Evaluation of Two CT Metal Artifact Correction Algorithms for a Novel Direction Modulated Brachytherapy Tandem Applicator

Our group has developed a novel, direction modulated brachytherapy (DMBT) tandem applicator designed for cervical cancer treatment. This MRI-compatible DMBT applicator contains metal in its construction (95% W, 3.5% Ni and 1.5% Cu, ρ=18g/cm3) and thus leads to poor quality CT images. In a phantom study, we have previously shown that use of a commercial metal artifact reduction algorithm, Orthopedic Metal Artifact Reduction (O-MAR) on a Phillips CT system, leads to improvements in the CT image quality. This work compares the results of an in-house developed normalized metal artifact reduction (NMAR) algorithm together with the commercial O-MAR algorithm to assess CT image quality of the DMBT tandem. A CT Acquisitions: Imaging was performed using a Philips Brilliance 16-slice CT scanner with our standard clinical pelvis protocol (1.5mm slice thickness, 120kVp, 600mAs, 1s rotation time, 16x0.75mm collimation, 0.692 pitch). A phantom, similar to the female pelvis in dimension, was filled with water and imaged twice.The first image set was in the absence of the applicator and served as a background image. The second image set contained the DMBT applicator (a novel design containing six channels of 1.3mm diameter each and an overall thickness of 5.4mm). This image set was reconstructed with and without the O-MAR algorithm. All three image sets were then sent to MatlabTM for analysis. B] Image Reconstruction and Analysis: NMAR was implemented in Matlab. From the original uncorrected CT image, a metal only image was created using CT numbers > 2500HU. Forward projections of both the metal only image and the original uncorrected image resulted in the metal only sinogram and the uncorrected sinogram, respectively. Masking of the uncorrected sinogram using the binary image from the metal only sinogram resulted in a gap which was then filled using linear interpolation of the surrounding data. This interpolated sinogram was used to reconstruct an image where CT numbers ranging from -200HU to 700HU were assigned the CT number of water to get a prior image. The prior image was forward projected to get the prior sinogram which was used for the normalization and denormalization steps in the NMAR method. The metal only image was finally inserted back in the NMAR corrected image to get the final NMAR corrected image. The dark and bright streakings present in the original uncorrected image were largely reduced in the images corrected by both the O-MAR and NMAR algorithms (Figure 1, upper panel). Bright streaking was still present in the corrected images and was more pronounced in the O-MAR corrected image than the NMAR corrected image. The black spots and the white streakings that were very close to the metal resulted in a larger loss of edge information in the O-MAR corrected image compared to the NMAR corrected image. The image profile comparison (Figure 1, lower panel) showed that both correction methods restored CT numbers that were very close to those of the background image. However, the closest match of the CT number to the background image was obtained on the NMAR corrected image. This work shows that both the commercial artifact reduction algorithm (O-MAR) and a research algorithm used in a preclinical setting (NMAR) resulted in substantial reductions of metal artifacts in the phantom images containing our novel DMBT tandem applicator. Although there were some remaining artifacts in the corrected images, they were subtle. The NMAR method restored image information that was closer to the background image compared to the O-MAR method. This work suggests that our MRI-compatible DMBT applicator can also be integrated into image guided adaptive brachytherapy workflow using CT, provided that a metal artifact reduction technique is applied to the CT images.