Reconstructing highly accurate relative stopping powers in proton computed tomography

Abstract

Proton computed tomography (pCT) is an evolving tomographic imaging modality with applications in proton and ion therapy. The method allows direct reconstruction of relative stopping power of patient tissues in a 3D-fashion. The pCT collaboration has built first experimental prototypes of pCT scanning systems [1] and has developed approaches to reconstruct proton CT images based on registering the coordinates and water equivalent path length (WEPL) of individual protons traversing the scanned volume. From these data one reconstructs the object boundary (hull) and initial image based on filtered back projection (FBP), calculates a most likely path (MLP) for each proton, and improves the initial image iteratively by solving a large linear system of equations of the form Ax = b using an iterative projection algorithm [2].

An Improved Method of Total Variation Superiorization Applied to Reconstruction in Proton Computed Tomography

“…Another benefit of NTVS is that it allows feasibility-seeking to be performed with a larger relaxation parameter λ than was appropriate with OTVS (λ = 0.0001, k = 12). It was found that with NTVS, the same RSP error can be obtained with λ = 0.0002 after performing k = 6 feasibility-seeking iterations without producing larger standard deviations, as previously experienced in practice with OTVS, which lead to the choice of λ = 0.0001 in previously published work with the simulated data set [42], [36]. Arriving at an acceptable solution in k = 6 feasibility-seeking iterations also offers substantial computational benefit.…”

“…Details of the pCT data preprocessing, calibration, and image reconstruction have been presented previously [36], [37]. For the purposes of this work, feasibility-seeking was performed using the DROP algorithm of [38] with blocks containing 3200 (simulated data set) and 25,600 (experimental data set) proton histories.…”

“…Figures 5(a) and 5(b) show the comparison of TV and standard deviation, respectively, for OTVS and NTVS with relaxation parameter λ = 0.0001, median filter radius r = 2 applied to the initial iterate [36], and 12 feasibility-seeking iterations. In each plot, the results for NTVS with and without inclusion of the TV reduction requirement are shown as a function of N .…”

Results from a Prototype Proton-CT Head Scanner

“…It is a challenge to complete the image reconstruction in a time interval comparable to the sixminute data acquisition, but Ordoñez demonstrated that it can be accomplished in only 4.5 minutes using ten compute nodes, each with twelve cores and two GPUs. Similarly, using different code Schultze demonstrated (Schultze, 2015) that a similar image could be reconstructed in 6.5 min on a single Xeon compute node using one NVIDIA K40 GPU. Figure 4 shows a reconstructed image of a CTP404 Geometry and Sensitometry module called the "sensitometry" phantom, a plastic cylinder with an assortment of cylindrical inserts (Catphan TM 504 phantom by The Phantom Laboratory, Salem, NY).…”

A Highly Accelerated Parallel Multi-GPU based Reconstruction Algorithm for Generating Accurate Relative Stopping Powers

“…Various concepts ranging from advanced detector designs to appropriately selecting the preprocessing techniques and parameters of the reconstruction algorithm have a major impact on the quality and accuracy of the pCT images [4], [5]. Calculating accurate RSPs is of great importance, but to take advantage of the pCT systems in a clinical setup, there is also a great need for real-time algorithms that can process hundreds of millions of protons in concise time frames not more than a few minutes.…”