engleski

Iterative Algorithms for Gaussian Mixture Model Estimation in 2D PET Imaging

In positron emission tomography (PET) the original measurement consists of pairs of gamma rays emitted from a radioactive substance forming a line, captured within a given plane (2D) or volume (3D). Traditional image reconstruction methods estimate intensity values in pixels or voxels on some predefined grid. In this paper, we investigate reconstruction of PET images directly from the lines of response, using a probabilistic Gaussian mixture model (GMM) for the underlying originals. Parameters are estimated by solving an overdetermined system of equations obtained directly from measurements. Experiments are performed on artificial data, using an iterative process resembling the expectation-maximization algorithm. Reconstruction is performed using several variations of the algorithm, which are compared by measuring the structural similarity index of the graphic representation of underlying distributions. The proposed segmentation method relies on the statistical properties of GMMs and appears to be robust, giving new insight and a new approach to traditional problems on real data.

positron emission tomography, Gaussian mixture models, expectation-maximization (EM) algorithm, image segmentation

nije evidentirano