Statistical image reconstruction methods in PET with compensation for missing data

We present the results of combining volume imaging with the PENN-PET scanner with statistical image reconstruction methods such as the penalized weighted least squares (PWLS) method. The goal of this particular combination is to improve both classification and estimation tasks in PET imaging protocols where image quality is dominated by spatially-variant system responses and/or measurement statistics. The PENN-PET scanner has strongly spatially-varying system behavior due to its volume imaging design and the presence of detector gaps. Statistical methods are easily adapted to this scanner geometry, including the detector gaps, and have also been shown to have improved bias/variance trade-offs compared to the standard filtered-backprojection (FBP) reconstruction method. The PWLS method requires fewer iterations and may be more tolerant of errors in the system model than other statistical methods. We present results demonstrating the improvement in image quality for PWLS image reconstructions of data from the PENN-PET scanner