Improved resolution via 3D iterative reconstruction for PET volume imaging

The authors have implemented iterative filtered backprojection (IFBP) and maximum likelihood by expectation maximization (ML-EM) algorithms in 3D space and applied them to phantom and real PET data. Transaxial resolution improves ≈50% and axial resolution improves ≈15% for IFBP at 15 iterations without a sieve compared to FBP. With a sieve, the improvements are reduced to ≈6%. 3D ML-EM reconstruction shows similar resolution improvement with a much slower convergence rate compared to IFBP. The improvements in resolution from both IFBP and ML-EM are apparent in 3D FDG brain data