Neural network-based position estimators for PET detectors using monolithic LSO blocks

The impinging position of a 511 keV photon onto a continuous scintillator can be obtained from the light distribution measured by a pixelated photodetector such as avalanche photodiode (APD) arrays. This information is extracted using neural networks trained for events with a particular incidence angle. Using a 20×10×10 mm block of lutetium oxyorthosilicate mounted onto a S8550 Hamamatsu APD matrix we achieved an intrinsic resolution of 1.9 mm full-width at half-maximum (FWHM) for perpendicular incident photons and 2.6 mm FWHM at a 40° incidence angle. A possible implementation for tomographic imaging is presented.