Purpose:
A clean and anatomically informative prior image is important in normalized metal artifact reduction (NMAR) method. To fully utilize the advantages of NMAR in terms of speed and computational cost, the prior image generation should desirably be fast and least case-dependent. In this study, we propose an efficient sinogram surgery method for computing prior image that does not require iteration or optimization.
Materials and Methods:
Initially reconstructed image using FBP has severe streaks and noises due to the metallic objects. To create a metal-replaced FBP image, a smoothing filter (Gaussian used in this study) was applied to mitigate the noise, and the values of metal regions were replaced by a constant value that has a similar scale with surrounded soft tissues. By forward projecting of it, the sinogram of reprojection was obtained.
Then the mathematically derived beam hardening corrector was modeled using segmented metal image. By forward projecting this pre-processed beam-hardening model, the correction mask was generated in the sinogram domain. It was utilized to remove the artifact corrupted region after the sinogram surgery. The sinogram surgery was operated to replace the metal trace of original sinogram with the sinogram of reprojection. Even the main artifact causes were removed through the correction mask, the sinogram inconsistencies were still remained. It was effectively reduced through the normalization process that performs a low-frequency matching. Finally, the operated sinogram was reconstructed using FBP to generate a prior image for later NMAR procedure.
For materials, polychromatic (80kVp) x-ray measurement data of XCAT phantom were acquired using simulation. The poisson random noises were added to create beam-hardening and photon starvation artifacts. Phantom 1 has two insertions of titanium with 360 views of projection data (SDD = 1300mm, SOD = 900mm). Phantom 2 has two insertions of titanium with 720 views of projection data (SDD = 2750mm, SOD = 2500mm).
Results:
The prior image show well-preserved anatomical structures while substantially reducing the metal artifacts when compared with the original FBP image and LI-MAR image. Consequently, the NMAR results with proposed prior method show best results in terms of structure resolution and artifact reduction
Conclusions:
In our study, an efficient way for computing the high-quality prior image that does not require any iteration or optimization was proposed. Therefore, the complementary use of this technique for NMAR is possible to preserve anatomical structures, and increase the soft-tissue contrast without compromising their benefits in terms of speed, and computational cost.
Keywords: prior image, metal artifacts, NMAR, sinogram surgery, normalization