Properties of minimum cross-entropy reconstruction of emission tomography with a morphologically based prior

S. Som*, B. F. Hutton, M. Braun

*Corresponding author for this work

    Research output: Chapter in Book/Report/Conference proceedingConference proceeding contributionpeer-review

    Abstract

    We have studied the properties of a minimum cross-entropy (MXE) algorithm for emission tomography reconstruction with a morphological prior. MXE is formulated with two terms: a maximum likelihood expectation maximization term and a penalty term for regularization within morphologically defined boundaries. The relative emphasis put on the two competing terms is controlled by the regularization constant β. Edge resolution and noise were compared for reconstructions with and without corresponding prior. The prior leads to significant edge enhancement with edge resolution converging to a theoretical limit independent of β. Normalized standard deviation (NSD) and resolution both illustrate that regularization within boundaries behaves predictably with more smoothing for larger β. Application of ordered subsets (OS) was also investigated. For OS, edge enhancement is fully preserved but NSD increases for low subset size. Results demonstrate that OS is applicable to MXE provided subset size is greater than 4. OS-MXE has appealing properties for regularized reconstruction.

    Original languageEnglish
    Title of host publicationIEEE Nuclear Science Symposium & Medical Imaging Conference
    EditorsO. Nalcioglu
    PublisherInstitute of Electrical and Electronics Engineers (IEEE)
    Pages1738-1742
    Number of pages5
    Volume2
    Publication statusPublished - 1997
    EventProceedings of the 1997 IEEE Nuclear Science Symposium - Albuquerque, NM, USA
    Duration: 9 Nov 199715 Nov 1997

    Other

    OtherProceedings of the 1997 IEEE Nuclear Science Symposium
    CityAlbuquerque, NM, USA
    Period9/11/9715/11/97

    Fingerprint Dive into the research topics of 'Properties of minimum cross-entropy reconstruction of emission tomography with a morphologically based prior'. Together they form a unique fingerprint.

    Cite this