An accelerated RAMLA reconstruction algorithm for X-ray cone-beam CT

Iterative image reconstruction algorithms have many advantages over analytical image reconstruction algorithms in computed tomography. A widely applied iterative algorithm is OSEM (ordered subsets expectation maximisation), which has good reconstructed image quality and costs less in computation time. Compared with the conventional OSEM algorithm, another OS method RAMLA (row action maximum likelihood algorithm) can not only bring about significant acceleration in the iterative reconstruction, but also outperforms the OSEM in its convergence rate. In this paper, an accelerated RAMLA algorithm (ARAMLA) is proposed and applied to X-ray cone-beam CT image reconstruction. By increasing the step size of the correction factor, the ARAMLA algorithm can further speed up the RAMLA algorithm while still retaining its convergence properties. A graphics processing unit (GPU)-based implementation of the ARAMLA is also developed for greatly reducing the computation time per iteration. Experimental results show that to achieve the same image quality as in RAMLA, ARAMLA, with an accelerating factor of 2, requires only about half the number of iterations as RAMLA.