From Light to Energy: Machine Learning Algorithms for Position and Energy Deposition Estimation in Scintillator–SiPM detectors
Yoav Simhony, Alex Segal, Ofer Amrani, Erez Etzion · 2025
Read From Light to Energy: Machine Learning Algorithms for Position and Energy Deposition Estimation in Scintillator–SiPM detectors by Yoav Simhony, Alex Segal, Ofer Amrani, Erez Etzion (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-00499.txt.
Introduction
Yoav Simhon, Alex Segal, Ofer Amrani, and Erez Etzion improve small particle detectors built from a prism-shaped scintillator with silicon sensors at the corners. Light from a passing particle spreads through the prism, and earlier analytic methods read that pattern to estimate where the particle hit and how much energy it left, with moderate accuracy. The team replaces the analytic step with machine learning.
Simulated cosmic muons and energetic oxygen ions, generated with the GEANT4 toolkit, trained gradient boosting models such as XGBoost and LightGBM. Both beat the analytic baseline at position and energy estimation. Hybrid strategies were tested too: plain hybrid boosting added little, but a probing strategy gave measurable gains, including for energy transfer estimates. The results suit compact detectors for physics, tomography, and space missions.
Catalog record
| CREATOR | Yoav Simhony, Alex Segal, Ofer Amrani, Erez Etzion |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-00499
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
From Light to Energy: Machine Learning Algorithms for Position and Energy Deposition Estimation in Scintillator–SiPM detectors Yoav Simhony, Alex Segal, Ofer Amrani, Erez Etzion Abstract Scintillator-SiPM Particle Detectors (SSPDs) are compact, low-power devices with applications including particle physics, underground tomography, cosmic-ray studies, and space instrumentation. They are based on a prism-shaped scintillator with corner-mounted SiPMs. Previous work has demonstrated that analytic algorithms based on a physical model of light propagation can reconstruct particle impinging positions and tracks and estimate deposited energy and Linear Energy Transfer (LET) with moderate accuracy. In this study, we enhance this approach by applying machine learning (ML) methods, specifically gradient boosting techniques, to improve the accuracy of spatial location and energy deposition estimation. Using the GEANT4 simulation toolkit, we simulated cosmic muons and energetic oxygen ions traversing an SSPD, we trained XGBoost and LightGBM models to predict particle impinging positions and deposited energy. Both algorithms outperformed the analytic baseline. We further investigated hybrid strategies, including hybrid boosting and probing. While hybrid boosting provided no significant improvement, probing yielded measurable gains in both position and LET estimation. These results suggest that ML-driven reconstruction provides a powerful enhancement to SSPD performance. 1 Introduction The ability to reconstruct charged particle trajectories and energy deposition with compact detectors has significant implications for cosmic-ray studies, spaceborne instrumentation, and applied particle physics. Scintillator detectors instrumented with Silicon Photomultipliers (SiPMs) are particularly attractive due to their compactness, low power consumption, and robustness against environmental conditions. A recent contribution by [ 1 ] described the Scintillator–SiPM Particle Detector (SSPD). This detector was introduced in [ 2 ] and later validated through in-orbit tests aboard the International Space Station [ 3 ] . The SSPD consists of a truncated prism scintillator read out by four SiPMs at its corners. The analytic model for this configuration translated sensor intensity signals into estimates of particle impinging position and energy deposition. By applying an SSPD array, one can achieve particle track reconstruction and Linear Energy Transfer (LET) estimation, resulting in millimeter-scale localization and LET uncertainties of 5-10\% under favorable conditions. Due to the geometrical distribution of the SiPMs and due to some numerical instability of the algorithm described in [ 1 ] , it underperforms in certain areas of the scintillator, mostly for particle events localized at the extremity of the SSPD. In parallel, the broader field has increasingly recognized the promise of machine learning methods for scintillator-based event reconstruction. For example, a 2021 study [ 4 ] demonstrated that boosted decision trees and neural networks can outperform analytic techniques in predicting event parameters from scintillation light distributions, particularly in complex or high-noise scenarios. This work demonstrated ML implementations to detector response modeling showing that data-driven methods are capable of capturing effects beyond the reach of analytic formulations. In