Radiation oncology treatment planning presents several challenges to CT imaging, including imaging in the presence of complex metal intracavitary applicators and the need to noninvasively measure photon cross sections. Using synthetic sinograms in which noise, a polyenergetic spectrum, and scatter are modeled, the performance of three algorithms is compared: filtered back projection, deterministic iterative deblurring, and a modern model-based stochastic iterative reconstruction technique. The model-based algorithm performs well, but requires knowledge of metal object locations and shapes to fully eliminate streaking artifacts.