Penalized maximum-likelihood image reconstruction for a two-panel PET scanner with image-based resolution modeling.

455 Objectives: Current whole-body positron emission tomography (PET) scanners have limited spatial resolution for imaging small structures in the human body, such as small lymph nodes (smaller than 1cm). We are developing a two-panel high-spatial-resolution PET scanner to improve imaging head and neck cancers. The two-panel scanner will be used after whole-body PET scan and has the advantage of patients’ comfort since the panels will be put on both sides of patients’ head, preventing claustrophobia. The challenge is that insufficient sampling in the two-panel scanner leads to artifacts in reconstructed images. To mitigate the limited-angle artifacts, we investigated a penalized maximum-likelihood (PML) image reconstruction method in which a prior image without limited-angle artifacts (low-spatial-resolution whole-body PET image) was used for regularization. A similar method has been used in time-resolved computed tomography (CT) image reconstruction when sufficient angular range for data acquisition is not achieved. Methods: The prior image was from a whole-body scanner. An image-based resolution model was incorporated into the regularization, blurring the image to be reconstructed for the two-panel scanner (the target image). The dissimilarity between the blurred target image and the prior image was penalized. We implemented a spatial-variant and asymmetrical Gaussian model for the blurring. Image reconstructions were performed for the simulation data of point source at different radial distances in the whole-body scanner field of view (FOV). Five parameters were used for fitting the reconstructed point source, including the internal radial sigma, the external radial sigma, the tangential sigma, the axial sigma, and the shift in the radial direction. Currently, only the dependency on the radial distance from the FOV center is taken into account. A polynomial quadratic function was then used for fitting the values for each parameter versus the radial distances. In this way, the parameters for each point inside the FOV can be calculated. Monte Carlo simulations were used for generating coincidence events. The two-panel scanner was based on 2 mm x 2 mm x 20 mm LYSO crystals. GE Discovery MI 4-ring PET scanner was used as the whole-body model. An 11-cm-diameter and 12.6-cm-long cylindrical phantom with uniform water attenuation was used in the simulations. Hot spheres were put in the phantom at the center slice, with the concentration ratio being 8:1. Results: The reconstructed images using the maximum-likelihood (ML) method showed strong limited-angle artifacts. The background region and hot spheres were elongated in the direction orthogonal to the panels. The elongation was mitigated in the image reconstructed using the PML method. The contrast recovery coefficient (CRC) or the contrast-to-noise ratio (CNR) of the small spheres (3 mm and 4 mm in diameter) in the image reconstructed using the ML and the PML method were similar. The 6-mm-diameter and 8-mm-diameter spheres reconstructed using the PML method had higher CRC or CNR compared to the spheres reconstructed using the ML method. Conclusions: The PML image reconstruction method investigated in this work can be used for mitigating the limited-angle artifacts in the image reconstruction for the two-panel high-spatial-resolution PET scanner. The ability of the scanner to image small features is retained.