Abstract
1770
Objectives: The research goal is to develop a non-iterative method for sinogram estimation with resolution recovery that is reproducible for use in emission tomography. Key criteria include incorporation of imaging system models to optimize resolution, good image signal-to-noise-ratios, and a computationally simple single pass algorithm for rapid throughput. Methods: The proposed method estimates projection images using a single iteration 2D local regression to describe shapes within the projection image. The mathematical relationship between the measured sinogram and the true object sinogram using imaging system spatial resolution models is then approximated yielding an expression that describes the measured sinogram as the linear sum of the true object sinogram and the second and fourth order derivatives of the measured sinogram weighted by the moments of the imaging system response function. This enables estimation of the true object sinogram which is reconstructed using filtered backprojection with a ramp filter. Simulation studies, phantom measurements, and small animal PET imaging data sets were used to test the performance of this algorithm. The simulation studies enabled comparisons of achievable spatial resolution and image signal-to-noise estimates with conventional reconstruction algorithms. Results: Imaging data from the Data Spectrum Micro-Deluxe phantom with cold spot inserts demonstrated that all cold spots could be clearly resolved, including 1.2 mm rods, using the proposed algorithm. Simulation studies demonstrate both enhanced spatial resolution and image SNR when compared to conventional reconstruction algorithms (see table). Imaging studies in mice demonstrate significant qualitative improvements in spatial resolution and image SNR consistent with the simulation and phantom study observations. Conclusions: The proposed method has been shown to improve transaxial image resolution while simultaneously controlling image SNR. The algorithm is simple to apply, is non-iterative, and overcomes convergence problems associated with iterative reconstruction algorithms.
Research Support (if any): Indiana Genomics Initiative/Lilly Endowment, Inc.
- Society of Nuclear Medicine, Inc.