Abstract
Evaluation of PET image quality is central to annual physics surveys, quality assurance, and laboratory accreditation. A common method is to image the American College of Radiology (ACR) PET phantom, which contains hot and cold structures of various sizes in a warm background. Performance evaluation involves qualitative assessment of hot and cold structure visibility and overall image quality. Some criteria are quantitative and rely on manually-drawn regions of interest (ROIs) to measure standardized uptake value (SUV). Fully automated scoring of ACR PET phantom images would improve efficiency, avoid observer-related dependencies, and may provide more robust evaluation of image quality. Methods: Software was developed to co-register PET images to a phantom template and to compute ROI measurements of hot vial activity (SUV-max) and background activity (SUV-mean) automatically. In addition, three-dimensional volumes of interest (VOIs) were generated to measure hot vial activity (“SUV-vial”), background activity, and cold rods contrast. Consistency of the ROI-based and VOI-based methods was evaluated using phantom data from a total of 17 annual physics surveys of three PET/CT scanners with the same PET detector designs. Results: The automated software processed all PET phantom datasets successfully. SUV consistency for hot vials was improved through use of cylindrical VOIs and through normalization with respect to assayed activities and dilution volumes used in phantom preparation. Average vial SUV standard deviation improved from 8.0% for standard SUV-max to 3.2% for normalized SUV-vial. Similarly, standard deviation for SUV ratio of 16mm to 25mm vials improved from 5.0% for SUV-max to 3.2% for SUV-vial. Background mean SUV had similar consistency between the ROI and VOI methods. Cold rods contrast was highly consistent, offering a potential alternative to qualitative visual assessment of low-contrast performance. Conclusion: Automated quantitative scoring of the ACR PET phantom is feasible and offers advantages of more efficient, consistent, and thorough performance characterization. Acceptance ranges for SUV values and ratios likely can be tightened if normalized VOI measurements are used. Further testing with phantom data from a variety of PET scanners is necessary to establish suitable quantitative thresholds for acceptable performance.