A systematic evaluation of five different image-derived input functions for the clinical implementation of 18F-NaF bone PET/CT in patients with chronic kidney disease–mineral and bone disorder

Abstract

IntroductionThe aim of this study was to investigate the impact of the use of varying input parameters on resulting bone plasma clearance (Ki) and other kinetic modelling parameters in a group of patients with chronic kidney disease–mineral and bone disorder (CKD-MBD).MethodsRaw PET/CT data and blood data were systematically analyzed using five different VOIs for the input functions in the left ventricle and in the thoracic aorta. Standardized VOIs were placed in four thoracic vertebrae and the results pooled and averaged. The basic image-derived input functions (IDIFs) were corrected for partial volume effect and spill-over and modified by substitution of the terminal image exponential with the corresponding plasma-exponentials derived from blood samples. Ki was then calculated using both a non-linear regression (NLR) analysis and a graphical Patlak analysis and compared.ResultsOur original results were reproducible with an inter-observer difference of approximately 6%. The correction factors varied with the VOI volumes from 0.73 ± 0.17 for the largest LV-VOI (48.7 ± 25.3 cm3) to 0.99 ± 0.10 for the AO-VOI (3.4 ± 1.2 cm3). The mean NLR-Ki results varied between 0.0378 ± 0.0112 and 0.0432 ± 0.0095 ml/min ml−1 with a fixed vB and 0.0408 ± 0.0111 and 0.045 ± 0.0102 ml/min ml−1 with a free-fitted vB. The corresponding Patl-Ki-results varied between 0.0302 ± 0.0071 and 0.0325 ± 0.0070 ml/min ml−1, having lesser differences and variances. The input functions with least variance and mean differences compared with NLR results were derived from the left ventricle with a VOI volume of 19.2 ± 11.3 cm3 corrected for PVE and Bg with a mean Ki-difference: 0.0097 ± 0.0370 ml/min ml−1 and 95% confidence limits (−0.023 to 0.004).ConclusionsOur results indicated that a VOI with a volume of approximately 20 cm3 with a correction factor of 0.83 ± 0.13 results in Patlak results with the least variance and difference compared with the NLR results. The use of free-fitted vB in the NLR analysis showed the most robust results in all input series. The Patlak results were in comparison generally lower than the NLR results (−17.3% to −23.4%) but very robust across the various input series and with results comparable to previously published data and are therefore recommended for future analysis.