Parametric imaging and quantitative analysis of the PET amyloid ligand [18F]flutemetamol

Abstract

ObjectivesThe amyloid imaging PET tracer [18F]flutemetamol was recently approved by regulatory authorities in the US and EU for estimation of β-amyloid neuritic plaque density in cognitively impaired patients. While the clinical assessment in line with the label is a qualitative visual assessment of 20min summation images, the aim of this work was to assess the performance of various parametric analysis methods and standardized uptake value ratio (SUVR), in comparison with arterial input based compartment modeling. MethodsThe cerebellar cortex was used as reference region in the generation of parametric images of binding potential (BPND) using multilinear reference tissue methods (MRTMo, MRTM, MRTM2), basis function implementations of the simplified reference tissue model (here called RPM) and the two-parameter version of SRTM (here called RPM2) and reference region based Logan graphical analysis. Regionally averaged values of parametric results were compared with the BPND of corresponding regions from arterial input compartment modeling. Dynamic PET data were also pre-filtered using a 3D Gaussian smoothing of 5mm FWHM and the effect of the filtering on the correlation was investigated. In addition, the use of SUVR images was evaluated. The accuracy of several kinetic models were also assessed through simulations of time–activity curves based on clinical data for low and high binding adding different levels of statistical noise representing regions and individual voxels. ResultsThe highest correlation was observed for pre-filtered reference Logan, with correction for individual reference region efflux rate constant k2′ (R2=0.98), or using a cohort mean k2′ (R2=0.97). Pre-processing filtered MRTM2, unfiltered SUVR over the scanning window 70–90min and unfiltered RPM also demonstrated high correlations with arterial input compartment modeling (MRTM2 R2=0.97, RPM R2=0.96 and SUVR R2=0.95) Poorest agreement was seen with MRTM without pre-filtering (R2=0.68). ConclusionsParametric imaging allows for quantification without introducing bias due to selection of anatomical regions, and thus enables objective statistical voxel-based comparisons of tracer binding. Several parametric modeling approaches perform well, especially after Gaussian pre-filtering of the dynamic data. However, the semi-quantitative use of SUVR between 70 and 90min has comparable agreement with full kinetic modeling, thus supporting its use as a simplified method for quantitative assessment of tracer uptake.