Task-Oriented and Study-Dependent Optimization of 3D and Fully 4D Reconstruction Parameters for ~(18)FFDG Imaging

摘要

3D and fully 4D dynamic PET iterative image reconstructions are usually performed with a predefined set of reconstruction parameters (number of iterations, level of smoothing, number and type of basis functions used in the 4D reconstruction). These parameters are often chosen without due attention to i) the specific task (reason for the scan) and ii) the unique characteristics of the acquired data at hand. For the task of functional parameter estimation (such as glucose metabolic rate), both the image reconstruction parameters and the statistics of the unique dataset have a significant impact on the final estimates. As such, there is a need for a more systematic approach to reconstruction parameter selection. This work investigates the impact of using both 3D and fully 4D reconstruction on kinetic parameter estimation (influx rate constant (K_i)) for an [~(18)F]FDG brain imaging data set acquired on the high resolution research tomograph (HRRT). Using a data-subsetting approach, it is shown that the choice of iteration number significantly affects the final kinetic parameter estimates (influx rate constant (K_i)) and hence the iteration number can be more optimally selected for each unique data set to deliver lower errors in the parameter estimates. As such, the approach advocates a study-dependent and task-oriented early stopping of the EM algorithm.