Can SUVpeak derived from PSF 18F-PET/CT images act as a surrogate for SUVmax derived from EARL compliant reconstructions?

Objectives: We explored the possibility of using SUVPEAK (defined as a 1mL sphere with the highest average SUV within a lesion/target) derived from 18F-FDG PET/CT images reconstructed with point-spread-function resolution modelling as a surrogate for SUVmax derived from images reconstructed with an EARL compliant protocol. SUVPEAK was already suggested as being a better alternative than SUVmax when derived from PSF images1. This assessment was performed using phantom images and clinical data from stage III and IV non-small cell lung cancer (NSCLC) patients. Methods The National Electrical Manufacturers Association (NEMA) image quality phantom was scanned for 5 minutes with a 10:1 sphere-to-background contrast ratio. Additionally, ten stage III and IV NSCLC patients underwent two baseline 18F-FDG PET/CT scans per day at 60min and 90min post injection, at two separate time-points within one week2. All data were reconstructed following two protocols: one compliant with EANM guidelines (EARL multicentre standard) and another with point-spread-function (PSF) resolution modelling. The latter protocol allows for improved lesion detection, but does not provide EARL compliant quantitative results. Patients' lesions were segmented using a semi-automatic isocontour at SUV=4.0 g/mL. NEMA spheres were manually delineated based on their diameter. The SUVPEAK PSF-derived (SUVPEAKPSF) was compared with SUVmax EARL-derived (SUVmaxEARL) with Pearson's correlations, t-tests and Bland-Altman plots. Results For the phantom scans, the mean relative difference between SUVPEAKPSF and SUVmaxEARL was -11.0% (SD=15.0%) and showed a correlation ρ=0.95. Considering only the three biggest spheres (volumes larger than 5 mL), the mean relative difference was 1.8% (SD=3.5%). The clinical dataset (total of 21 lesions analysed) presented similar results, with a perfect correlation between SUVPEAKPSF and SUVmaxEARL (ρ=1.00) and a mean relative difference of -4.3% (SD=5.2%). This highlights that, overall, SUVPEAKPSF values are slightly lower than SUVmaxEARL values. Only two lesions were identified as smaller than 5 mL using the EARL reconstruction, three with PSF. The relative difference between SUV implementations for those small lesions was -7.1% (SD=4.3%). Moreover, uptake time did not affect the differences seen between SUVPEAKPSF and SUVmaxEARL, which were -4.2% (SD=5.0%) and -4.3% (SD=5.6%) for the 60 and 90 minutes post-injection data (p=0.89), respectively. Conclusion This pilot study