IEEE Transactions on Nuclear Science · 1999 · 26 citations · 36 references
MedicineNuclear MedicineRadiologyDiagnostic ImagingDiagnostic RadiologyThoracic RadiologyCancer DetectionImage ReconstructionMedical Image AnalysisMedical ImagingThoracic Oncology ImagingTumor DetectionAbnormality LocalizationGaussian FiltersObserver Performance EvaluationReconstruction AlgorithmsRegularizationThoracic Malignancy DetectionThoracic Mass LocalizationTumor Localization
Iterative reconstruction algorithms are usually regularized by applying a penalty function, by post-filtering, or by halting the algorithm after some number of iterations. It is difficult to know a priori what is the optimal combination of regularization methods. One method of selection is to use a localization receiver operating characteristic (LROC) study. LROC extends ROC analysis by incorporating a search-and-localize component into the task. Using LROC, the authors investigated the combination of iteration number and 3D Gaussian filter which will maximize the detectability and localization accuracy of 1-cm gallium-avid tumors in maximum-likelihood (ordered-subset) expectation-maximization SPECT reconstructions of the chest region. In the authors' study, 5 observers read 200 images per test condition, divided equally over 2 reading sessions. In each case, the observer indicated the most probable location of the lesion in the image and provided a confidence rating (as in an ROC experiment). The best observer performance was achieved using a reconstruction with 8 iterations of MLEM followed by filtering with a 3D Gaussian filter having a 4-pixel (1.3 cm) FWHM, although the difference between this test condition and others is not significant over a broad range of the parameters considered.
36
Model observers for assessment of image quality.
Harrison H. Barrett, Jie Yao, Jannick P. Rolland et al. · Proceedings of the National Academy of Sciences · 1993 · 525 citations · Full text