Evaluation of a new regularization prior for 3D PET reconstruction including PSF modelling

A general limitation in PET is represented by the poor spatial resolution of the system. To compensate for this limitation by using iterative reconstruction algorithms it is possible to account for the response of the PET system (Point Spread Function, PSF) in the reconstruction scheme to improve PET image quality and quantitative accuracy. Unfortunately, a common behaviour of iterative reconstruction techniques is the increase of noise as the iterations proceed due to the ill-posed nature of the reconstruction process. On the other hand a high number of iterations is usually needed to recover a significant percentage of the signal and to reach the convergence, especially when including resolution modelling. To solve this dilemma, regularization strategies could be employed to control the noise amplification as the iterations proceed. In this work a new prior for variational Maximum a Posteriori regularization is proposed to be used in a 3D ML-OSEM reconstruction algorithm which also accounts for the PSF of the PET system. The new regularization prior is characterised by a strong smoothing component for regions in the image with a magnitude of the gradient below a given threshold (set to discriminate between background and signal), while preserving edges above the threshold. The new algorithm has been validated on phantom and clinical data. The results showed that the use of the proposed regularization prior allows a better control of the noise, while maintaining high enough signal recovery thanks to the PSF modelling. To obtain the best results using the proposed prior, i.e. the best compromise between noise control and loss of recovered signal, an optimization of the regularization parameters is therefore required.