Phantom-based Noise Assessment in 3D-PET Image Reconstruction

Abstract

In addition to traditional iterative PET reconstruction methods like OSEM, some new algorithms take into account resolution recovery and regularization, has gradually gained attention. Comparing and evaluating these reconstruction methods has become an important issue. The phantom in this research contains a large anthropomorphic chamber and spheres with diameter in 1 cm and 2 cm injected with fluorine-18 radionuclide. To perform the reconstruction, different algorithms are applied from the scanner manufacturer’s software: (1) Time-of-flight (TOF) (2) Non-time-of-flight (NTF) (3) SharpIR (4) Q.Clear. Data is obtained from three PET scanners, and different number of iterations is considered for each algorithm. Three different definitions of noise are measured to assess the quality of the PET images: (1) Image roughness (IR), (2) Background variability (BV) and (3) Ensemble noise (EN). As observed from the noise-contrast plots, images reconstructed with time-of-flight, resolution recovery (SharpIR) and regularization algorithm (Q.Clear) usually lead to better contrast with lower noise. However, for OSEM images without resolution recovery, more iterations (greater than 5) do not necessarily result in higher quality. On the contrary, the image quality will decline because the contrast is almost unchanged, while the noise increases rapidly