Comparison of Nonlinear Position Estimators For Continuous Scintillator Detectors In PET

Several positioning algorithms are tested to extract position information from the measured scintillation light distribution generated in monolithic LSO blocks of various shapes and read out by a Hamamatsu S8550 APD array. The intrinsic detector resolutions of photons impinging at different angles e.g. 0deg, plusmn 10deg, plusmn 20deg, plusmn 30deg are studied. To this end, we evaluate the following positioning algorithms : Neural Networks trained with error back propagation (Levenberg-Marquardt), Neural Networks trained with an algebraic method and Support Vector Machines (SVM).