Optimized Momentum Steps for Accelerating X-ray CT Ordered Subsets Image Reconstruction

Recently, we accelerated ordered subsets (OS) methods for low-dose X-ray CT image reconstruction using momentum techniques, particularly focusing on Nesterov’s momentum method. This paper develops an “optimized” momentum method that is faster than Nesterov’s method. Drori and Teboulle’s original version requires substantial memory space and computation time per iteration. Therefore, we design an efficient implementation approach of the optimized momentum method that uses storage and computation comparable to Nesterov’s method. We also propose to combine it with OS methods. We examine the acceleration of the proposed algorithm using 2D X-ray CT simulation data.