TY - JOUR T1 - Assessment of PET/CT reconstruction harmonization through Gaussian post- filtration. JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1917 LP - 1917 VL - 57 IS - supplement 2 AU - John Sunderland AU - Paul Kinahan AU - Joel Karp AU - Margaret Daube-Witherspoon AU - Darrin Byrd AU - Joseph Panetta AU - Josh Scheuermann AU - Levent Sensoy Y1 - 2016/05/01 UR - http://jnm.snmjournals.org/content/57/supplement_2/1917.abstract N2 - 1917Objectives Differing quantitative performance characteristics of PET/CT scanners with varied reconstruction settings confound the ability to meaningfully combine and compare data in multi-scanner settings, including both clinical trials and clinical practice. Previous work explored identifying scanner model-specific reconstruction parameter sets that resulted in quantitatively similar performances as defined by contrast recovery coefficient (CRC) curve harmonization. In this work we explore a simpler harmonization approach whereby post-reconstruction Gaussian filtration is used as the mechanism either to align CRC curves to a standardized or reference curve (e.g. EANM EARL) or to align the CRC curves of one scanner model to another.Methods A modified NEMA IQ phantom was imaged on 8 different PET/CT models (total of 23 PET/CT scanners) for 30 minutes at 9.7:1 contrast using 12 sphere sizes 8.5-44mm (6 at a time). Images were reconstructed using 3-5 different clinically relevant reconstruction parameter sets per scanner spanning a range from higher to lower resolution. Scanner model and reconstruction specific CRC curves were generated for SUVmax, SUVpeak and SUVmean. SUVmax data is presented here. Two harmonization approaches and analyses were performed. First, Gaussian filters of varying FWHM (0.3mm steps) were convolved with model and reconstruction specific image sets and CRC curves generated. The Gaussian FWHM filter that resulted in a CRC curve best aligned with the EANM EARL CRC curve (a well-established standard for quantitative validation of PET/CT scanners for clinical trials) was defined as the optimized harmonization filter (OHF) FWHM. OHF FWHM was determined for the each of the reconstructions for the 8 PET/CT models. Second, using similar methodologies, a CRC-based Gaussian OHF FWHM was identified for all scanner model/reconstruction pairs allowing for direct filter-based harmonization between any two scanner models, avoiding constraint to a pre-determined reference curve. The higher resolution model/reconstruction combination had the Gaussian filter applied to match the CRC of the lower resolution set.Results Clinically-used reconstructions have demonstrated widely divergent quantitative performance. Gaussian filtration proved, in general, to be an effective quantitative harmonization strategy. OHF FWHM values ranging from 2.0-6.2mm were identified to best smooth data to the mean EANM EARL CRC curve for 3-5 different reconstructions from each of the 8 PET/CT models. In 11 of the 34 cases the original reconstruction parameter set resulted in CRC curves below the reference curve; thus additional Gaussian filtration would shift the curve further from the reference curve. In general this occurred when original reconstructions used Gaussian filters >6mm. Predictably, some model scanner’s CRC curves aligned more closely with the reference curve than others. For pair-wise scanner matching, OHF FWHM values (ranging from 1.8mm-7.2mm) were identified for 232 different scanner model/reconstruction pairs associated with the 8 model scanners analyzed to date. OHF FWHM was only calculated for the higher resolution model/reconstruction set. Pair-wise reconstructions predictably resulted in better curve matching (as defined by sum of square differences) than matching to a standard curve.Conclusions PET/CT reconstruction harmonization through Gaussian post-filtrating is a viable approach to reconstruction harmonization and can be used to match a reference curve or to provide a pairwise approach where two individual scanner/reconstructions can be harmonized to one another. Many sites have more than one PET/CT scanner, and having the capability to meaningfully compare a subject’s quantitative SUV values from one scanner to another (for example in response to therapy situations) has significant value. This work presents a large set of usable OHF FWHM values for each of these cases. Supported by NIH/NCI R01CA169072 ER -