TY - JOUR T1 - A comprehensive PET-CT scanner characterization performance assessment paradigm and database. JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1398 LP - 1398 VL - 62 IS - supplement 1 AU - John Sunderland AU - Ashok Tiwari Y1 - 2021/05/01 UR - http://jnm.snmjournals.org/content/62/supplement_1/1398.abstract N2 - 1398Introduction: New SiPM PET scanner hardware is widening the range of imaging performance between old and new generation scanners. New advanced reconstructions result in generally higher contrast images, and higher reconstructed resolution is achievable but with a noise penalty that impacts both quantitative accuracy and reproducibility. The choice as to how to optimally acquire and reconstruct PET images for a given PET/CT system is complex and unclear. We believe there would be value in creating and sharing a comprehensive library of well-curated, standardized, PET/CT scanner model-specific phantom images data sets with associated analyses. This has potential to benefit both the clinical and clinical trial enterprises as well as providing well-characterized image datasets to entities developing imaging segmentation and analysis software. Described below is a prototype phantom-based approach to standardly and fully characterize a particular PET/CT scanner model’s 1) reconstructed resolution, 2) noise properties, 3) recovery coefficient (RC) performance, and 4) image quality across a broad range of reconstructions. The method employs three currently available commercial phantoms, a prototype phantom, and publicly available analysis software. Results are preliminarily presented on four PET/CT platforms, but is intended to ultimately include all PET/CT systems. Methods: In this prototype performance measurement paradigm, the GE Discovery MI, Siemens Vision 600, Siemens mCT, and Siemens TruePoint are characterized. Four phantoms were used: the 20 cm uniform phantom, standard NEMA IQ phantom, a modified (12 sphere) NEMA IQ Phantom, and CTN Oncology Phantom. Acquisitions and analyses were performed as shown in Table 1. The number and types of reconstructions performed cover the full clinical range of reconstruction parameter sets, including advanced reconstructions. Analyses were performed with the publicly available online automated SNMMI Phantom Analysis Tool (PAT) http://snmmi.org/PAT, except for the modified NEMA 12 sphere phantom. Axial and radial resolution were measured on the DMI (42 different reconstruction parameter sets), Vision (48), mCT (48) and TruePoint (18) using the uniform phantom1. RC curves were generated using high statistics data (30 min/bed) using recommended CTN and EARL fills for the CTN and NEMA phantoms, respectively. Clinical statistic RC curves were generated for all systems and reconstructions using 17 replicate 4-minute/bed acquisitions (decay corrected) of the 12 sphere NEMA phantom to assess reproducibility of clinically relevant SUV measurements. NEMA noise and image roughness measurements were made using the CTN and modified NEMA phantom data collected with clinically relevant counts. Lesion detection and image quality measures were manually performed using a Likert scale on the 2 image quality features available in the CTN Oncology phantom. Results: Data was successfully collected, reconstructed, and analyzed for the 4 PET/CT systems, resulting in 2500 total phantom datasets for the 4 systems. Axial and Radial reconstructed resolution were successfully measured using PAT for all scanners and reconstructions (Fig 1). The 17 replicate 4-minute scans enabled generation of RC curves (SUVmax, SUVpeak, SUVmean) with associated standard deviation of the measurements to assess reproducibility and bias as a function of reconstruction. These RC curves acquired with clinically relevant counts can be compared to the RC curves generated with the high statistic, lower noise 30 min/bed acquisitions from the CTN and NEMA phantoms. Conclusion: A prototype performance measurement paradigm for PET/CT scanners has been tested and appears to generate meaningful quantitative information that could inform both clinical practice and clinical trials. Image sets, metadata, and analysis have been uploaded to an SNMMI administered platform for sharing.1Lodge et al. JNM 2018, 59(11) 1768-1775 View this table:Phantom Acquisitions and Measures ER -