A preliminary study on 3D position reconstruction of monolithic crystal readout

Abstract

Compared to pixelated crystal detectors, monolithic crystal detectors present more advantages. Without reflective layer between crystal pixel, monolithic crystal detectors provide higher detection efficiency while being cheaper. Another important feature is that monolithic crystal detectors can provide depth of interaction (DOI) information, which can improve image quality in nuclear imaging systems. In this article, we aim to study a new positioning scheme of this kind of detector in order to push its applications in nuclear imaging systems. Fan-beam collimation and convolutional neural network are combined to reconstruct interaction points in crystal. Firstly, we built a monolithic detector model based on Geant4 and validated this positioning scheme with simulation data. Then, we constructed a monolithic GAGG(Ce) detector and evaluated its performance with experimental data. With proposed positioning scheme, in experiment, we obtained ~ 1.7 mm FWHM resolution on average in x-/y-direction, while ~ 2.5 mm FWHM resolution on average in DOI direction for a 33 × 33 × 10 mm3 monolithic GAGG(Ce) crystal coupled with 8 × 8 SiPM array. Through simulations and experiments, we validated our positioning scheme, i.e., fan-beam collimation and convolutional neural network. Convolutional neural network can reconstruct 3D positions of gamma-rays interaction points in a high resolution, and fan-beam collimation can provide high calibration efficiency; with combination of them, we established a high position resolution and high calibration efficiency positioning scheme for monolithic crystal detector.