Multi-resolution image reconstruction for a high-resolution small animal PET device

Imaging systems that generate data having widely different spatial uncertainties arise in several contexts. While it is desirable to use all information generated by these devices, conventional reconstruction methods such as filtered backprojection cannot readily account for the differing intrinsic quality of the measurements. Embedding the reconstruction problem in a statistical framework can eliminate this issue. A maximum likelihood reconstruction method is described for a proposed small animal PET system that records high-, medium-, and low-resolution events during a single data acquisition depending on the annihilation photon interactions. Unlike filtered backprojection, the statistically-motivated method naturally accounts for the different measurement uncertainties. A computationally efficient implementation based on a common tomographic projection operator with individual spatially invariant projection-space smoothing operators is described. Additionally, a method is introduced whereby the value of each dataset in terms of reducing the variance in reconstructed images at a given resolution can be quantified.