Motion-compensated 90Y-PET/MR image reconstruction for enhanced localization and quantification of liver 90Y dose maps

Objectives: 90Y PET has been employed to image and quantify the absorbed 90Y dose distribution in liver tumors and surrounding normal tissue, following 90Y radioembolization [1]. However, PET/CT liver dose maps can suffer from respiratory motion-induced blurring, due to the relatively poor 4D motion tracking capabilities of the CT, and the low 90Y PET signal, due to low positron yield [2]. In fact, clinical PET/CT studies have shown 90Y dose differences of as low as 25% between responding and non-responding lesions [3]. Nevertheless, the advent of simultaneous PET/MR enabled the utilization of the superior soft-tissue contrast of the MR to improve tumors localization and determine a stronger correlation between delivered dose and tumor’s response [4]. In this study, we aim to exploit the superior MR-based motion tracking capabilities of clinical PET/MR and employ a 4D respiratory motion-compensated 90Y PET image reconstruction method to improve the accuracy of the 90Y liver dose maps without further amplifying the statistical noise. Methods: A cohort of 21 patients (66.7 +/- 9.9 y.o., 77.7 +/- 14 kg) underwent a PET/MR scan at a Siemens Biograph mMR scanner, starting 3.3 +/- 0.8 h post injection of a 2 +/- 1 GBq of 90Y dosage. The PET/MR protocol consisted of a 20min list-mode (LM) PET acquisition, in parallel to a 19sec VIBE MR sequence with 2-point Dixon fat/water separation, to support PET attenuation correction with 4-class tissue (air, lungs, fat, water) segmentation, and a 20min 3D radial VIBE MR prototype sequence with self-navigation to track the respiratory motion phase and accordingly sort the MR and the synchronized LM PET data into 5 gates from end-expiration to end-inspiration. The LM PET gates were later histogrammed into 5 sinogram gates, respectively. Hierarchical diffeomorphic image registration was then applied to the 5 MR gates to derive the 3D motion transformations between a reference gate (end-expiration) and each of the other gates (forward model) and vice-versa (reverse model). Subsequently, we developed a sinogram-based 4D motion-compensated image reconstruction (MCIR) method by incorporating the forward and reverse motion components into the forward- and back-projection operators, respectively, of an Ordinary Poisson Ordered Subsets Expectation Maximization (OP-OSEM) algorithm. Finally all the gated PET sinograms were passed to the 4D MCIR algorithm to produce a) a motion-compensated (MC) 90Y dose map and compare it against dose maps reconstructed from b) non-gated non-MC, as well as c) gated (end-expiration) 90Y data. Results: Visual inspection and quantitative evaluation along line profiles intersecting points of suspected focal liver uptake demonstrated that the in-vivo application of the 4D MCIR method on 90Y-PET/MR liver data for all 21 cases resulted in significantly higher signal contrast recovery (12% higher line profile peaks on average) compared to non-gated non-MC dose maps for the same noise levels. In addition, the dose maps derived with the 4D MCIR method were associated with significantly lower noise levels and higher signal-to-noise ratio (smoother line profiles with more concentrated peaks), compared to the gated dose maps at end-expiration position. Finally, the alignment between the 90Y dose maps and the MR-defined liver regions was significantly improved with the 4D MCIR method, compared to non-gated non-MC dose maps, as suggested by the elimination of 90Y dose outside the liver after 4D MCIR in nearly all cases. This is of high importance in clinical practice, as occasionally a percentage of 90Y dose may be observed in lungs due to air embolization.

Conclusions: 90Y-PET/MRI exploiting MR-based respiratory motion models and 4D motion-compensated PET image reconstruction may provide highly accurate 90Y dose liver maps to significantly enhance hepatic tumor response assessments after 90Y radioembolization. Research Support: This work was supported by NIH/NHLBI R01HL071021 grant.

• Download figure
• Open in new tab
• Download powerpoint