The Utah PET lesion detection database

Task-based assessment of image quality is a challenging but necessary step in evaluating advancements in PET instrumentation, algorithms, and processing. We have been developing methods of evaluating observer performance for detecting and localizing focal warm lesions using experimentally-acquired whole-body phantom data designed to mimic oncologic FDG PET imaging. This work describes a new resource of experimental phantom data that is being developed to facilitate lesion detection studies for the evaluation of PET reconstruction algorithms and related developments. A new large custom-designed thorax phantom has been constructed to complement our existing medium thorax phantom, providing two whole-body setups for lesion detection experiments. The new phantom is ~50% larger and has a removable spine/rib-cage attenuating structure that is held in place with low water resistance open cell foam. Several series of experiments have been acquired, with more ongoing, including both 2D and fully-3D acquisitions on tomographs from multiple vendors, various phantom configurations and lesion distributions. All raw data, normalizations, and calibrations are collected and offloaded to the database, enabling subsequent retrospective offline reconstruction with research software for various applications. The offloaded data are further processed to identify the true lesion locations in preparation for use with both human observers and numerical studies using the channelized non-prewhitened observer. These data have been used to study the impact of improved statistical algorithms, point spread function modeling, and time-of-flight measurements upon focal lesion detection performance, and studies on the effects of accelerated block-iterative algorithms and advanced regularization techniques are currently ongoing. Interested researchers are encouraged to contact the author regarding potential collaboration and application of database experiments to their projects.