Superresolution tomography using a moving detection system and convex projections

Abstract

Tomographic imaging systems such as those used in PET and SPECT are typically limited in their performance by the size of the detectors used in acquiring the data. By analogy to findings in the context of optical image-plane applications,1-2 we show that, by moving a tomographic detection system to expand the measurement set, image detail can be recovered by methods such as convex projections. It is demonstrated that, by underrelaxing the data constraints in a convex projection algorithm, excellent results can be obtained in the presence of the high levels of quantum noise typically encountered in the motivating applications. In addition, the method presented is shown to be robust to reasonable misspecifications of the spatial response functions of the imaging system. A “wobbling” PET system is used for purposes of demonstration; however, the concept can be applied in a variety of ways to various tomographic imaging configurations to improve achievable spatial resolution, either in image space or in projection space.