### Abstract

The autocovariance matrix provides full description of nonwhite and nonstationary SPECT image noise in terms of magnitude and texture and is useful in calculating image quality indices. The height and shape of the autocovariance function provide information about the noise magnitude and texture respectively. A method is presented for computing the autocovariance matrix at any point in a SPECT image obtained from noniterative reconstruction if the reconstruction method and noise-free projection data are given and the projection noise is Poisson distributed. Simple disk phantoms as well as a more complicated cardiac-chest phantom are used to demonstrate how different reconstruction filters affect the nonstationary noise properties. We have studied the effects of attenuation compensation methods such as arithmetic mean, geometric mean, and intrinsic filtering on the reconstructed image noise properties. It is shown that for a uniformly emitting and attenuating disk the noise magnitude and signal-to-noise ratio varies as a function of radial position and attenuation compensation algorithm. The noise texture exhibits much less radial dependence. For the simple disk phantoms noise texture largely reflects the reconstruction filter but in the case of the complex cardiac-chest phantom the texture is dependent on both the projection data and filter function used for reconstruction. The understanding of nonstationary image noise characteristics is useful in filter design and choice of compensation methods in SPECT image reconstruction.

Original language | English (US) |
---|---|

Number of pages | 1 |

Journal | Annals of biomedical engineering |

Volume | 19 |

Issue number | 5 |

State | Published - Dec 1 1991 |

Event | 1991 Annual Fall Meeting of the Biomedical Engineering Society - Charlottesville, VA, USA Duration: Oct 12 1991 → Oct 14 1991 |

### ASJC Scopus subject areas

- Biomedical Engineering

## Fingerprint Dive into the research topics of 'Nonstationary noise characterization for SPECT images produced using several reconstruction filters and attenuation compensation techniques'. Together they form a unique fingerprint.

## Cite this

*Annals of biomedical engineering*,

*19*(5).