I. INTRODUCTION Advances in model-based iterative reconstruction (IR) methods for x-ray CT and cone-beam CT (CBCT) imaging

C-arm cone-beam CT offers great potential in image-guided interventions, but conventional analytic reconstruction methods are associated with limited image quality, particularly for soft-tissue imaging. While model-based iterative reconstruction (IR) methods improve image quality and/or reduce radiation dose, long reconstruction time limits utility in clinical workflow. Additionally, in contrast to diagnostic CT, C-arm cone-beam CT (CBCT) involves complexities of lateral truncation, an incomplete orbit, and relatively few projections. Lateral truncation in particular slows reconstruction convergence and introduces large errors in the reconstruction. Faster IR algorithms are therefore essential for broader adoption in CBCT-guided procedures. This work examines the acceleration achieved by modifying the ordered-subset, separable quadratic surrogates algorithm for solving the penalized-likelihood (PL) objective to include Nesterov’s method, which utilizes “momentum” from image updates of previous iterations to better inform the current iteration and provide significantly faster convergence. Reconstruction performance was assessed in C-arm CBCT of an anthropomorphic head phantom and cadaveric torso, demonstrating that Nesterov’s method provides equivalent image quality while reducing the reconstruction time by an order of magnitude. Despite the slower convergence of IR with truncated C-arm CBCT, implementation of Nesterovaccelerated PL reconstruction on relatively inexpensive GPUs reduced reconstruction time from ~100 min for the ordered subset, separable quadratic surrogates method to as little as ~2 min.