Abstract
96
Objectives System response blurring degrades image quality. The objective of this study was to develop a novel method to model shift-variant point spread function (PSF) of PET systems and incorporate into listmode reconstruction for resolution recovery.
Methods High count listmode data were acquired on a clinical ToF PET system using a point source of 1 mm in diameter placed in air at multiple positions cross the field of view. The data were rebinned into uniformly spaced (0.5 mm) sinograms, from which FWHM values were measured in both axial direction and the transverse plane for each angle. The FWHM values were investigated and then fitted into polynomial functions of radial and angular variables. Incorporated in listmode reconstruction on GPU, the PSF for each LOR is modeled as a Gaussian function based on the FWHM values obtained from the polynomial functions to calculate contributions from neighbor voxels. This method was evaluated using a Jaszczak phantom and a patient brain study with (6 iterations) and without (3 iterations) polynomial modeling of PSF.
Results The FWHM in transverse plane increased with the radial distance, ranging from 3.9 mm at center to ~8.5 mm at 25 cm off-center, and also changed with the angle. FWHM was approximated with a 2nd order polynomial function of radius, whose coefficients are linear functions of the angle. Surface plots showed that the modeled FWHM agreed in trend and range with those from the measured data. The averaged FWHM in the axial direction with no z-tilt was relatively constant, ranging from 4.2 to 5.3 mm. The images reconstructed with PSF modeling provided more details with higher resolution and signal-to-noise ratio as compared to those without PSF modeling.
Conclusions A method based on polynomials has been developed to allow fast modeling of shift-variant PSF of ToF PET systems and incorporated into list-mode reconstructing to provide improved image quality.