ESTIMATION STRATEGIES FOR NUCLEAR MEDICAL IMAGING

描述（改编自申请人摘要）：长期目标

DESCRIPTION (Adapted from Applicant's Abstract): The long term aim of this research is to develop techniques for in vivo quantification of radioisotope localization in small animals in order to speed and simplify the design and evaluation of new radiolabeled pharmaceuticals. As part of another project, the applicants are completing the construction of a high-resolution single photon ring tomograph modeled after SPRINT, but specifically designed for imaging 99mTc, 123I, and 131I in mice and rats with millimeter resolution. The applicant have proposed to use uniform Cramer-Rao (CR) lower bound as a tool to evaluate the intrinsic quality of data obtained from an imaging system, and to compare the performance of several estimators for the task- specific objectives of determining whole organ or tumor activity concentration and distribution within the organ. The initial application and testing will be for the case of small animals, but it is anticipated that the methods developed will be applicable to the imaging of humans. Methods will be developed to compute the CR bound when the Fisher matrices are poorly conditioned. Several measures derived from the bias image will be compared to the bias gradient length to determine their usefulness as a measure of image quality. The uniform CR bound will be used to characterize imaging system performance, estimator performance, and the robustness of the system and object models. Two relatively recent estimation algorithms, Penalized Weighted Least Squares, (PWLS), and Space-Alternating Generalized Expectation- maximization, (SAGE), will be evaluated and compared, and performance limits determined for the quantification tasks both for the case where side information is available, e.g. in the form of estimates of organ and tumor boundaries obtained from nuclear magnetic resonance (NMR) images and were no such information is available. Methods will be devised to parameterize boundaries from NMR images, construct estimators for joint estimation of boundaries and specific activity, and characterize the bias and variance of these estimates. Theoretical developments will be tested by simulation and phantom and small animal imaging experiments.