Metal Artifact Reduction of Dental Fillings in Head and Neck CT Images

Introduction: The issue of metal artifact and its reduction is as old as the clinical use of computed tomography itself. When metal objects such as dental fillings, hip prostheses or surgical clips are present in the computed tomography (CT) field of view (FOV), make severe artifacts that reduce the image quality and accuracy of CT numbers. They can lead to unreliable clinical results due to inability in true tumor volume delineation and uncertainty in dose calculation of treatment planning software. So far, no generally accepted solution to this problem, especially for small size dental fillings, has been found. Herein, an approach is presented to reduce the dental filling artifacts in CT images of head and neck patients with multiple dental fillings. The proposed approach is compared with the commercial orthopedic metal artifact reduction (O-MAR) algorithm by Philips company. Materials and Methods: Our algorithm consists of six steps: 1) metallic object segmentation by thresholding, 2) obtaining the prior image by multi thresholding of the initial image, 3) normalization of the original image sinogram by prior image, 4) interpolation (replacing the affected projection data by previous slice unaffected projection data, 5) denormalization of the corrected sinogram and 6) application of a two dimensional adaptive filter to the final image. The forth step can be implemented by substitution of affected projections with opposite view projections. Both methods were tested and previous slice substitution was superior. Quantitative evaluation of the commercial and proposed algorithms was done in terms of peak signal-to-noise ratio (PSNR), signal-to- noise ratio (SNR) and structural similarity index (SSIM) on a CT image of a head and neck patient with four dental fillings. The posterior part of the head in original image was considered as the reference image. Also two uniform regions of interest (ROI) were considered in the image. ROI1 was posterior region of the head away from dental fillings, while ROI2 was restricted to the tongue region near to the dental fillings. Results: The PSNR, SNR, SSIM values obtained by the proposed method and O-MAR algorithm were 37.05, 36.13, 22.29, 21.37, 0.93 and 0.84, respectively. Using our approach, the standard deviation of CT numbers in ROI1 was 27 times less than the original image and 7% better than the commercial competitor. However, for ROI2 the commercial algorithm was 36% more successful in reduction of variations due to metal artifact. Conclusion: The proposed approach can be applied successfully for dental filling artifact reduction in head and neck patients. Although the performance of the commercial method was superior to ours in near dental regions, for the farther regions with more critical organs was vice versa. However, weighting the interpolation and adaptive filtering steps may result in even better results.