Evaluation of 3D image reconstruction methods for a dual-head small-animal PET scanner

A high-sensitivity small-animal PET (muPET) scanner can be obtained by employing two large-area detector panels in a compact, stationary configuration. However, the compact geometry also results in substantial parallax errors, therefore producing images that show resolution significantly worse than the scanner´s intrinsic resolution and offsetting the advantages gained in the sensitivity. In theory, parallax errors can be corrected for in image reconstruction. In practice, such correction is limited by the difficulty in obtaining an accurate system response matrix (SRM) for the scanner. With the dual-head configuration, we are able to exploit the symmetry properties available to drastically reduce the complexity in computing the SRM. As a result, we can employ Monte-Carlo (MC) simulations to generate the SRM. We employ the calculated SRM to examine the parallax errors in, and evaluate the image spatial resolution of, our prototypical muPET scanner. As expected, when not including the SRM in reconstruction, the spatial resolution deteriorates substantially and is non-isotropic: the resolution is -1.8 mm in directions parallel to the detectors and becomes -3.4 mm in the direction perpendicular. In comparison, when incorporating the SRM in reconstruction, such resolution degradations can be eliminated, reaching an isotropic spatial resolution of -(1.2 mm)³. We also investigate the noise properties of the reconstructions and show preliminary results obtained from real data.