UMass Chan Medical School Faculty Publications

Title

Choosing anatomical-prior strength for MAP SPECT reconstruction to maximize lesion detectability

UMMS Affiliation

Department of Radiology

Publication Date

2007-10-26

Document Type

Conference Proceeding

Disciplines

Analytical, Diagnostic and Therapeutic Techniques and Equipment | Bioimaging and Biomedical Optics | Biological and Chemical Physics | Nuclear | Nuclear Engineering | Radiology

Abstract

With the widespread availability of SPECT/CT systems it has become feasible to incorporate prior knowledge about anatomical boundaries into the SPECT reconstruction process, thus improving observer performance on tasks of clinical interest. We determine the optimal anatomical-prior strength for lesion search by measuring area under the LROC curve using human observers. We conclude that prior strength should be chosen assuming that only organ boundaries are available, even if lesion boundaries will also be known some of the time. We also test whether or not the presence of anatomical priors affects the observer's strategy, and conclude that mixing images with and without priors does not hurt reader performance when priors are not available. Finally, we examine whether using an anatomical prior in SPECT reconstruction helps observer performance when the observer already knows the possible lesion location, and conclude for this task anatomical priors do not provide the same improvement seen in search tasks.

Keywords

Lesions, Image reconstruction, Computed tomography, Humans, Testing, Anatomy, Computational modeling, Iterative algorithms, Feedback, Predictive models

DOI of Published Version

10.1109/NSSMIC.2007.4437049

Source

Lehovich A, Gifford HC, Schneider PB, King MA. Choosing anatomical-prior strength for MAP SPECT reconstruction to maximize lesion detectability. IEEE Nucl Sci Symp Conf Rec (1997). 2007 Oct 26;6(1):4222-4225. doi: 10.1109/NSSMIC.2007.4437049. PMID: 19779594; PMCID: PMC2749318. Link to article on publisher's site

Related Resources

Link to Article in PubMed

Journal/Book/Conference Title

IEEE Nuclear Science Symposium conference record. Nuclear Science Symposium

PubMed ID

19779594

Link to Full Text

Share

COinS