by blob based FSNP are blurred by the weighted summation of neighboring voxels.


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)

FDK reconstructions of Shepp-Logan phantom.From left to right are cross-section, sagittal view and coronal view, respectively. Row (A): phantom images; Row (B): the reconstruction with ray-driven projection and ray-driven back-projector (RMSE = 7.11%); Row (C) reconstruction with distance-driven projection and distance-driven back-projector (RMSE = 5.06%); Row (D): the reconstruction with linear interpolation based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 5.30%); Row (E): the reconstruction with linear interpolation based FSNP (512 sampling points per ray) and voxel-driven back-projector (RMSE = 5.22%); Row (F): the reconstruction with blob based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 2.98%).
© Copyright Policy
Related In: Results  -  Collection

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

pone.0142184.g004: FDK reconstructions of Shepp-Logan phantom.From left to right are cross-section, sagittal view and coronal view, respectively. Row (A): phantom images; Row (B): the reconstruction with ray-driven projection and ray-driven back-projector (RMSE = 7.11%); Row (C) reconstruction with distance-driven projection and distance-driven back-projector (RMSE = 5.06%); Row (D): the reconstruction with linear interpolation based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 5.30%); Row (E): the reconstruction with linear interpolation based FSNP (512 sampling points per ray) and voxel-driven back-projector (RMSE = 5.22%); Row (F): the reconstruction with blob based FSNP (256 sampling points per ray) and voxel-driven back-projector (RMSE = 2.98%).
Mentions: We performed FDK reconstruction with Ram-Lak filter. 3-D Shepp-Logan phantom was used with the system configuration in Table 2 (low resolution mode), and the phantom data was projected into 360 angles. The results are shown in Fig 4. We do not provide the result for the voxel-driven projections because the grid artifacts in such projection mode often results in severe artifacts in FDK reconstruction [26]. With respect to the phantom images in Fig 4(A), obvious streak artifacts can be observed in Fig 4(B), 4(C), 4(D) and 4(E); the combination of linear interpolation based FSNP and voxel-driven back-projection operator can provide better results than the classic ray-driven projector and back-projector pair. We can also see in Fig 4(F) that the blob based FSNP leads to the reconstructions with effective artifact suppression and the lowest MSE values, but at the cost of blurred edges. This is because the simulated projections generated by blob based FSNP are blurred by the weighted summation of neighboring voxels.

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