Evaluation of NEUROSPECT reconstruction algorithms: deterministic and statistical errors

The authors have recently developed a novel iterative reconstruction algorithm (IRA) for the reconstruction of images from projection data which have been degraded by convolution of the true radioactivity distribution with a finite-resolution distance-dependent detector response. When the reconstructed images from 25 or 50 statistically independent noisy realizations were averaged, there was a slight degradation relative to the noise-free case. Although the performance of the IRA was still far superior to that of filtered backprojection for a single realization in the presence of noise, there was a further degradation for an IRA image reconstructed from a single noisy realization. In order to determine a strategy for reducing this degradation, the authors present an analysis of the three components (two of which are deterministic and one of which is statistical) which contribute to the overall error