Evaluation of compartment models and semi-quantitative measures for analysing [18F]FDDNP studies

Introduction: [18F]FDDNP is a PET ligand that has been introduced for imaging neurofibrillary tangles and beta-amyloid fibrils in the brain. Increased cerebral uptake of [18F]FDDNP in patients with Alzheimer's disease (AD) has been described[1]. The purpose of the present study was to evaluate quantification of [18F]FDDNP binding using standard compartment models[2]. In addition, potential semi-quantitative measures for routine clinical use were investigated. Methods: The following kinetic and semi-quantitative Methods: were assessed: plasma input single tissue (1T-2k), two tissue irreversible (2T-3k) and two tissue reversible (2T-4k) compartment models[2], simplified (SRTM) and full (FRTM) reference tissue models[3], standard uptake values (SUV) and SUV ratios (SUVr). Both simulations and clinical studies were used to assess the effects of binding potential (BP=0.1-0.4), flow and fractional blood volume (Vb=0.025-0.075) on precision and accuracy of estimated parameters. Various [18F]FDDNP time activity curves (TAC) were simulated at a 15% noise level using either reference tissue (RI) or plasma input (PI). In addition, [18F]FDDNP studies were performed in 4 control and 3 AD subjects. Methods: were compared using several frontal cortex TACs (typical AD regions). Grey matter cerebellum was used as reference tissue. Results: Simulations showed best BP accuracies for SRTM (<5% bias), with lower bias for higher BP. For the 2T-4k model, bias reduced (to <29%) with decreasing flow (R1=0.6-0.9), but was still best for SRTM (2.5% bias) and FRTM (5% bias). Effects of Vb on BP accuracy were small for 2T-4k (<4%), SRTM (<10%) and FRTM (<10%), although occasionally SRTM and FRTM showed outliers. According to the Akaike information criterion (AIC)[4], the 2T-4k model (72%) was preferred over 1T-2k (0%) and 2T-3k (28%) models in healthy subjects. This preference was even 100% in AD subjects. SRTM was preferred (72%) over FRTM in both control and AD subjects. Good correlations were found between BP-SRTM and BP-FRTM (R2=0.92) or SUVr over 60-90 min (R2=0.82). In this limited series of subjects, none of the Methods: provided a significant contrast between AD and healthy subjects, although 2T-4k BP showed a trend (p<0.1) for decreased binding in AD and 2T-3k a trend (p<0.1) for increased Ki in AD. The 2T-4k BP decrease might be explained by an observed increase of k4, possibly caused by metabolites entering the brain. SRTM did not show a significant difference as well. Simulations, however, indicated that reference tissue models are sensitive for changes in k4 between target and reference region. Conclusions: AIC indicates a preference for 2T-4k for plasma input and SRTM for reference tissue models. This quantification study will be extended to assess the impact of metabolites entering the brain (as potential cause of change in k4) on pharmacokinetic analysis.