The influence of noise in full Monte Carlo ML-EM and dual matrix reconstructions in positron emission tomography

Studying the properties of the updating coefficients in the OSEM algorithm for iterative image reconstruction in PET

We have studied the properties of the pixel updating coefficients in the 2D ordered subsets expectation maximization (OSEM) algorithm for iterative image reconstruction in positron emission tomography, in order to address the problem of image quality degradation-a known property of the technique after a number of iterations. The behavior of the updating coefficients has been extensively analyzed on synthetic coincidence data, using the necessary software tools. The experiments showed that the statistical properties of these coefficients can be correlated with the quality of the reconstructed images as a function of the activity distribution in the source and the number of subsets used. Considering the fact that these properties can be quantified during the reconstruction process of data from real scans where the activity distribution in the source is unknown the results of this study might be useful for the development of a stopping criterion for the OSEM algorithm.

A parallelizable compression scheme for Monte Carlo scatter system matrices in PET image reconstruction

Scatter correction techniques in iterative positron emission tomography (PET) reconstruction increasingly utilize Monte Carlo (MC) simulations which are very well suited to model scatter in the inhomogeneous patient. Due to memory constraints the results of these simulations are not stored in the system matrix, but added or subtracted as a constant term or recalculated in the projector at each iteration. This implies that scatter is not considered in the back-projector. The presented scheme provides a method to store the simulated Monte Carlo scatter in a compressed scatter system matrix. The compression is based on parametrization and B-spline approximation and allows the formation of the scatter matrix based on low statistics simulations. The compression as well as the retrieval of the matrix elements are parallelizable. It is shown that the proposed compression scheme provides sufficient compression so that the storage in memory of a scatter system matrix for a 3D scanner is feasible. Scatter matrices of two different 2D scanner geometries were compressed and used for reconstruction as a proof of concept. Compression ratios of 0.1% could be achieved and scatter induced artifacts in the images were successfully reduced by using the compressed matrices in the reconstruction algorithm.