Investigation of super-sampling techniques with blob-based super-resolution reconstructions for PET imaging

Spatial resolution in positron emission tomography (PET) is still a limiting factor in many imaging applications. Mechanical movement (e.g., controlled scanner or bed motion) has been introduced to PET systems to increase sampling and improve spatial resolution. The objective of this work is to investigate super-sampling techniques with an advanced blob-based super-resolution reconstruction algorithm. We conducted experiments with a small-animal PET scanner (A-PET) using a micro hot-rod phantom. The phantom was mounted on three motorized linear stages with freedom in the three perpendicular axes. We generated motion synchronized digital pulses and recorded them as control-events in list-mode data to separate different data sets from different positions. We investigated different sampling strategies: 1) stationary acquisition, 2) stepping the phantom in 3-D within a motion cube, 3) translating the phantom along the scanner axial direction and 4) translating the phantom along a controlled but deliberately misaligned direction. To fully take advantage of the super-sampling techniques, we developed and implemented a dedicated blob-based super-resolution reconstruction approach based on the MLEM and OSEM algorithms. The object offset was incorporated in the forward and backward projections at each iteration. The high-resolution images were reconstructed from the multiple list-mode data sets using the blob-based super-resolution method. The reconstructed images of the hot-rod phantom were compared for the four acquisitions with different sampling strategies. We demonstrated that both grid and linear super-sampling with deliberately misaligned direction can improve the overall image-quality with the same acquisition time. In conclusion, super-sampling is an effective and economic way to improve sampling and spatial resolution for PET imaging, which has an important implication in preclinical and clinical PET imaging applications.