Dictionary-based software for proton dose reconstruction and submilimetric range verification

Abstract

Objective: This paper presents a new method for fast reconstruction (compatible with in-beam use) of deposited dose during proton therapy using data acquired from a PET scanner. The most innovative feature of this novel method is the production of noiseless reconstructed dose distributions from which proton range can be derived with high precision. Approach: A new MLEM $\&$ Simulated Annealing (MSA) algorithm, developed especially in this work, reconstructs the deposited dose distribution from a realistic pre-calculated activity-dose dictionary. This dictionary contains the contribution of each beam in the plan to the 3D activity and dose maps, as calculated by a Monte Carlo (MC) simulation. The MSA algorithm, using a priori information of the treatment plan, seeks for the linear combination of activities of the precomputed beams that best fits the observed PET data, obtaining at the same time the deposited dose. Main results: the method has been tested using simulated data to determine its performance under 4 different test cases: 1) dependency of range detection accuracy with delivered dose, 2) in-beam vs offline verification, 3) ability to detect anatomical changes and 4) reconstruction of a realistic spread-out Bragg peak. The results show the ability of the method to accurately reconstruct doses from PET data corresponding to 1-Gy irradiations, both in intra-fraction and inter-fraction verification scenarios. For this dose level (1 Gy) the method was able to spot range variations as small as 0.6 mm. Significance: out method is able to reconstruct dose maps with remarkable accuracy from clinically relevant dose levels down to 1 Gy. Furthermore, due to the noiseless nature of reconstructed dose maps, an accuracy better than one millimeter was obtained in proton range estimates. These features make of this method a realistic option for range verification in proton therapy.

GPU accelerated Monte Carlo scoring of positron emitting isotopes produced during proton therapy for PET verification

“…Throughout this work we have compared FRED to the QGSP_BIC hadronic models of Geant4, however the scoring of activation within FRED may also be performed with any cross sections, which can be loaded into the GPU kernel at run time. This allows use of in house or experimentally measured cross sections as in (Parodi et al 2007, Onecha et al 2022. We have also only considered production of isotopes relevant for offline imaging, however scoring of any isotope of interest, for example 12 N which is important for on-line imaging (Ozoemelam et al 2020), is possible.…”

“…The main motivation for implementation of PEI production scoring capabilities using GPU accelerated codes is the significant decrease in total simulation time. We note that an equivalent variance reduction technique has previously been introduced into the FLUKA code (Parodi et al 2007), and was also recently implemented in TOPAS for use in generating a dictionary of activities for dose reconstruction (Onecha et al 2022), however the MC simulation was notably still the most time consuming step. Such an approach may also be implemented for the GATE toolkit.…”

“…Analytical convolution techniques for calculation of the activation of the patient have been proposed in order to reduce dependency on computationally intensive Monte Carlo (MC) codes (Parodi and Bortfeld 2006, Attanasi et al 2009, Frey et al 2013, however these techniques may suffer from the same difficulties faced by analytic dose calculation engines in highly inhomogeneous regions (Schaffner et al 1999), and therefore may not provide the same verification accuracy as that of MC (Paganetti 2012). Many studies on patient activation utilise full MC toolkits based on the Geant4 code, such as GATE (Jan et al 2013, Robert et al 2013, Meiner et al 2019, and TOPAS (Perl et al 2012, Onecha et al 2022, or FLUKA (Parodi et al 2007, Augusto et al 2018. Such toolkits provide the most accurate modelling of all relevant physical processes, however are limited by large simulation times to achieve sufficient statistics for clinical plans.…”

“…Speedup is then limited by the available computing resources and the required statistics of the simulation. Flux based scoring techniques have previously been implemented in such codes using experimentally determined cross sections (Parodi et al 2007, Onecha et al 2022. Such techniques allow for simulation of fewer primary particles to calculate the PEI distribution, and can achieve better statistics thanks to variance reduction, however still rely on computationally intensive codes.…”

UMC-PET: a fast and flexible Monte Carlo PET simulator

“…Nuclear activation during irradiation generates positron emitter fragments which may be used for range verification (Kraan 2015, Bauert et al 2019, Masuda et al 2020. MC simulations of PET signals are required to have a reliable estimation of detector response and proton range reconstruction in clinical scenarios (Jan et al 2013, Choi et al 2020, Onecha et al 2022.…”