Maximizing the detection and localization of Ga-67 tumors in thoracic SPECT MLEM(OSEM) reconstructions

Iterative reconstruction algorithms are usually regularized by applying a p enalty function, by postfiltering, or by halting the algorithm after some n umber of iterations. It is difficult to know a priori what is the optimal c ombination of regularization methods. One method of selection is to use a l ocalization receiver operating characteristic (LROC) study. LROC extends RO C analysis by incorporating a search-and-localize component into the task. Using LROC, we investigated the combination of iteration number and 3D Gaus sian filter which will maximize the detectability and localization accuracy of l-cm gallium-avid tumors in maximum-likelihood (ordered-subset) expecta tion-maximization SPECT reconstructions of the chest region. In our study, five observers read 200 images per test condition, divided equally over two reading sessions. In each case, the observer indicated the most probable l ocation of the lesion in the image and provided a confidence rating (as in an ROC experiment). The best observer performance was achieved using a reco nstruction with 8 iterations of MLEM followed by filtering with a 3D Gaussi an filter having a 4-pixel (1.3 cm)FWBM, although the difference between th is test condition and others is not significant over a broad range of the p arameters considered.