Optimization of CeBr[formula omitted] position-sensitive calorimeter module

Abstract

For the next generation of MeV range gamma-ray telescopes, position sensitive calorimeters based on a monolithic scintillator coupled to a pixelated photodetector could be an important building block. In this paper, we present the optimization of the position reconstruction algorithms using machine learning, for a detector based on a 51×51×10mm3 CeBr3 crystal. For that purpose, we used an automated test bench and collimated radioactive sources to generate experimental data of known energy and position by irradiating the detector with gamma rays. We found in these data different gamma-ray interaction morphologies for which position reconstruction algorithms perform differently, and we developed an algorithm to automatically classify them. We also conducted an extensive optimization of the artificial neural networks that perform the 3D position reconstruction using the Keras Python library with Theano backend. We found that at 662keV, 90% of events have a morphology that facilitates position reconstruction. The optimized position reconstruction algorithms give for those events a rms error in the plane of the detector of 1.8mm on each axis. The rms error in the depth of the crystal is found to be 2mm.