Research Article
CT Metal Artifact Reduction Method Based on Improved Image Segmentation and Sinogram In-Painting
Table 3
Computation cost (in CPU seconds) for each steps in the Bal algorithm and the proposed correction method.
| | ClinicalImage1 | ClinicalImage2 | ClinicalImage3 | PhantomImage |
|
Bal’s Method in [7] | Pre-filtering step | 0.98 | 0.97 | 0.96 | 0.97 | Segmentation step | 67.23 | 23.44 | 24.21 | 14.13 | Inpainting step | 1.32 | 1.57 | 1.22 | 0.85 |
| | Total | 69.53 | 25.98 | 26.39 | 15.95 |
| Proposed correction (unparallelized) | Pre-filtering step | 237.56 | 236.92 | 237.17 | 236.32 | Segmentation step (metal) | 34.45 | 26.26 | 28.34 | 12.39 | Segmentation step (artifacts) | 26.21 | 16.96 | 16.92 | 4.93 | Inpainting step | 44.02 | 49.09 | 64.78 | 52.81 |
| | Total | 342.24 | 329.23 | 347.21 | 306.45 |
| Proposed correction (CUDA parallelized) | Pre-filtering step | 6.23 | 5.97 | 6.39 | 5.98 | Segmentation step (metal) | 34.45 | 26.26 | 28.34 | 12.39 | Segmentation step (artifacts) | 26.21 | 16.96 | 16.92 | 4.93 | Inpainting step | 1.42 | 1.58 | 2.06 | 1.74 |
| | Total | 33.86 | 24.51 | 25.37 | 12.65 |
|
|