Automatic Parameter Tuning for X-ray Computed Tomography Reconstruction

Iterative reconstruction algorithms are able to significantly enhance the quality of X-ray CT images by incorporating more realistic imaging models and favorable prior information. However, the determination of parameters, such as step sizes in optimization algorithms, for good performance usually suffers laborious manual tuning. In this work, we propose schemes to automatically determine parameters in a two-stage reconstruction framework based on constrained total variation (TV) optimization. The data fidelity constraints are enforced through projection onto convex sets (POCS) and TV minimization is achieved through adaptive steepest descent. The relaxation parameter of POCS is determined by the projection data, while the step size of steepest descent is decided by the difference of POCS update either in projection domain or in image domain. The performance of proposed methods is evaluated using simulated data and physical phantom. Our results demonstrate that proposed algorithms with automatic parameter tuning can achieve satisfactory reconstruction for sparse-view CT data.