Evaluation of Parallel Level Sets and Bowsher’s Method as Segmentation-Free Anatomical Priors for Time-of-Flight PET Reconstruction
G. Schramm, M. Holler, A. Rezaei, K. Vunckx, F. Knoll, K. Bredies, F. Boada, J. Nuyts
IEEE Trans Med Imaging
Image reconstruction in Positron Emission Tomography (PET) is challenged by high Poisson noise in the acquired data and limited spatial resolution due to finite detector size and photon acollinearity.
The limited spatial resolution in the reconstructed images makes the reconstruction problem
ill-posed, and results in Gibbs artifacts (ringing) in unconstrained reconstructions with resolution recovery. Assuming that the PET tracer distribution follows anatomical boundaries, penalized-likelihood reconstructions using anatomical information derived from high-resolution magnetic resonance (MR) images can be used to suppress noise and Gibbs artifacts while limiting the loss of resolution caused by the regularization.
In this work, we evaluate the concept of Parallel Level Sets (PLS) and Bowsher's method to include anatomical prior information into penalized iterative PET reconstruction. In addition, we present an efficient algorithm for the solution of the PLS-regularized PET reconstruction problem that is suitable for clinical time-of-flight PET reconstructions.
The analysis of data from dedicated simulations and a clinical TOF PET/MR data set shows that penalized-likelihood reconstructions using PLS and Bowsher's method are superior to post-smoothed Maximum Likelihood Expectation Maximization with ordered subsets (OSEM) in terms of bias-noise characteristics in different regions of interest. At very high regularization weights, all investigated anatomical priors suffer from the transfer of non-shared gradients.