Modified simultaneous iterative reconstruction technique for faster parallel computation

Three-dimensional iterative reconstruction of high-resolution computed tomography data poses significant difficulties due to the associated computational burden. In previous work, we have shown that implementing distributed computing techniques in addition to ordered subsets is an effective approach to decreasing the total reconstruction run-time. However, we also established that interprocessor communication accounts for a considerable portion of the total run-time. In this work, we first analyze the simultaneous iterative reconstruction technique (SIRT) to establish its convergence. We then modify the SIRT algorithm in order to substantially decrease the interprocessor communication requirements, and thus the final run-time, while maintaining convergence. We include error reduction statistics and timing results gathered from a reconstruction of a mouse data set to demonstrate the advantages of the modified SIRT algorithm