Streak Preventive Image Reconstruction with ART and Adaptive Filtering

Streaks arise in computed tomograms for a variety of reasons, such as presence of high-contrast edges and objects, aliasing errors, patient movement, and use of very few views. The problem appears to be an inherent difficulty with all reconstruction methods, including backprojection (with convolution) and the algebraic reconstruction technique (ART). This paper presents the derivation and results of an ART-like algorithm (SPARTAF) oriented towards prevention of streaks via optimization of a cost function based on features of streaks, subject to the constraints of the given projection data. The object-dependent method employs pattern recognition of streaks and adaptive filtering during iterative reconstruction by ART. Results of experiments with a test pattern and of application of the method to reconstructive tomography from radiographic films are presented and the convergence properties demonstrated.