The long term objectives of this project are: (1) to implement a Bayesian image processing (BIP) method as a unified reconstruction algorithm to facilitate quantitative single photon emission computed tomography (SPECT); (2) to implement the BIP method as usual statistical reconstruction algorithm to facilitate quantitation in position emission tomography (PET); (3) to develop and evaluate the BIP method as an effective imaging algorithm for cardiac imaging studies. Quantitative SPECT has been limited mainly by the inadequate number of detected photons in the projections containing Poisson noise, photon attenuation and scattering within patient body, and the variations in collimation variation. Different a priori source information models will be studied and incorporated into BIP to suppress the unpredictable image degrading effects of Poisson noise and accidental coincidences. For cardiac imaging studies, since the radioisotope concentration of the blood filling the cardiac chambers is much higher than that of the surrounding heart tissues, such quite restrictive a priori information will be investigated and utilized to remedy the image degrading effects due to blood moving and ventricular wall motion, as well as the non-uniform attenuation of the body. The compensation of attenuation and scattering will use an attenuation map of the body. The collimation variation will be compensated using a Gaussian function with different full-width-half-maximums corresponding to different distances from the detector. The a priori source information models, such as entropy, range of concentration levels, fuzzy patterns of concentration distribution etc, will be considered. BIP will be evaluated for convergence and stability as a function of iteration, accuracy of system model, and photon count density. It will be compared with other statistical estimators, such as maximum likelihood, maximum entropy, and maximum mutual information. Simulated and experimental projection data from phantoms will be reconstructed to evaluate BIP for improvement in lesion detectability, noise propagation, resolution recovery, and quantitation of update ratios for distributions which reflect clinical imaging geometries. Results from preliminary investigations are encouraging, indicating that BIP can provide quantitative compensation in reconstructed images and improve image quality by considering valid a priori information. Presently, BIP is the only statistical reconstruction algorithm which has demonstrated simultaneous compensation and the ability to utilize a priori information. This proposed evaluation of alternatives for BIP development is intended to help establish a strategy for enhancing the use of nuclear isotope imaging as quantitative function imaging.
