Image reconstruction for positron-emission tomography when anatomical boundaries are known

Summary form only given, as follows. A method is proposed for improving estimates of radioactivity distributions obtained with PET (positron emission tomography) by utilizing anatomical information derived from MRI (magnetic resonance imaging) or X-ray CT (computered tomography). An algorithm is identified for computing regularized maximum-likelihood (ML) estimates of radioactivity distributions, subject to the constraint that a region of the image with known boundaries contains an unknown but constant intensity of radioactivity. Regularized ML estimates are computed using an expectation-maximization algorithm. A regularization method with spatially varying sieve and resolution kernels is used to prevent artifacts while preserving sharp boundaries where appropriate. The method is evaluated using a simulation and found to produce more accurate estimates than filtered backprojection or the usual ML algorithm. The variance of the constrained estimator is also studied.<>