Hybrid MPI+OpenMP parallelization of image reconstruction in proton beam therapy on multi-core and many-core processors

The advantage of proton beam therapy is that the lethal dose of radiation is delivered by a sharp increase toward the end of the beam range, known as the Bragg peak (BP), with no dose delivered beyond. By using these characteristics of the BP, radiation dose to the tumor can be maximized, with greatly reduced radiation dose to the surrounding healthy tissue. If the secondary gamma rays that are emitted through interaction of the protons in the beam with atoms in the patient tissue could be imaged in (near) real-time during beam delivery, it could provide a means of visualizing the delivery of dose for verification of proper treatment delivery. However, such imaging requires very fast image reconstruction to be feasible. This project focuses on measuring the performance of a new parallel version of the CCI (Compton camera imaging) image reconstruction algorithm. We show two conclusions: (i) The new hybrid MPI+OpenMP parallelization of the code on the many-core Intel Xeon Phi KNL processor with 68 computational cores makes fast reconstruction times possible and thus enables the use of CCI in real time during treatment. (ii) A compute node with two of the latest multi-core Intel Skylake CPUs with 24 cores performs even better in a first comparison of both types of processors available on Stampede2 at the Texas Advanced Computing Center (TACC).