Extending gPET for Multi-Layer PET Simulation

Depth-of-interaction (DOI) encoding is an effective strategy for reducing parallax error and preserving spatial resolution in positron emission tomography (PET), particularly in compact small-animal scanners. To enable efficient simulation-driven design of DOI-capable systems, we extend the GPU-accelerated Monte Carlo toolkit gPET to support flexible multi-layer detector geometries. The original three-level hierarchical detector model in gPET (panel-module-crystal) was expanded by introducing an intermediate “layer” level, enabling parameterized modeling of stacked scintillator architectures. The photon transport algorithm was correspondingly updated to sample interactions across multiple layers and detector panels while preserving GPU-efficient memory usage. The framework was validated using three scanner configurations: a conventional single-layer ring (H2RSPET-1CL), an aligned split-layer design (H2RSPET-1CL-split), and an offset dual-layer design (H2RSPET-2CL). System performance was evaluated following NEMA NU4-2008 protocols using sensitivity, spatial resolution, and Derenzo phantom simulations with CASToR-based maximum likelihood expectation maximization reconstruction. The H2RSPET-1CL and H2RSPET-1CL-split configurations produced statistically identical hit distributions, while H2RSPET-2CL exhibited the expected offset interaction patterns. Sensitivity of H2RSPET-2CL remained comparable to H2RSPET-1CL, generally within about 2-5 percent, while radial spatial resolution improved substantially (0.8-1.6 mm vs. 1.0-4.2 mm from the center to a 50 mm radial offset). Runtime performance remained essentially unchanged between configurations. The extended gPET framework therefore enables fast and flexible simulation of multi-layer PET detectors and supports efficient optimization of DOI-enabled PET system designs.

💡 Research Summary

The paper presents an extension of the GPU‑accelerated Monte Carlo PET simulation toolkit gPET to support multi‑layer detector geometries, enabling efficient design studies of depth‑of‑interaction (DOI) capable small‑animal PET scanners. The original gPET architecture models detectors with a three‑level hierarchy (panel‑module‑crystal). The authors inserted a new “layer” level between module and crystal, creating a four‑level hierarchy (panel‑module‑layer‑crystal). This required adding parameters for the number of layers, per‑layer material, density, crystal size, spacing, and offsets to the configuration file, updating the geometry reader, and labeling each interaction with crystal, layer, module, and panel indices. The photon transport algorithm retained the Woodcock tracking method for free‑path sampling, but now includes a layer‑selection step based on the photon’s depth coordinate in the panel’s local frame, loading the appropriate material properties for interaction sampling. The design preserves GPU‑friendly memory usage and computational throughput, ensuring that adding layers does not significantly increase runtime or memory consumption.

To validate the extension, three scanner configurations were simulated: (1) a conventional single‑layer ring (H2RSPET‑1CL), (2) a split‑layer version where each 10 mm crystal is replaced by two 5 mm crystals stacked radially (H2RSPET‑1CL‑split), and (3) an offset dual‑layer design (H2RSPET‑2CL) where two 5 mm layers are stacked with a half‑crystal axial offset, resulting in a 50 × 50 inner array and a 51 × 51 outer array. All scanners used LSO/LYSO crystals (density 7.4 g/cm³) and identical panel dimensions (radius 85 mm, axial length 34.1 cm). Hit distributions were recorded and compared. The single‑layer and split‑layer configurations produced virtually identical DOI histograms, confirming that the layer split does not alter physics. The offset dual‑layer design showed fewer hits in the shallow DOI region (0–4.5 mm) due to the smaller crystal count in the inner layer, and a clear lateral shift in the global x‑y hit map, matching the intended geometric offset.

Performance evaluation followed the NEMA NU‑4‑2008 standard. Sensitivity was measured with four energy windows (150‑700, 250‑700, 350‑700, 450‑700 keV) across radial offsets (0‑70 mm) and axial offsets (0‑170 mm). The dual‑layer scanner’s sensitivity differed by only 2‑5 % from the single‑layer system, indicating that DOI‑enhancing geometry does not sacrifice detection efficiency. Spatial resolution was assessed using a 0.1 mm³ voxel air phantom with a point source and 12 iterations of 3‑D MLEM reconstruction (CASTOR v3.1.1). The single‑layer system showed radial FWHM degrading from ~1.0 mm at the center to ~4.2 mm at 50 mm radius, whereas the offset dual‑layer design achieved 0.8‑1.6 mm over the same range, demonstrating substantial resolution improvement due to DOI information. Derenzo phantom simulations (rod diameters 0.3–1.0 mm) further confirmed enhanced rod visibility and contrast for the dual‑layer configuration, especially for the smallest rods.

Computational performance was benchmarked on a single NVIDIA TITAN Xp GPU (12 GB, CUDA 11.6). Ten repeated runs of the Derenzo activity distribution yielded average wall‑clock times that differed by less than 7 % among the three configurations, confirming that the added layer handling imposes negligible overhead. Memory usage remained comparable to the original gPET implementation.

In summary, the authors successfully extended gPET to model multi‑layer PET detectors while preserving its GPU‑efficient architecture. The extended toolkit accurately reproduces DOI‑dependent hit patterns, maintains sensitivity, and delivers the expected spatial resolution gains associated with DOI encoding. Runtime and memory footprints are essentially unchanged, making the tool practical for rapid, iterative optimization of DOI‑enabled PET system designs, particularly for compact small‑animal scanners where parallax errors are most pronounced.