RT Journal Article
SR Electronic
T1 Simultaneous SUV/Patlak-4D Whole-Body PET: a Multi-Parametric 4D Imaging Framework for Routine Clinical Application
JF Journal of Nuclear Medicine
JO J Nucl Med
FD Society of Nuclear Medicine
SP 367
OP 367
VO 57
IS supplement 2
A1 Nicolas Karakatsanis
A1 Martin Lodge
A1 Guillaume Fahrni
A1 Michael Casey
A1 Yun Zhou
A1 Rathan Subramaniam
A1 Habib Zaidi
A1 Arman Rahmim
YR 2016
UL http://jnm.snmjournals.org/content/57/supplement_2/367.abstract
AB 367Objectives Single-frame whole-body (WB) PET imaging relies on the established metric of standardized uptake value (SUV), which is however considered semi-quantitative as it does not account for blood tracer concentration and depends on the post-injection scan time. On the other hand, dynamic PET and subsequent post-reconstruction parametric imaging has been constrained to a single bed and may trigger high noise levels, thus hampering its translation into clinical routine. In this study, we propose a novel, multi-parametric and clinically feasible SUV/Patlak 4-dimensional (4D) WB PET imaging framework, supporting the production of both SUV and quantitative 4D Patlak WB images from a single WB PET scan. Although this approach can be applicable to a range of isotopes and tracers, we have validated and optimized here for the well-established F18-FDG tracer.Methods The proposed 4D WB PET multi-pass scan protocol involves the same amount of administered radiotracer dosage and standard-of-care acquisition time-window routinely employed in the clinic for conventional WB single-pass studies (60-80min post-injection for FDG PET scans) to ensure high clinical adoptability. The WB SUV PET images can be simply generated by first reconstructing all dynamic PET images and then adding them together for each bed position (image-based synthesis). Alternatively, the dynamic sinograms may first be added together for each bed, followed by 3D PET reconstruction of the synthesized sinogram at each bed position with the proper normalization coefficients (projection-based synthesis). In both cases the average count statistics of a single-pass scan can be attained in the synthesized SUV images, while achieving better temporal uniformity across bed positions. In addition, we apply a highly quantitative and robust direct 4D parametric reconstruction algorithm that employs a nested iterative generalized Patlak (gPatlak) reconstruction method along with a cross-validated population-based input function in order to i) further enhance quantification in the case of non-negligible tracer uptake reversibility (k4) and ii) efficiently reconstruct tracer uptake rate (Ki) images with lower noise levels and high convergence acceleration rate. Finally, we proposed initializing the non-linear gPatlak 4D reconstruction algorithm with the Ki image estimates of the initial iterations of a linear standard Patlak (sPatlak) 4D reconstruction to achieve correct convergence. The methods have been validated on a set of 5 WB dynamic patient scans.Results Our evaluation of the proposed method has demonstrated i) its clinical feasibility, ii) a very good SUV agreement between single-pass scans and image-based synthesis (<10% difference) and iii) contrast-to-noise ratios that were improved by 60-80% for the direct 4D gPatlak compared to indirect post-reconstruction Patlak analysis. Although synthetic SUV images exhibited suppressed noise levels, tumor to background ratios (TBR) and contrast-to-noise (CNR) scores were consistently higher by at least 5% in 4D Patlak Ki images, thus suggesting enhanced lesion detectability when gPatlak-4D reconstruction complements conventional SUV image generation. Moreover, projection-based synthesis improved TBR and CNR scores only by <7% compared to simpler image-based synthesis, implying the latter type of synthesis can be adequate and more practical. Finally, at least ~60 ML-EM sPatlak iterations (3 full OS-EM iterations with 21 subsets) are recommended for proper gPatlak initialization, while a total of 20 nested (s/g)Patlak iterations appears to provide the optimal convergence rate.Conclusions We designed a readily adoptable, efficient and robust clinical PET imaging framework for the simultaneous generation of established SUV as well as highly quantitative 4D Patlak images in the clinical routine and without affecting current patient throughput.