Use of a neural network to exploit light division in a triangular scintillating crystal

Abstract This work presents a novel way of exploiting light division in a scintillating crystal to localise the point of interaction using a neural network. Because sensitivity at the centre of a positron tomograph would be significantly increased for a spherical geometry compared to a cylinder with the same area of detection, we explored the possibility of developing a position-sensitive triangular block detector that might be used ultimately as an element of a spherical shell. With 3 PMTs coupled on each side of a triangular BaF 2 crystal of about 8 cm 2 × 1 cm, we obtain spatial resolutions less than 4 mm FWHM over the whole surface of the triangle and energy resolutions of about 14% FWHM for 662 keV impinging photons. Experimental measurements are compared with Monte Carlo simulations and prospects for different configurations are discussed.