their work, the authors compared boosting algorithms such as XGBoost, probabilistic neural networks and an analytic model in various geometries and sizes of scintillators in the case of muon particles on a track perpendicular to the scintillator’s surface. They showed that the best results are achieved using a square scintillator and XGboost. Related research on the application of machine learning to particle tracking and detector optimization has demonstrated promising results across various detector types. For example, Yaary et al. [ 5 ] compared tree-based algorithms and neural networks for enhancing real-time tau-lepton selection in proton–proton collisions. Additional studies have explored similar approaches in diverse contexts, including event reconstruction, and position estimation and particle classification [ 6 , 7 , 8 , 9 , 10 , 11 ] . Motivated by these developments, the present study seeks to combine the analytic physics-based algorithm described in [ 1 ] with machine learning algorithms. Specifically, we investigate whether gradient boosting algorithms - XGBoost and LightGBM - can improve track reconstruction and LET estimation in the SSPD configuration of [ 1 ] . Beyond direct application of boosting, we also explore hybrid strategies that integrate analytic-physics-based and ML-driven approaches, including hybrid boosting and probing [ 12 ] . By comparing analytic, pure ML, and hybrid models, we aim to assess the practical gains, in terms of estimation accuracy in SSPDs, achievable through machine learning. We test our methods, similarly to [ 1 ] , on GEANT4 [ 13 ] simulated muon particles and high energy oxygen ions. 2 Methods 2.1 Detector Geometry and Simulation We adopted the geometry from the original work in [ 1 ] (see Figure 1 ). The detector comprises a prism-shaped polyvinyl-toluene scintillator with truncated corners (dimensions: 70 mm × 70 mm × 6.7 mm), with each corner coupled to one of four SiPMs. The scintillator’s dimensions and optical properties were modeled according to published specifications [ 1 , 14 ] . We generated two datasets using GEANT4 (version 10.7): • Muon dataset: 10^{5} simulated 4 GeV muons with randomized incidence positions and angles. • Oxygen dataset: 10^{5} simulated 18 GeV oxygen ions with randomized incidence positions and angles. For each simulated event, we recorded the number of photons measured by each SiPM (photons vector), the particle’s energy deposition along the track within the scintillator, and the location of incidence. Figure 1 : Top view of the SSPD [ 1 ] . The four SiPMs are positioned at the truncated corners of a prism-shaped scintillator. A particle impinging on the scintillator at position (x_{in},y_{in}) is shown, along with the angles (\alpha_{11},\alpha_{12},\alpha_{21},\alpha_{22}) defined between the impinging point and the edges of the four SiPMs. 2.2 The Physics-Based Analytic model The physics-based analytic model (AM) introduced in [ 1 ] is based on the fact that the number of photons reaching each SiPM is proportional to the angle \alpha_{ij} between the edges of the SiPM and the particle incidence location ( \left(\alpha_{11},\alpha_{12},\alpha_{21},\alpha_{22}\right) in Figure 1 ): N=C_{s}\cdot\textrm{LET}\cdot\frac{\alpha}{2\pi}, (1) where LET is the particle’s linear energy transfer and C_{s} is a constant dependent on the scintillator’s geometry, density and efficiency in converting particle deposited energy to electromagnetic energy (photons). Using this model, one may pre-compute the resulting measurements (up to a multiplicative constant) on a grid with a given resolution, and use the obtained results to estimate the particle’s impinging location. For details see Algorithm 1 in [ 1 ] . After the particle’s impinging location is obtained, the particle’s LET may be approximated using the analytic model. The analytic approach is computationally efficient and physically interpretable, making it suitable for deployment in resource-constrained environments such as space missions. However, due to numerical instability of the algorithm, nonlinearities in events where the particle impinges the scintillator very close to a SiPM, and due to the variability of the Geometric Dilution of Precision (GDOP) across the SSPD’s area it under-performs at certain areas of the scintillator, mostly near the edges, where the GDOP is small (see Figure 2 (a)). 3 Machine learning methods 3.1 Gradient Boosted Regression (XGBoost and LightGBM) We first explore direct regression from photon vectors to incidence location using gradient boosting. Two widely used frameworks are considered: • XGBoost is a highly effective tree boosting algorithm [ 15 ] . It works by constructing additive ensembles of decision trees using second-order gradient information. It is robust to nonlinear feature interactions and performs well on structured datasets. Since its introduction, XGBoost,