Comparative study of analytical metal artifact reduction methods in CT imaging

Over the past few decades, computed tomography (CT) imaging has been merged as one of the leading cross-sectional imaging techniques in a wide range of clinical applications in diagnostic radiology, oncology, and multimodal molecular imaging. Despite the recognized value of this imaging modality, the quality and accuracy of CT images can be compromised by a number of implants. The presence of metal objects such as dental fillings, hip or knee prostheses, heart pacemakers, war fragments, and spinal cages can cause severe image artifacts. These types of artifacts appear as black and white streaks in the CT images, obscuring the structures and tissues around the metal implant which decreases the diagnostic values of the images. Metal artifacts also affect the accuracy of radiation therapy treatment planning, which relies on X-ray images to determine electron density and estimate the relative stopping power of particles. In this regard, different algorithms of the Metal Artifact Reduction (MAR) have been proposed over the decades to address this issue. In this study, five commonly used MAR algorithms in clinical practice have been evaluated using simulated and clinical datasets. These algorithms include linear interpolation (LI_MAR) of the degraded data in the sinogram space, reduction of metal artifacts by normalization method (NMAR), metal deletion technique (MDT), Orthopedic metal artifact reduction (OMAR), and a method based on iteration algorithms (MAP). Clinical CT images in different anatomical regions of the body, with different dimensions and types of metal implants, have been studied to evaluate the performance of the MAR algorithms. In order to quantitatively evaluate the quality of CT images corrected by the different MAR algorithms, the Normalized Root Mean Square Error (NRMSE) metric was employed. The quantitative analysis demonstrated the overall superior performance of the NMAR algorithm in effective metal artifact reduction compared to the other algorithms. The NMAR method exhibited relatively less signal distortion and reasonable processing time which make it a dependable solution in clinical practice.