Neural Network-Driven Parameter Calibration for 3D CdZnTe Detector Simulators

Open Webpage

A. Valverde Mahou, S. Hauberg, I. Kuvvetli, S. Ringsborg Howalt Owe, M. Kossakoswki, G. Arvanitidis

Published: 01 Jan 2025, Last Modified: 27 Jan 20262025 IEEE Nuclear Science Symposium (NSS), Medical Imaging Conference (MIC) and Room Temperature Semiconductor Detector Conference (RTSD)EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Accurately localizing photon interactions in radiation detectors is essential for imaging applications. This study explores the use of neural networks to enhance the localization of photon interactions by inverting a photon interaction simulator for a 3D CdZnTe drift strip detector developed for space application. We form a differentiable implementation of the simulator to finetune the simulator's parameters, aligning them more closely with real-world laboratory data. Our empirical results demonstrate that this parameter optimization improves the neural network's performance in predicting interaction positions, leading to more precise localization and an enhanced spatial resolution for the detector.