Metal artifact reduction based on multi-level sinogram segmentation and sequentially applied MAP-EM reconstruction method

Metal artifacts and their reduction is a common problem in imaging field because artifacts arising from these high density objects often hinder the underlying anatomical structures. This study investigates the performance of sequentially applied maximum a posteriori expectation maximization (MAP-EM) for metal artifact correction. Firstly, positions of metal objects were identified with a novel multi-level segmentation method based on weighted Otsu's threshold level. The sinogram bins representing the metal objects were regarded as missing data which are modeled in the system matrix of the sequentially applied MAP-EM method with the spatial domain median filtering. Regularization level in the MAP-EM was decreased gradually throughout the sequences. Qualities of reconstructed images were investigated both qualitatively and quantitatively on a numerical jaw phantom with different amount of metals. NMSE% and line profile analysis results were in parallel with the visual impression that as penalization is reduced, reconstructed images have higher contrast with sharper boundaries between anatomical structures. This study provides encouraging results for using sequentially iterative algorithms such as sequentially applied MAP-EM in order to have a more accurate reconstruction of intensity values.