The overall goal of this grant is to develop a rigorous theoretical and experimental frame-work for objective assessment of image quality and to apply it to the development of practical, clinically useful algorithms for single-photon emission computed tomography (SPECT). Essential to the full realization of this goal is adequate computer power. The grant currently uses a distributed-memory multiprocessor system, but it is now proposed to use a cluster of workstations. Recent advances in networking technology make it possible to meet the interprocessor communications requirements with a cluster, and algorithms developed for the cluster can easily be transported to other institutions with multiple workstations. Another key ingredient in the proposed research is development of objective, task-based methods for the assessment of image quality. Continuing the productive research in this area in the previous funding period, the proposed research will refine existing mathematical-observer models and validate them by means of extensive psychophysical studies. New methods for realistic simulation of SPECT images will be developed and used to generate images for investigations into image quality. The performance of human observers on complex detection tasks will be studied and compared to the model observers in order to elucidate the factors that limit the performance of human observers. These methods of image-quality assessment will be used to evaluate a variety of existing SPECT reconstruction algorithms and determine whether different algorithms lead to different signal-detection performance by model or human observers. New SPECT reconstruction algorithms will be developed that accurately account for scattered radiation. Bayesian reconstruction algorithms with various forms of prior information will be investigated, and the priors them-selves will be evaluated in terms of objective image quality. To expand the theoretical understanding of image quality and its relation to reconstruction algorithms and system design, the role of artifacts, nonlinear constraints, nonlinear reconstruction algorithms, and modeling errors will be investigated.
Showing the most recent 10 out of 31 publications
