From Light to Energy: Machine Learning Algorithms for Position and Energy Deposition Estimation in Scintillator-SiPM detectors

TL;DR

The paper tackles reconstructing particle impinging positions and energy deposition in scintillator-SiPM detectors using gradient-boosted regression and hybrid physics-ML strategies. It leverages a physics-based analytic model, $N = C_s \cdot \text{LET} \cdot \frac{\alpha}{2\pi}$, and augments it with XGBoost and LightGBM to improve localization and LET estimation on GEANT4-simulated muons and high-energy oxygen ions. The key contributions are (i) a systematic comparison of pure ML, pure analytic, and hybrid approaches (Boosting and Probing), (ii) demonstrated RMSE improvements of roughly 30% over the analytic baseline and additional gains from probing, and (iii) evidence that edge artifacts can be mitigated while preserving interpretability and speed. The results indicate that ML-enhanced reconstruction can significantly improve SSPD performance, with practical implications for compact, low-power detectors in space missions and related applications.

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.

Figures (6)

• Figure 1: Top view of the SSPD Simhony2024. 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.
• Figure 2: Comparison of localization error heatmaps for the (a) analytic model, (b) pure machine‑learning, and (c) hybrid approach. Each subfigure uses the same color scale to facilitate direct comparison.
• Figure 3: Comparison of localization error heatmaps in a centered $50mm \times 50mm$ area of interest of the SSPD for (a) the analytic model, (b) pure machine‑learning using LightGBM, and (c) hybrid approach with LightGBM. Each subfigure uses the same color scale to facilitate direct comparison.
• Figure 4: Mean error direction and size vector maps for the (a) analytic, (b) machine learning and (c) the hybrid models. Notice that the errors are distributed differently on each axis, depending on the location. This difference is due to variability of GDOP across the area of the scintillator and nonlinear effects very close to the SiPMs.
• Figure 5: (b) Localization error heatmap for the hybrid probing model assuming a theoretical perfect classifier, i.e., a classifier that always selects the model yielding the most accurate estimate for each measurement. For convenience, subfigure (a) reproduces the localization error map already shown in Figure \ref{['fig:oxygen_sim_lgb']}.