Feasibility of Population-Based Input Function for Kinetic Analysis of [11C]-DPA-713

Abstract

Quantitative PET studies of neurodegenerative diseases typically require the measurement of arterial input function (AIF), an invasive and risky procedure. The aim of this study was to assess the accuracy of population-based input function (PBIF) for [11C]DPA-713 PET kinetic analysis. The final goal is to possibly eliminate the need for AIF. Eighteen subjects from two [11C]-DPA-713 PET protocols, including six (6) healthy and twelve (12) Parkinson Disease (PD) subjects, were included in this study. Each subject underwent 90min dynamic PET imaging on a Siemens Biograph mCT™ scanner. Five of the six healthy subjects underwent a Test/Retest within the same day to assess the reproducibility of the kinetic parameters. Kinetic modeling was carried out with 2-tissue compartment model (2TCM) as well as with the Logan VT model using the PBIF, and again with the patient-specific AIF (PSAIF, gold standard). Using the leave-one-out cross validation method, we generated a PBIF for each subject from the remaining 17 subjects after normalizing the PSAIFs by three techniques: (a) patient weight×injected dose (b) Area Under AIF Curve (AUC), and (c) weight×AUC. The variability in the total distribution volume (VT) and non-displaceable binding potential (BPND) due to the use of PBIF was assessed for some brain regions of interest using Bland-Altman analysis, and for the three normalization approaches. Systematic bias was noticed with the test-retest scans, but this was removed by normalizing with gray matter. Better repeatability was obtained with the Logan VT model where the 95% limits of agreement (LoA) lie within ±20% for all the brain regions. Also, % relative difference between PBIF and PSAIF is significantly different across the normalization techniques, with the normalization by weight×AUC yielding the least % relative difference. For the Bland-Altman analysis, the mean % difference for VT lies within ±2% and the 95% LOA lies within ±40%. For the BPND, the mean difference lies within ±4% and the corresponding 95% LOA is ±80%. In all cases, the variability between PBIF and PSAIF lie within the test-retest repeatability. This study shows that PBIF-based kinetic modelling is feasible, and that better repeatability is achieved with Logan VTmodelling.