Performance evaluation of various reference tissue input parametric methods: For [18F]FDDNP

Introduction: [18F]FDDNP is a PET ligand that has been introduced for imaging neurofibrillary tangles and beta-amyloid fibrils in the brain. Recently, Kepe et al.[1] studied [18F]FDDNP binding in patients with Alzheimer Disease (AD) using reference Logan analysis[2]. The purpose of the present study was to study performance of several reference tissue parametric Methods: for measuring [18F]FDDNP binding. Methods: The following parametric Methods: were evaluated: reference Logan[2], two basis function Methods: (RPM1 & RPM2[3,4]), and various multi-linear Methods: (MRTMo, MRTM, MRTM2[5]) and two newly developed multi-linear Methods: based on MRTMo (MRTM3 and MRTM4). RPM2 and MRTM2,3,4 all include fixing the reference tissue clearance rate (k2'; using RPM1, MRTM2, MRTMo and MRTM1 respectively). Both simulations and clinical data were used to determine the effects of flow, fractional blood volume (Vb) and binding potential (BP) on accuracy and precision of parametric BP. Simulations were done by varying only one parameter and fixing other parameters to default values (R1=0.9, k2=0.07 & BP=0.2). Various simulated [18F]FDDNP time activity curves (TAC) were generated at a 15% noise level using plasma input. Clinical [18F]FDDNP data were obtained from 3 controls, 3 MCI and 3 AD subjects. TAC and/or regional average parametric data were derived for 8 regions of interest (ROI) across the frontal cortex. Grey matter cerebellum was used as reference tissue. ROI were projected on parametric images for comparison with BP obtained using SRTM. Results: Simulations over a range of BP values (0.1-0.4) showed a maximum BP bias of 7% (compared with SRTM) for all Methods:, which decreased with increasing BP. Lowest BP bias was found for MRTM2 (<0.7±18%) and reference Logan (< 1.7±23%). MRTM showed very poor precision of 9-216% for BP = 0.4 to 0.1. BP bias did not exceed 4% for all Methods and for a range of simulated flows (R1=0.6 to 0.9). Precisions equaled ∼10% except for MRTM1 (10-65%). MRTM2 yielded lowest bias (11±10%) over a range of Vb (0.025-0.075). For most Methods: clinical parametric BP data showed correlations of R2=∼0.8 with BP-SRTM. The only exception was MRTM (R2=0.49). Lowest biases were found for MRTM2, RPM2 and MRTM4 (1, 3 and 3%, respectively). For all other Methods: bias exceeded 10% compared with SRTM. Although parametric data showed a trend towards increased BP in AD subjects, none showed a significant difference with healthy subjects, in line with the findings from SRTM. Visual inspection of parametric images showed artifacts for voxels near and in CSF and skull for most multi-linear Methods:. RPM2 provided 'best' BP images without these artifacts and these images appeared less noisy. Conclusion: Most parametric Methods: evaluated showed similar performance compared with SRTM for brain tissue voxels. Based on observed accuracies, precisions and image quality, RPM2 is recommended for generating parametric BP (and R1) images.