Robust Maximum- Likelihood Position Estimation in Scintillation Cameras

Abstract

The classical maximum-likelihood (ML) estimator for the position of a scintillation event in a gamma camera, as derived by Gray and Macovski in 1976, requires exact knowledge of the light-spread function (LSF) of each photomultiplier tube. In practice, one must determine each LSF from noisy measurements corrupted by Poisson noise, quantization error, and electronic noise and bias. Since the ML position estimator involves derivatives of each LSF, even small measurement errors can result in degraded estimator performance. In this paper we derive a robust ML position estimator that accounts for the statistical uncertainty in LSF measurements. The form of the robust estimator diminishes contributions from the tails of the LSF, where the relative measurement errors are the largest.