Preliminary examination of the use of case specific medical information as "prior" in Bayesian reconstruction
In this paper we attempt to answer one specific question regarding the role of "prior" distributions in Bayesian reconstructions in Emission Tomography. The question is: Can prior information on some areas of an imaging field improve the results of a reconstruction in other areas of the same image? We answer the question for the specific case of a simple image with a large square of activity surrounding a smaller square, with different ratios of activity in the two regions. For the case of ratios of 10:1 we find that feasible MLE reconstructions exhibit some effect in the internal region due to the presence of the external region and that those effects can be reduced by a factor of ∼0.5 by using as prior information the known mean of activity in the large outer region. Two methods of obtaining reconstructions with pixel dependent constraints are developed in order to obtain the above results. The conclusion of the study is, principally, that "priors" should have a local action, i.e., should apply directly to the region of interest if a substantial improvement in reconstruction quality is to be achieved.
KeywordsBayesian reconstruction Prior information Medical prior information Maximum a posteriori (MAP) Successive Substitutions
Unable to display preview. Download preview PDF.
- Chen C-T, Ouyang X, Ordonez C, Wong WH, Hu X and Metz CE (1990). Sensor Fusion in Image Reconstruction. pres. 1990 IEEE Nucl. Sci. Symp., Arlington, VA.Google Scholar
- Geman S and Geman D (1984). Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images. IEEE Trans. Pattern Anal. Machine Intel., PAMI-6, 721–741.Google Scholar
- Hildebrand FB (1974). Introduction to Numerical Analysis, 2nd ed., McGraw-Hill, New York.Google Scholar
- Huesman RH, Gullberg GT, Greenberg WL and Budinger TF (1977). Donner Algorithms for Reconstruction Tomography. Publ. 214, Lawrence Berkeley Laboratory.Google Scholar
- Levitan E and Herman GT (1987). A Maximum A posteriori Probability Expectation Maximization Algorithm for Image Reconstruction in Emission Tomography. IEEE Trans. Med. Imaging, MI-6, No. 3, 185–192.Google Scholar
- Liang Z, Jaszczak R and Greer K (1989). On Bayesian Image Reconstruction from Projections: Uniform and nonuniform a priori source information. IEEE Trans. Med. Imaging, MI-8, 227–235.Google Scholar
- Llacer J and Veklerov E (1989). Feasible Images and Practical Stopping Rules for Iterative Algorithms in Emission Tomography. IEEE Trans. Med. Imaging, MI-8, 186–193.Google Scholar
- Llacer J and Bajamonde AC (1990). Characteristics of Feasible Images Obtained from Real PET Data by MLE, Bayesian and Sieve Methods. To be publ. Proc. of SPIE 1990 Int. Symp. on Optical and Optoelectronic Applied Science and Eng, San Diego, CA.Google Scholar
- Nuñez J and Llacer J (1989). Maximum Entropy and the Concept of Feasibility in Tomographic Image Reconstruction. Proc. SPIE Medical Imaging III Conf., 1090, 359–372.Google Scholar
- Nuñez J and Llacer J (1990). A Fast Bayesian Reconstruction Algorithm for Emission Tomography with Entropy Prior Converging to Feasible Images. IEEE Trans. Med. Imaging, MI-9, No. 2, 159–176.Google Scholar
- Shepp LA and Vardi Y (1984). Maximum Likelihood Reconstruction for Emission Tomography. IEEE Trans. Med. Imaging, MI-1, 113–122.Google Scholar
- Veklerov E and Llacer J (1987). Stopping rule for the MLE algorithm based on statistical hypothesis testing. IEEE Trans. Med. Imaging, MI-6, 313–319.Google Scholar