Abstract
1510
Introduction: Our group recently developed Prism-PET - a segmented light guide that enhances spatial and depth-of-interaction (DOI) resolution in multi-crystal, single-ended readout PET modules. The deterministic and anisotropic light-sharing characteristic of Prism-PET may be leveraged to decompose signal from scattered gammas in the photopeak and subsequently recover the location of the primary interaction within the scintillator crystal array, further improving spatial resolution. In this work, we connect the readout signal of a Prism-PET module with its constituent gamma-scintillator interactions via Monte Carlo simulation.
Methods: The Prism-PET module design utilized for this study consists of a 20 × 20 × 20 mm3 block of LYSO scintillator, pixelated into a 16 × 16 array of crystals, with the prism light guide coupled to the entrance end of the block, and an 8 × 8 array of silicon photomultiplier (SiPM) readout pixels fixed to the opposite end. Spatial offsets between the prisms and SiPMs enable a unique, anisotropic SiPM readout pattern for each scintillator crystal in the array. SiPM readouts were constructed from simulated gamma-scintillator interactions by combining GATE and TracePro Monte Carlo simulations with custom-built R code for data processing and visualization. GATE simulation captures the interactions of each incident gamma with the scintillator array, and outputs a set of hits detailing the position and energy deposition of each interaction. The hits are then aggregated by the crystal volume in which they occurred to construct the set of firing crystals. This set is filtered to remove crystals with less than 50 keV deposited to generate the set of firing crystals C that produce detectable SiPM patterns. The energy deposited on each crystal in this set ci is mapped to an SiPM matrix S(i) whose elements give the deposited energy at each SiPM readout pixel. TracePro simulation of optical photon transport generates this mapping based on the energy-averaged DOI of hits within the crystal, and position of the crystal within the array. The observable SiPM readout is generated by summing across S(i).
Results: Figure 3 illustrates a sample result for a single incident gamma, where 7 hits are rolled up to 3 crystals, each of which is mapped to an SiPM matrix. These matrices are then summed to give the observable SiPM readout. The anisotropy of the SiPM readout pattern suggests the viability of machine-learning approaches to decompose such compound events into their constituent crystals. Simulations of up to 10,000 events show that 2-crystal events are the most likely, with an average of 2.2 firing crystals per incident gamma in the photopeak.
Conclusions: We have built and validated a simulation pipeline that connects the SiPM readout pattern with its constituent photon-matter interactions. Data generated by this pipeline may then enable us to decompose readout patterns by computational methods, and recover the primary interaction crystal for scattered events. Figure 1: Schematic illustration of a single event (incident gamma ray) generating 3 hits (photon-matter interactions) within the scintillator crystal array. Photons generated by each hit are shared across the underlying SiPMs (diagonal shading) in a deterministic manner by the prism array atop the scintillator crystals (crosshatch shading). Figure 2: Schematic illustration of process flow linking the physical ground truth of radiation-matter interaction generated by Monte Carlo simulation, to the simulated SiPM readout. Figure 3: Visualizations generated for a single sample event. (a) illustrates individual hits in 3-dimensions, (b) shows the firing crystals (i=1, 2, 3), (c) gives the SiPM pattern S(i) for each firing crystal, while (d) gives the SiPM readout matrix, generated by summing the constituent SiPM patterns in (d). Figure 4: Distribution of events in the photopeak by the number of firing crystals. Of 10,000 simulated events, 3,682 were fully absorbed in the detector and are included in the distribution.