A metal projection segmentation algorithm based on Random walks for dental CBCT metal artifacts correction

The introduction of flat-panel detectors into the cone-beam computed tomography (CBCT) has a lot of benefits. Metallic implants have higher attenuation coefficient and it form shadows in the raw projection data. This shadow will cause streak artifacts which influence image quality and it is still a challenge to reduce the metal artifacts. There are many algorithms to reduce the metal artifacts and projection data preprocessing method is much more efficient. The vital step of this method is to segment the metal shadows in projection data. The goal of this paper is to find a method to segment the metal projection. In this problem, it is difficult to segment the projection only once to get a good result. But it is easy to find background regions that contains the metal projection and former regions which is inside the metal projection. Segmentation based on random walks utilizes the two regions and calculates every pixel's probability that it first reaches the former regions. Based on the obtained probability values, metal shadows are segmented. In comparison with other methods, the algorithm based on random walks gives the best result and it shows the clear boundary of metal projection. Modify the metal projection with total variation (TV) inpainting model, the reconstruction image quality has improved and the nearby soft- tissue regions are much clearer.