? Iterative reconstruction algorithms that significantly improve image quality over filtered backprojection methods have been developed by us and others. However, the full potential of these algorithms have not yet been realized. For example, iterative reconstruction methods can incorporate sophisticated modeling of the expected image that is specific for classes of clinical applications, and use of iterative methods can substantially remove constraints on PET instrument designs (e.g., allow the use of non-standard detector geometries). The goal of this grant is to develop a theoretical framework that can be used in practice to obtain the optimum iterative reconstruction algorithm as well as the optimum PET instrument design for each clinical application and to validate the new approaches in order to attain the full potential of PET. The potential application of this work is enabled by the availability of high performance computers that allow the use of accurate and statistically sophisticated iterative reconstruction algorithms. We propose to incorporate stochastic models for the target and background as a major extension of our previous work. The new approach to the optimization of the reconstruction algorithm will use the initial noisy data set to estimate the resolution and the noise characteristics of the target and the background at each element of the image. Numerical observers will be used to analytically compute task-specific figures of merit. A spatially variant image prior model will then be designed to achieve the optimal lesion detection and quantitation across the whole region of interest. Finally, the theoretical results of the optimum reconstruction, combined with ensemble information of patients and of the disease pertaining to a specific organ, will be used to discover the optimal instrument design for each clinical task. We will study both whole-body PET systems and application specific PET systems with non-standard geometries (e.g., PET for imaging breast and prostate). All results will be validated using Monte Carlo simulations, phantom scans, and real patient data from a whole body scanner. ? ?
Showing the most recent 10 out of 47 publications