Parameter estimation in a model based approach for tomographic perfusion measurement

The algorithmic aspects of tomographic estimation of perfusion model parameters from projections acquired with a rotating imaging system are presented. The approach is stated as a model based parameter optimization problem. A parametric model is used to describe the time attenuation curves of the dynamic image under consideration. Computed forward projections of this model considering the temporal and geometric properties of the acquisition system are compared to the measured projections in a least squares optimization. An iterative Lanczos algorithm is used to solve this problem numerically. An efficient implementation is presented which exploits a decomposition of the problem into the spatial and temporal domain. The ill-posed nature of the problem is tackled by a Tikhonov-type regularization method that makes use of a spatial/temporal decomposition along with a data driven optimization of the regularization parameter. The potential of this approach is shown for two different setups: a brain perfusion software phantom, as well as reconstructions from a standard CT perfusion acquisition of an acute stroke patient. Experiments show that accurate estimates of the underlying time-intensity curves can be recovered on a pixel-by-pixel base.