Fig 1.
Diagram of the proposed method.
The proposed method consists of three steps. Step 1 shows truncation artifact correction. Step 2 shows small FOV sinogram synthesizing. Step 3 shows sinogram inpainting based MAR method.
Fig 2.
Representative images of the XCAT data.
Each row corresponds to a different part of the body. Each column represents the original XCAT image, metal implant, reconstructed image, and reconstructed image with truncation artifact correction. The red circles in the images represent the small FOV for each case. Display window width/window level = 2000 HU/0 HU.
Fig 3.
Siemens X-ray spectrum at 120 kVp.
Table 1.
Parameters for data acquisition and reconstruction for both the simulations and the experiments.
Fig 4.
Representative images of the clinical data.
Each row corresponds to a different part of the body. Each column represents the original clinical image, metal implant, reconstructed image, and reconstructed image with truncation artifact correction. The red circles in the images represent the small FOV for each case. Display window width/window level = 2000 HU/0 HU.
Fig 5.
(a) Experimental benchtop system and (b) semi-top view of the disk phantom.
Fig 6.
Each row corresponds to a different part of the body. Each column represents the reference image, uncorrected metal artifact image (corrected for only truncation artifact), and LMAR and NMAR result images using the previous and proposed methods. Display window width/window level = 2000 HU/0 HU.
Fig 7.
Sinograms and central profiles (189th view) of the XCAT shoulder results.
(a) Originally measured uncorrected sinogram, (b) small FOV based sinogram, (c,e) sinograms with the previous LMAR and NMAR methods, and (d,f) sinograms with the proposed LMAR and NMAR methods. The red dotted lines in the sinogram images represent the central profiles at the 189th view, and the gray shading in the central profiles represent the metal trace regions. The blue lines in the plots represent the sinogram values at the 189th view, and the red lines represent the reference sinogram values at the 189th view.
Table 2.
NMSE and SSIM results with XCAT images.
Fig 8.
Each row corresponds to a different part of the body. Each column represents the reference image, uncorrected metal artifact image (corrected only truncation artifact), and LMAR and NMAR result images using the previous and proposed methods. Display window width/window level = 2000 HU/0 HU.
Table 3.
NMSE and SSIM results with clinical images.
Fig 9.
Reference image and uncorrected metal artifact images (corrected for only truncation artifact) using the disk phantom experimental data. The red box represents the ROI which the NMSE and SSIM were computed. Display window width/window level = 3000 HU/-500 HU.
Fig 10.
Results using disk phantom experimental data.
Each row corresponds to the number of metal objects outside the small FOV. Each column represents the LMAR and NMAR result images using the previous and proposed methods. Display window width/window level = 2500 HU/-750 HU.
Table 4.
NMSE and SSIM results with experimental images.