Abstract
2339
Objectives Biograph mMR scanner is the state-of-the-art simultaneous PETMR scanner that has great potential in clinical applications. However, the Point Spread Function (PSF) reconstruction is not yet available for this scanner because of the compatibility of MRI field. In this project, we build an accurate PSF model using analytical calculation and Monte Carlo simulation. In addition, we develop a Maximum a Posteriori (MAP) reconstruction to reduce the ring artifact caused by the PSF.
Methods We first calculate the geometrical system matrix incorporating the block structure of scanner and the gaps between detector blocks. We then model the sinogram blurring (or PSF) by Monte Carlo simulation considering the physical processes, such as photon pair nonlinearity, inter-crystal scatter and penetration, as well as the block structure and the gaps between blocks [1]. A spatially variant quadratic penalty function is applied with Poisson log-likelihood function to achieve count-independent resolution and reduce the ring artifact. Normalization, attenuation, scatters and randoms corrections are included in the reconstruction. A Jaszczak phantom filled with FDG is scanned in mMR scanner for 120 minutes and then the list mode data is sorted into 30 sinograms with103 M counts each. The resulting sinograms are reconstructed by both OSEM3D without PSF and MAP with PSF.
Results The images reconstructed with PSF demonstrate higher resolution than the ones reconstructed without PSF; the profiles cross the hot and cold regions show better edge-preserving features with PSF reconstruction. MAP reconstruction with spatially variant smoothing demonstrates better noise performance at same resolution level compared with OSEM3D without PSF.
Conclusions PSF modeling is essential to the resolution recovery of mMR scanner. Our preliminary results indicate that MAP reconstruction with PSF can achieve both higher resolution and better bias-variance trade-off