Influence of Reconstruction Techniques on PET/CT Image Quality and Quantitative Accuracy: a Phantom Study

Purpose To investigate the influence of different reconstruction techniques on the quantitative accuracy and image quality of PET/CT. Methods The NEMA NU2-2018 image quality phantom was scanned on a GE Discovery Elite PET/CT scanner and the spatial resolution was measured based on NEMA NU2 standard. The same raw data were reconstructed using five different algorithms: ordered subset expectation maximization (OSEM), OSEM with point spread function (PSF) modeling, OSEM with time-of-flight (TOF), OSEM with TOF and PSF, and filtered back-projection (FBP). The recovery coefficient (RC), contrast recovery coefficient (CRC), and contrast-to-noise ratio (CNR) were calculated for the six hot spheres, the percent background variability (PBV) and coefficient of variation (COV) were calculated for the background, and the residual error (RE) was calculated for lung insert in different image slices. Results The incorporation of PSF modeling showed the smallest transverse FWHM and FWTM at both 1 and 10 cm radical offsets. The combination of PSF modeling and TOF improved RC mean and CRC for all spheres and resulted in the highest ratings for the detectability of 10 mm spheres in human observer assessment. PSF modeling played a role in reducing the COV within the background region of interest and increasing the CNR of the spheres, and decreased background noise ratings in human observer assessment. Besides, TOF significantly reduced the RE in lung insert. Neither PSF modeling nor TOF had a significant effect on PBV. Compared to FBP, the OSEM algorithm showed significant advantages in PBV, COV, CNR and RE and human observer ratings of image quality, but worse results for RC max , RC mean and CRC. Conclusions The integration of TOF and PSF modeling into the OSEM algorithm achieves improvements in both quantitative accuracy and image quality, providing distinct advantages. PSF modeling improves the spatial resolution and decreases the visual appearance of background noise. The OSEM algorithm shows significantly better image quality than the FBP algorithm but no distinct advantages concerning quantitative accuracy.