The effect of reduced administered activity and image reconstruction on SUVR quantification in F-18 flutemetamol amyloid PET-MR scanning

Abstract

Aim/Introduction: Quantifcation of brain amyloid deposits and the classifcation of patients as amyloid positive or negative has potential value in Alzheimer research and clinical diagnosis [1]. Suitable radiotracers have been developed together with methods to standardize quantifcation across patients and radiotracers [2]. The use of reduced administered activity may enable applications such as repeat longitudinal scanning and screening. We evaluate whether signifcant reductions in injected activities are possible without signifcant changes in amyloid deposit quantifcation and to evaluate the role of diferent reconstruction algorithms. This evaluation was conducted by subsampling full dose [18F] futemetamol (FLUT) PET-MR data and using commercially available reconstruction algorithms. Materials and Methods: PET data acquired on a PET-MR GE Signa of 7 scans (6 healthy elderly participants), corresponding to 90-110minutes postinjection of ~185MBq of FLUT was subsampled 10 times to mimic 10% injected activity.PET reconstructions: Three reconstruction algorithms were assessed: VPFX (OP-OSEMTOF) with/without resolution modelling (SharpIR) (both 4 iterations-28 subsets) and Q.Clear regularised reconstruction. For VPFX a variety of Gaussian post-reconstruction flters and for Q.Clear a variety of beta values were assessed using Flangeless Esser-PET-phantom™ (visual and quantifcation assessment). Optimal β values resulted in β=250 (fulldose) and β=1000 (10% dose).Analysis: The images, fulldose OP-OSEM-TOF with 4mm-Gaussian (clinical standard) and QClear β=1000 for 10% dose, were read by an expert neurologist to determine the amyloid classifcation and confdence of classifcation. Statistical parametric mapping (SPM12) and centiloid regions were used to calculate SUVR values of cortical to cerebellum grey matter (GM) concentrations. Spatial normalisation was conducted using a T1 -weighted MPRAGE MR sequence and to refne regions to predominantly GM voxels, with both non-refned and GMrefned SUVR values assessed. Results: Five participants were classifed as amyloid negative and one positive. All images were found to return high diagnostic-confdence levels with exception of one participant (intermediate confdence) at 10% dose (Qclear β=1000), a participant that displayed some signifcant atrophy. Quantitatively, very small diferences in SUVR values were observed between dose levels and diferent reconstructions that were notably less than the diferences between participants and regions (non-refned or GM-refned). Estimates of imprecision from resampling at a 10% dose were also small. Conclusion: Large reductions of the doses are possible without signifcant deterioration in SUVR accuracy and precision. Furthermore, although some deterioration is visible, images can still be read and classifed with confdence. Further work is necessary with a larger more clinically relevant cohort.