Noise properties of filtered-backprojection and ML-EM reconstructed emission tomographic images

The authors compare noise properties of emission tomographic images reconstructed using maximum likelihood-expectation maximum (ML-EM) and filtered-backprojection (FB) algorithms. Noise comparisons were made in terms of the covariance matrix, which gives information on the noise magnitude (variance) and noise texture (covariance). Noise properties were studied as a function of iteration for ML-EM and as a function of noise apodization filter for FB. It is shown that little noise variance is spread from high count density regions into low count regions and that the noise is correlated over shorter distances than for FB. It is concluded that ML-EM reconstruction has the potential to exceed FB in terms of detectability in low count density regions of emission tomographic images