Development of a 3D position sensitive scintillation detector using neural networks

This work presents a novel method of exploiting light division in a scintillating crystal to localise in three-dimensions the point of interaction of an impinging photon. For this, light output of a detector block is mapped for known irradiation positions of a collimated photon beam and used to train a set of three independent multilayer neural networks. Spatial resolutions obtained with a Monte Carlo simulation of a cubic 2×2×2 cm³ CsI(Tl) crystal fully covered with 24 Si PIN photodiodes are better than 2 mm fwhm in the three directions over the entire volume of the detector block. For comparison, the spatial resolution reconstructed in two dimensions from a simple Anger logic based on the photodetectors coupled to the rear side of the crystal is about 3.5 mm fwhm for the same experimental setup