Much of my current research effort is focused on developing methods to implement the task-based assessment of image quality for various medical imaging systems. I have participated in published work in this area for SPECT, MRI and OCT imaging systems. 1 am also currently involved with image quality projects for DOT and CT systems. In all of these areas our research team has developed mathematical observers, object and imaging system simulation methods, and task-based measures of observer performance that can be computed either directly from data or from reconstructed images. The tasks we have studied include detection of abnormalities, estimation of clinically relevant parameters, and tasks that combine classification and estimation. We have also developed some general theory on the computation of ideal-observer performance including MCMC techniques, the estimation ROC curve, and surrogate figures of merit based on Fisher information. Any computation of a task-based figure of merit involves some error, and we have published work on methods for quantifying this error and predicting the degree to which it can be reduced by increasing the number of cases and/or observers. We have recently published work on a general theory for the combination of linear compartmental models with dynamic imaging. The result is what we call global compartmental models to distinguish them from localized compartmental models which are often used in imaging studies. This in turn has led to some work on estimating the identifiable kinetic parameters from imaging data in both local and global compartmental models. The question of which kinetic parameters are identifiable from a given data set is often overlooked in imaging studies. In addition, even for identifiable parameters, the methods used to estimate them are often suboptimal from a statistical point of view.
Image quality, image reconstruction, image science, information theory, statistical decision theory, pharmacokinetics