Voxel S value (VSV) kernels underpin internal radionuclide dosimetry but conventionally require intensive Monte Carlo (MC) transport. Surrogates trained on analytical dose point kernel approximations would only reproduce a closed-form expression; we therefore trained machine learning (ML) surrogates on MC-derived kernels: 3000 voxel dose kernels from the open-access OpenDose collection (voxel integration of MC dose point kernels; 10 radionuclides; 60 voxel sizes from 0.1 to 6.0 mm; 5 tissue media; 1.57 million symmetry-unique radial samples within r ≤ 25 voxels). A 28-feature physics-informed representation trained gradient boosted trees (GBT), random forests (RF) and a multilayer perceptron, evaluated on held-out kernels, by leave-one-radionuclide-out (LORO), leave-one-voxel-size-out (LOVS) and leave-one-medium-out cross-validation, by uncertainty calibration and by sphere-dose validation against a deterministically re-integrated reference. In-distribution accuracy was high (RF median relative absolute error 0.41%, R2 = 0.99986; GBT 1.54%). Resolution interpolation matched it (LOVS median 1.40%), degrading only below 1 mm. Cross-radionuclide extrapolation was poor (LORO median 78%): moderate for radionuclides resembling the training span but catastrophic for Y-90, the only radionuclide without appreciable photon emission. Radial offset relative to the β range governed the predictions. Recalibrated intervals reached 94.1% coverage, and the predictive uncertainty separated both held-out radionuclides tested from the training span. Median sphere-dose errors were 3.1–12.3% (1.3–7.5% with an energy-weighted loss), partly because the distributed kernels carry near-field numerical artefacts. Surrogates interpolate accurately across resolution and within the training decay-physics span, but do not extrapolate to unseen emission spectra; we report this boundary transparently.
Low Gain Avalanche Detectors (LGADs) have emerged as key sensors for High Energy Physics and related applications due to their excellent time resolution and enhanced signal-to-noise ratio. However, their use in low-energy ion spectrometry has generally been limited by gain suppression effects, arising from the high ionization densities produced by such ions within the multiplication layer, which leads to a reduction of the gain and a non-linear detector response with ion energy.In this work, we investigate an alternative detection configuration aimed at mitigating this effect by shifting the region of highest ionization density away from the gain layer. Alpha spectrometry measurements in the energy range 3–9 MeV were performed by irradiating the detector from both the frontside (multiplication layer side) and the backside of the devices. Additionally, potential gain suppression effects when the detected particles are photons were evaluated using an X-ray tube operated at 40 kV, with irradiation from both the front and back sides.The results show that backside irradiation leads to a higher and more uniform gain and a linear relationship between generated charge and alpha particle energy. In particular, a gain of 3–4 is obtained for backside irradiation, while a nearly constant gain of ≈ 2 is achieved for frontside irradiation. In addition, the achieved energy resolution (1%–2%) is comparable to that of standard PiN detectors fabricated under the same conditions. Finally, the gain response to X-rays was found to be very similar to that observed for alpha particles, suggesting that no gain suppression is observed in the studied LGADs for backside irradiation.These findings demonstrate the potential of LGADs for low-energy ion spectrometry, particularly in applications where low deposited energy makes enhanced signal-to-noise ratio a critical advantage.
The three-dimensional (3D) neutron flux distribution in a self-developed heterogeneous head phantom was characterized by integrating Au-wire activation measurements, a Monte Carlo simulation using an analytical geometry model, and a treatment planning system (TPS) prediction based on computed tomography (CT) derived voxel geometry. This integration established a three-way comparison framework to rigorously validate both the source term description and the TPS in clinical boron neutron capture therapy (BNCT). The phantom mimicked a head structure using four distinct materials to explicitly define the skin, skull, and brain tissue, incorporating an internal cavity and a tumor region. A total of 35 bare and Cd-covered Au wires were embedded along five selected paths within the phantom. This configuration mapped the 197Au(n,γ)198Au reaction rates across various locations and interfaces, providing a 3D measurement dataset for direct comparison with numerical predictions. The detector-embedded phantom was irradiated at the Tsing Hua Open-pool Reactor BNCT facility. Both calculation models well-reproduced the measured results, with most discrepancies within 10%, validating the source term definition and TPS capability. The results demonstrate that this heterogeneous head phantom serves as a reliable benchmarking platform for BNCT dosimetry quality assurance and inter-facility comparisons.
Electron-beam (EB) irradiation offers an initiator-free route for controlling network formation in highly filled polymer systems, yet the extent to which radiation-induced structural changes are translated into fire and mechanical responses in halogen-free flame-retardant composites remains insufficiently resolved. This study examines dose–network–property relationships in LDPE/EVA composites containing aluminum trihydroxide (ATH), magnesium hydroxide (MH), and zinc borate (ZB). Composites containing 60–64 wt% inorganic filler were subjected to EB treatment up to 250 kGy and characterized by gel-content analysis, hot-set testing, thermogravimetric analysis, tensile testing, limiting oxygen index (LOI), smoke-density measurements, and cone calorimetry. Increasing absorbed dose promoted polymer-network development, directly evidenced by increasing gel content and improved hot-set performance. The accompanying changes in LOI, tensile strength, smoke generation, and heat release were formulation dependent and are therefore interpreted as property responses associated with, rather than direct measures of, radiation-induced network development. MH-containing systems generally showed lower smoke and heat-release parameters than the corresponding ATH-based systems, while ZB-containing formulations showed further performance improvements within the investigated formulation space. Because ZB addition also increased total inorganic content from 60 to 64 wt%, these changes cannot be assigned exclusively to a specific ZB interaction. Overall, the results establish absorbed EB dose as a controllable radiation-processing variable whose effects on macroscopic performance depend strongly on flame-retardant composition.
