Data rebinning and reconstruction in 3D PET/CT oncological studies: a Monte Carlo evaluation

Abstract

An accurate identification and interpretation of neoplastic lesions by PET is related to PET image quality, depending on several factors including data processing for image formation. Aim of this work was to assess the influence of data rebinning and reconstruction on lesion detectability and quantification for a high-resolution 3D PET/CT system, in order to optimize 3D PET/CT oncological protocols. Oncological 18F-FDG PET studies were Monte Carlo (MC) simulated, varying lesion size (LS), lesion-to-background ratio (LBR), statistics, and including or not attenuation and scatter effects. Single slice rebinned (SSR), Fourier rebinned (FORE) and fully 3D sinograms were considered. 2D/3D ordered subset expectation maximization (OSEM) reconstruction was applied. Human observers evaluated images in terms of clinical parameters characteristic of lesion detectability and quantification: lesion number, LS, and LBR. By comparison with the known activity map (input for MC simulations), identified lesions were classified as true and false positive (TP, FP), and % errors on LS and on LBR were calculated The results show that 2D rebinning allow more lesions to be identified than fully 3D, but SSR induces more FP than FORE. Quantitatively, smaller errors on LBR and on LS were found using FORE rebinning and fully 3D, respectively. Our findings suggest that FORE+2D OSEM is more suitable for lesion detectability and quantification, fully 3D OSEM is better for lesion spatial characterization