The Effects of Incorrect Modeling on Noise and Resolution Properties of ML-EM Images.

The effects of incorrect compensation for collimator blur in single-photon emission computed tomography (SPECT) images are studied in terms of the noise and resolution properties of the reconstructed images. Qualitative analysis of images of the Hoffman brain phantom reconstructed using nonlinear maximum-likelihood-expectation maximization (ML-EM) show the behavior of longer noise correlations for high-pass filtered images. These qualitative observations are confirmed with more quantitative noise measures. The differences also appear in images reconstructed using linear Landweber iteration. However, the signal-to-noise ratio, in terms of the noise-equivalent quanta, remains largely unchanged. We conclude that the compensation model affects SPECT image properties, though the effect on human task performance remains to be studied.