Emission computed tomography (ECT) is an important tool in the detection of lesions and in the study of in vivo physiological functions. It therefore plays a critical role in the localization of tumors and determination of their metastatic status. It also contributes to the development of new medical technologies requiring quantification of metabolic processes, for example monoclonal antibody protocols. The objective of this project is to maximize the amount of diagnostic information extracted from ECT data to perform these clinical tasks. The mechanism that will be employed to accomplish this objective is a statistical model for the physical processes that generate ECT data and their associated source distributions. The statistical model developed in this project incorporates parameters that directly relate to tumor detection and localization. Estimation of these parameters from observed ECT data permits quantification of mean lesion activity levels, lesion volumes, and provides for an. automatic segmentation of ECT images into activity differentiated regions. It also permits clinician interaction in the reconstruction of ECT images, thus increasing the accuracy of resultant reconstructions through the incorporation of clinical knowledge of anatomical structures. Finally, the proposed model permits information from cross-correlated (spatially matched) structural images (e.g. magnetic resonance or computed tomography images) to be incorporated into the reconstruction of functional ECT images. Importantly, this structural information need not be exactly specified: the statistical model permits shifts of predetermined boundaries according to the observed data likelihood function. An important aspect in the clinical evaluation of ECT images is the degree of certainty with which image features have been determined. The statistical model developed in this project permits direct assessment of such image uncertainty. Thus, statistical tests for the existence of tumors are available and confidence intervals for tumor activities and volumes can be constructed.
|Higdon, D M; Bowsher, J E; Johnson, V E et al. (1997) Fully Bayesian estimation of Gibbs hyperparameters for emission computed tomography data. IEEE Trans Med Imaging 16:516-26|