Optimization of Q.Clear reconstruction for dynamic 18F PET imaging

Elisabeth Kirkeby Lysvik; Lars Tore Gyland Mikalsen; Mona-Elisabeth Rootwelt-Revheim; Kyrre Eeg Emblem; Trine Hjørnevik

doi:10.1186/s40658-023-00584-1

Optimization of Q.Clear reconstruction for dynamic ¹⁸F PET imaging

EJNMMI Phys. 2023 Oct 20;10(1):65. doi: 10.1186/s40658-023-00584-1.

Authors

Elisabeth Kirkeby Lysvik^{1

2}, Lars Tore Gyland Mikalsen^{3

4}, Mona-Elisabeth Rootwelt-Revheim^{5

6

7}, Kyrre Eeg Emblem³, Trine Hjørnevik^{3

5}

Affiliations

¹ Department of Physics and Computational Radiology, Division of Radiology and Nuclear Medicine, Oslo University Hospital, Building 20, Gaustad Sykehus, Sognsvannveien 21, 0372, Oslo, Norway. eliski@uio.no.
² Institute of Clinical Medicine, University of Oslo, Oslo, Norway. eliski@uio.no.
³ Department of Physics and Computational Radiology, Division of Radiology and Nuclear Medicine, Oslo University Hospital, Building 20, Gaustad Sykehus, Sognsvannveien 21, 0372, Oslo, Norway.
⁴ Department of Life Sciences and Health, Oslo Metropolitan University, Oslo, Norway.
⁵ Institute of Clinical Medicine, University of Oslo, Oslo, Norway.
⁶ The Intervention Centre, Oslo University Hospital, Oslo, Norway.
⁷ Department of Nuclear Medicine, Division of Radiology and Nuclear Medicine, Oslo University Hospital, Oslo, Norway.

Abstract

Background: Q.Clear, a Bayesian penalized likelihood reconstruction algorithm, has shown high potential in improving quantitation accuracy in PET systems. The Q.Clear algorithm controls noise during the iterative reconstruction through a β penalization factor. This study aimed to determine the optimal β-factor for accurate quantitation of dynamic PET scans.

Methods: A Flangeless Esser PET Phantom with eight hollow spheres (4-25 mm) was scanned on a GE Discovery MI PET/CT system. Data were reconstructed into five sets of variable acquisition times using Q.Clear with 18 different β-factors ranging from 100 to 3500. The recovery coefficient (RC), coefficient of variation (CV_RC) and root-mean-square error (RMSE_RC) were evaluated for the phantom data. Two male patients with recurrent glioblastoma were scanned on the same scanner using ¹⁸F-PSMA-1007. Using an irreversible two-tissue compartment model, the area under curve (AUC) and the net influx rate K_i were calculated to assess the impact of different β-factors on the pharmacokinetic analysis of clinical PET brain data.

Results: In general, RC and CV_RC decreased with increasing β-factor in the phantom data. For small spheres (< 10 mm), and in particular for short acquisition times, low β-factors resulted in high variability and an overestimation of measured activity. Increasing the β-factor improves the variability, however at a cost of underestimating the measured activity. For the clinical data, AUC decreased and K_i increased with increased β-factor; a change in β-factor from 300 to 1000 resulted in a 25.5% increase in the K_i.

Conclusion: In a complex dynamic dataset with variable acquisition times, the optimal β-factor provides a balance between accuracy and precision. Based on our results, we suggest a β-factor of 300-500 for quantitation of small structures with dynamic PET imaging, while large structures may benefit from higher β-factors.

Trial registration: Clinicaltrials.gov, NCT03951142. Registered 5 October 2019, https://clinicaltrials.gov/ct2/show/NCT03951142 . EudraCT no 2018-003229-27. Registered 26 February 2019, https://www.clinicaltrialsregister.eu/ctr-search/trial/2018-003229-27/NO .

Keywords: Dynamic PET; Q.Clear; Quantitation; Recovery coefficient; β-factor.

Associated data

ClinicalTrials.gov/NCT03951142

Grants and funding

303249/Norges Forskningsråd