Image reconstruction and signal detectability in dual-head small animal PET

Dual-head flat-panel positron emission tomography (PET) is the dedicated system for small-animal imaging because of its high spatial resolution and detection sensitivity. Unlike the conventional ring-based PET system, the dual-head system is versatile and can be reconfigured readily to accommodate different sample sizes and/or to achieve an optimal system performance. However, the unique imaging geometry also leads to severe depth-of-interaction (DOl) blurring. Monte Carlo simulation has been employed to account for DOl effects in the system response matrix with high accuracy, but the long simulation times make routine search for an optimal configuration impractical. To facilitate the system optimization while considering the parallax error, we employ an efficient numerical method to model the DOl effect in the system matrix for image reconstruction and accelerate the computation with the graphics processing units (GPUs). Among the system response matrices corresponding to different geometric configurations, a favorable one can be selected by exploiting the statistical detection theory. Computer simulation studies were carried out to validate and quantitatively evaluate the proposed method.