Spectral analysis of PET projection data

Spectral analysis is a general modelling approach that allows the determination of unit impulse response functions (IRF) from tracer kinetic data. The technique can be applied at a pixel level enabling calculation of parametric images independent of an assumed compartmental structure. We have investigated the validity of applying spectral analysis to projection data (SpecSino) motivated by the advantages that: (i) the number of reconstructions is reduced by an order of magnitude and (ii) iterative reconstruction of parametric images becomes practical (as a result of (i)) which potentially offers improved signal-to-noise ratio (SNR). A software phantom based on the Zubal phantom and typical [ ¹¹C]thymidine kinetics was used to evaluate SpecSino and assess the trade-off between bias and variance using filtered back projection (FBP) and OS-EM reconstruction. We found that the two approaches are not equivalent due to properties of the non-negative least squares fitting algorithm. However, the differences are small (<5%) and mainly affect parameters related to early and late time points on the IRF (K₁ and, to a lesser extent, VD). The optimal number of EM iterations (or subsets) was found to be 15-30 with up to a 2-fold improvement in SNR. The technique has been applied to a range of PET ligand studies (both in 2D and 3D) and yields accurate parametric images with high SNR without imposing a compartmental model