Improved SPECT radioactivity quantification using MRI side information.

Abstract

Conventional evaluation of radiopharmaceuticals is an in vitro animal dissecting and counting procedure where multiple animals are sacrificed at different time points in order to determine tracer distribution as a function of time. Organ uptake, transplanted tumor growth and uptake can vary greatly from animal to animal. For this reason, measurements from different animals at different time points will give much less information than an in vivo measurement process, where the response of each tumor or organ can be followed over the entire time course. In vivo evaluation of radiotracer distribution using Single Photon Emission Computed Tomography (SPECT) can potentially overcome these disadvantages. However, the modest SPECT resolution impedes the potential of SPECT in vivo quantification. In this dissertation, we propose a joint estimation approach which incorporates high resolution, correlated Magnetic Resonance Image (MRI) anatomic region information into SPECT reconstruction to improve the SPECT functional tracer distribution quantification. This approach is developed using a joint penalized Maximum Likelihood objective function and a joint Space-Alternating Generalized EM algorithm to exploit the MRI region information that matches the SPECT functional information and to reduce artifacts caused by mismatched MRI anatomical information. Both the simulation and experimental results confirm that the joint estimation surmounts the difficulty of mismatch between SPECT and MRI data while retaining the benefit of matched MRI region information. One problem with the method is the sensitivity of the performance to a hyperparameter due to the non-convexity of the objective function. We have applied a deterministic annealing procedure to reduce the effect, and have discussed some possible solutions to the problem. This dissertation also addresses several topics related to Animal Single Photon RIng Tomography (AS-PRINT) SPECT. These include ASPRINT sampling geometry related correction for FBP reconstruction, system weights generation, mathematical model of ASPRINT and statistical model of projection acquisition process. Other issues, such as phantom design, image registration and segmentation, are also addressed. We have also proposed a set theoretic estimation method using EPC algorithm for SPECT. The algorithm gives a confidence region and an image estimate. Methods of accelerating the EPC algorithm and incorporating smoothness constraints are also developed.