Embedded artificial intelligence for position sensitivity in thick scintillators

Abstract

The study and application of AI algorithms for photon interaction position reconstruction in a 3-inch thick scintillator is here presented. Decision Tree and Neural Network algorithms have been designed to retrieve the 3D position inside the scintillator in order to compensate for the Doppler Effect in nuclear physics measurements. A full explanation of the training phase, which is critical in imaging applications with thick crystals, both for planar coordinates (XY plane) and Z-Axis, is reported. Furthermore, an AI Filter to improve spatial resolution on the XY plane will be introduced showing how it has been trained and all the pros and cons about this new solution. Finally, algorithms have been implemented in an Artix-7 FPGA to allow Real Time position reconstruction, and a comparison is reported.