Image reconstruction in positron emission tomography (PET) is reaching a maturing stage where the main algorithmic classes (e.g. analytical vs. iterative) are largely delineated, modeling of the main physics processes (e.g. attenuation, scatter, randoms, and more recently the point spread function) is well established, and more complex issues such as respiratory motion compensation and direct kinetic parameter estimation are aggressively being pursued. Once of the greatest challenges when investigating new PET algorithms and technologies is the need for efficacious assessment of image quality. The most widely accepted method for objective assessment of PET image quality relies upon measuring performance for specific clinical tasks (or surrogates thereof), which requires significant expertise and challenging development of test datasets and methodologies. Building on our previous experience in these areas, we propose to create a database of PET imaging data designed and tuned for evaluating observer performance in detecting focal warm lesions, modeling the clinical task of cancer detection and staging. This database will be comprised of existing datasets acquired under a prior R01 project and via research relationships with two major PET tomograph manufacturers. The database will include scans from four PET tomographs from two vendors operated in 2D, fully-3D, and 3D + time-of-flight (TOF) acquisition modes. The database is intended to provide a collaborative shared resource enabling the broad PET research community to easily assess lesion-detection performance improvements for their own developmental algorithms and technologies. As such, the database will be designed with portable data formats, plus include collaborative access to research reconstruction software and observer study tools. These software tools have been developed and used in the investigator's laboratory for the past decade. Since the datasets and observer studies are quite complex, data will not be shared anonymously;rather, collaborative access is targeted where the grantee will be collaboratively involved in study design and execution. The database, once prepared and organized, will enable rapid and repeat localization receiver operating characteristics (LROC) studies with both model and human observers in timeframes of several weeks (as opposed to months to years currently required for groups new to such studies to get them up and running). This will be exemplified in the second phase of the proposed research, where the database will be used to resolve and optimize several unanswered questions regarding parameter selection for current iterative PET reconstruction algorithms. Successful completion of this project will provide a new research resource enabling widespread objective LROC studies for PET lesion-detection by a number of groups, as well as optimize several reconstruction parameters for the most common clinical PET application of cancer detection and staging.
Positron emission tomography (PET) has experienced tremendous growth in past years, and new technologies are emerging that will push the modality even further. One of the great challenges, however, is evaluating these technological advances in terms of how the changes in image quality affect clinical tasks. This project will create a new collaborative resource for performing task-based assessment of PET image quality, and then use this resource to investigate optimal approaches for reconstructing PET images.
|Morey, A Michael; Kadrmas, Dan J (2013) Effect of varying number of OSEM subsets on PET lesion detectability. J Nucl Med Technol 41:268-73|