The proposed research extends our previous work on evaluation of scatter and attenuation correction methods using task-dependent criteria to more complex models incorporating realistic anatomy. Estimation performance from simulated and phantom images, reconstructed by filtered backprojection of data corrected by these methods, will be compared to the best possible performance from the measured data. This ideal performance will be represented by the Cramer-Rao lower bound (CRB) on variance of estimated model parameters. Values of the CRB will be calculated from the expected sinograms, which will be determined by Monte Carlo (MC) simulation. Any difference between actual and ideal performance would represent the potential advantage to be gained by iterative reconstruction. We will also continue our development of task-based figures-of-merit, particularly those which apply to low-count conditions or to tasks which incorporate a priori information. The imaging tasks on which performance measures will be based include clinical classification tasks related to Alzheimer's disease, as well as the mathematically modeled estimation tasks. The proposed research is also directed toward several objectives beyond the major goal of evaluation of correction methods. The first of these is to improve estimation performance by altering the data collection strategy. An analysis completed during the previous project period has shown that collimators with centrally peaked sensitivity functions can significantly improve task performance. During the proposed project period, we will validate these improvements by phantom experiments. We are also addressing the question of the value of a priori information, such as intermodality transfer of structure boundaries, in estimation tasks. We will evaluate, using a clinical classification task, a method, developed during the previous project period, of estimating structure activity using such information. Progress in these first two areas is expected to have significant clinical impact. At present deep brain structures, although known to be involved in dementia, are rarely examined because of difficulties in quantifying their activity. These problems are related to high noise levels at the center of the brain, the small size of the structures, and imprecision of defining boundaries of these structures from the functional SPECT images, the very issues addressed by the proposed research. Finally, we will continue the development of methods to handle estimation performance at low signal-to-noise ratio (SNR), where the CRB is not attainable.
