Spatially Penalized Methods for Linear Parametric Imaging in Dynamic PET

Dynamic emission tomography can provide the estimation of physiological and biochemical parameters through the use of tracer kinetic modeling techniques. Compared to the conventional region-of-interest method for kinetic data analysis, parametric imaging is becoming preferable since it can provide the spatial distribution of physiologically important parameters. However, parametric images obtained by the conventional methods are usually noisy because the time activity curve of each pixel has high noise. Here we use Markov random field as an image prior to improve the quality of parametric images. Compared to the ridge regression method where spatial regularization was introduced through a mean image, Markov random field prior explicitly models the spatial correlation between neighboring pixels. We have derived monotonically convergent iterative algorithms for estimating parametric images. The method is evaluated using computer simulation and also applied to real dynamic PET data.