Evaluation of multiclass model observers in PET LROC studies

Author

Gifford, H.C. ; Kinahan, P.E. ; Lartizien, C. ; King, M.A.

Author_Institution

Univ. of Massachusetts Med. Sch., Worcester, MA

Volume

7

fYear

2004

fDate

16-22 Oct. 2004

Firstpage

4068

Lastpage

4071

Abstract

We report on results of a localization ROC (LROC) study that evaluated several multiclass model observers as predictors of human tumor-detection performance. The study used image slices extracted from simulated whole-body FDG-PET volumes. Lesions were located in the liver, lungs, and soft tissue of a mathematical phantom, and the data simulation modeled a full-3D acquisition mode with an ECAT HR+ scanner. Reconstructions were performed with the FORE+AWOSEM algorithm. In the LROC study, observers read separate sets consisting of either coronal, sagittal, or transverse slices from the same set of cases. Multiclass versions of the channelized nonprewhitening observer demonstrated good agreement with the human observers

Keywords

image reconstruction; liver; lung; medical image processing; phantoms; positron emission tomography; tumours; ECAT HR+ scanner; FDG-PET volumes; FORE+AWOSEM algorithm reconstructions; PET LROC studies; channelized nonprewhitening observer; coronal; data simulation model; full-3D acquisition mode; human tumor-detection performance; image slices; lesions; liver; localization ROC study; localization receiver observer characteristics studies; lungs; mathematical phantom; multiclass model observers; sagittal; soft tissue; transverse slices; Biological tissues; Data mining; Humans; Imaging phantoms; Lesions; Liver; Lungs; Positron emission tomography; Predictive models; Whole-body PET;

fLanguage

English

Publisher

ieee

Conference_Titel

Nuclear Science Symposium Conference Record, 2004 IEEE

Conference_Location

Rome

ISSN

1082-3654

Print_ISBN

0-7803-8700-7

Electronic_ISBN

1082-3654

Type

conf

DOI

10.1109/NSSMIC.2004.1466787

Filename

1466787