GPU‐accelerated regularized iterative reconstruction for few‐view cone beam CT

Abstract

Considering the image quality and execution times, this method may be useful for reconstruction of low-dose clinical acquisitions. It may be of particular benefit to patients who undergo multiple acquisitions by reducing the overall imaging radiation dose and associated risks.

Evaluation of the OSC‐TV iterative reconstruction algorithm for cone‐beam optical CT

“…Iterative reconstruction has been shown to significantly improve image quality in x-ray CT, 15,16 and, considering the similar forward models, is expected to demonstrate similar behavior in optical CT. For this work, OSC-TV has been selected among numerous iterative methods because of its ability to reduce noise and mitigate streaking artifacts in xray CT, and because it depends on very few free parameters. 17 It is worth outlining the mathematical formulation and some intrinsic properties of the OSC-TV algorithm in order to highlight its potential usefulness in cone-beam optical CT.…”

“…Iterative reconstruction is computationally intensive and may appear unnecessary in a context where the projection data have a high SNR. Nevertheless, it has the potential to yield more accurate 3D images of optical attenuation coefficient distributions in gels, as shown in the field of x-ray CT. 15,16 A robust iterative statistical reconstruction approach, the ordered subsets convex algorithm with regularization via total variation minimization (OSC-TV), 17 was used in this work. It incorporates a penalized Poisson loglikelihood objective function that is iteratively optimized via expectation-maximization.…”

System matrix computation vs storage on GPU: A comparative study in cone beam CT

“…Together with a high thread count, these properties conferred a 1.52 times lower execution time to the fully stored system matrix method. Considering these observations, this latter method is interesting for reconstruction algorithms for which the forward projection and backprojection may be implemented as a single CUDA kernel, such as OSC‐based and algebraic (ART‐based) methods . In the case where forward projection results have to be stored in the GPU RAM and analyzed separately before backprojection, the OTF approaches seem more promising, because of the faster forward projection operator.…”

“…Several symmetries may be useful if storing the system submatrices. A basic usage of symmetries was presented by Matenine et al . and a slightly extended formalism is presented below.…”

“…Several symmetries may be useful if storing the system submatrices. A basic usage of symmetries was presented by Matenine et al 39 and a slightly extended formalism is presented below. A detailed description of the conditions which ensure different symmetries may be found in Appendix A (available in the supplementary material).…”