Iterative CT reconstruction of real data with metal artifact reduction

We present an iterative tomographic reconstruction procedure suitable for processing of real projection datasets. This method is based on a polychromatic sinogram formation model that takes the beam hardening effect into account and thus reduces the incidence of streak artifacts due to metal inserts in the imaged body. It involves an optimized implementation and a novel measurement uncertainty model aimed at improving the conditioning of the problem and reducing the runtime of each iteration. Reconstruction of realistic-size images was performed on both synthetic and actual projection data. Comparison with results provided by existing techniques indicates that a significant reduction of the runtime is achieved, with no loss in image quality.