Reconstruction-dependent differences in brain [18F]FDG PET: a large-scale same-acquisition comparison of Q.Clear and VUE Point FX

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Objective To determine whether Q.Clear and VUE Point FX (VPFX) reconstructions obtained from the same brain [18 F]fluorodeoxyglucose positron emission tomography ([18 F]FDG PET) acquisition are quantitatively interchangeable, and to characterise the regional, reference-region and noise-texture dependence of any differences. Methods We retrospectively analysed paired Q.Clear and VPFX reconstructions from 1,145 clinical brain PET folders. The reviewed protocol used a 10-min acquisition, Q.Clear QCFX-S with beta 150, SharpIR and no z-axis filter, and VPFX-S with 3 iterations, 16 subsets and a 4.0-mm filter. Digital Imaging and Communications in Medicine metadata were audited, and 1,022 readable same-examination pairs were processed in MATLAB R2026a and statistical parametric mapping version 25. A common transform estimated from mean(Q.Clear, VPFX) was applied to both reconstructions. Absolute and standardised uptake value ratio (SUVR) differences normalised to whole brain, cerebellum and pons were assessed voxel-wise and in Automated Anatomical Labelling atlas 3 regions. DICOM-derived local high-frequency and robust global coefficients of variation (CVs) were also compared. Results Q.Clear increased the whole-brain, cerebellar and pontine reference means by 0.953%, 0.595% and 0.818%, respectively. Absolute regional differences were heterogeneous; after normalisation, most cortical and limbic Q.Clear-minus-VPFX differences were negative. The medial temporal whole-brain-normalised difference was − 0.00625 (Cohen dz -2.09), equivalent to approximately − 0.69% at an illustrative VPFX SUVR of 0.90. Covariate-adjusted voxel-wise T maps were nearly identical to no-covariate maps ( r = 0.9996–0.9999; sign concordance > = 99.96%). Local high-frequency CV was lower with VPFX in 689 pairs and with Q.Clear in 333, whereas robust global CV was lower with Q.Clear in 903 pairs, with one tie. Conclusions Q.Clear and VPFX were not quantitatively interchangeable under the evaluated protocol. The study establishes protocol-specific reconstruction-dependent differences, not the direction of quantitative accuracy, because no ground truth was available. Harmonised reconstruction and validation are required when quantitative brain PET data are compared or pooled.