Dynamic PET reconstruction algorithms using empirical mode decomposition regularisation

Dynamic PET enables quantitative analysis of in-vivo metabolic activity. Commonly, each image in a temporal sequence is reconstructed independently using standard methods developed for static PET.We present reconstruction methods which use Empirical Mode Decomposition (EMD) based regularization. The methods extend conventional static OSEM reconstruction to ensure consistency between temporal frames. Various data-driven denoising methods are evaluated. The EMD denoising scheme has advantages over conventional Gaussian smoothing and wavelet denoising. We perform 1D and 2D+t dPET simulations to compare the new algorithms with conventional FBP and OSEM. The methods can accommodate a wide range of activity curves and pharmacokinetic models. The new methods result in lower minimum MSE and larger maximum SNR after fewer iterations than conventional algorithms.