Speaker
Description
MeV gamma-ray imaging is important for astrophysics, nuclear physics, nuclear security, and medical physics, but conventional collimated systems often suffer from low detection efficiency. Compton imaging provides a promising alternative by reconstructing the incident gamma-ray direction from event topology.
In this study, we investigate a compact hybrid Cherenkov–scintillation Gamma tracker using Geant4 simulation and Transformer-based reconstruction. The simulated detector consists of a central 5 cm CsI scatterer, surrounded by BGO absorber-bar arrays and SiPM readout planes. The current preliminary study focuses on 1 MeV gamma rays from a fixed source position, selecting events with the first interaction in the CsI scatterer and full energy deposition in the BGO absorber. A multi-task Transformer model is developed to reconstruct event-level quantities from variable-length SiPM photon-hit sequences. The model simultaneously predicts the recoiled electron direction, electron energy, scattered photon direction, and incident gamma-ray direction. The incident gamma opening-angle error is used as the main angular-resolution metric.
Across several simulated incident-direction configurations, including random, fixed-grid, shifted-grid, and radial-ring samples, the model achieves stable incident gamma reconstruction with p68 angular resolution of about 1.6°–1.9° and p90 containment of about 2.4°–2.8°. These preliminary results demonstrate the feasibility of combining a compact hybrid detector concept with machine-learning-based reconstruction for event-by-event MeV gamma-ray imaging.
| 请选择分会 | 粒子物理实验技术 |
|---|
