Optimization of reconstruction parameters of a block sequential regularized expectation maximization algorithm for 68Ga-DOTOTAC PET/CT studies

Thomas Quist Christensen, Mie Holm Vilstrup, Anne Lerberg Nielsen, Poul-Erik Braad

Research output: Contribution to journalConference abstract in journalResearchpeer-review


Aim/Introduction: Image reconstruction in nuclear medicine
has widely been using the ordered-subset-expectationmaximum (OSEM) algorithm. In recent years, a commercially
available block sequential regularized expectation maximization
(BSREM) algorithm was introduced to improve reconstruction
convergence with suitable noise control. The aim of the present
study was to investigate the impact of the BSREM penalizing
factor (β) on image noise and activity quantitation, and to
investigate if optimizations might facilitate lower scan times or
activity administrations in 68Ga-DOTOTAC PET/CT. Materials and
Methods: The background of a NEMA/IEC torso phantom was
flled with a homogeneous aqueous 68Ga-solution at an activity
concentration corresponding to that of the liver in clinical 68GaDOTOTAC examinations. Spherical inserts (diameters: 10-37
mm) were flled at a 5:1 sphere-to-background activity contrast.
Scans were performed on a Discovery MI 25 cm axial feld-ofview PET/CT scanner (GE Healthcare) at acquisition times of 10,
20, 35, 60 and 90 seconds/bed and reconstructed using either
PSF-OSEM (4 iterations, 17 subsets, 5 mm Gaussian post-flter) or
BSREM with β-values in the interval 300-3000. Contrast recovery
curves (RC) were calculated for each sphere. Four 50 mm VOIs
was placed in the background and image noise compared
across scan and reconstruction settings. Six anonymized
clinical 68Ga-DOTOTAC-examinations were acquired in listmode
at 90 s per bed-position and retrospectively reframed and
Eur J Nucl Med Mol Imaging (2021) 48 (Suppl 1): S1–S648
reconstructed similar to phantom studies. The image quality
in blinded patient and phantom scans were visually graded by
two experienced nuclear medicine physicians. Results: BSREM
with a β-value of 550 showed similar RC to PSF-OSEM while
lower or higher β-values resulted in, respectively, superior or
inferior RC’s, especially in larger spheres. In phantom studies,
the noise at an acquisition time of 90 s was signifcantly lower
for BSREM (β: 550) than for PSF-OSEM. Similar noise levels to 90
s PSF-OSEM, was obtained at 35 s with BSREM with no change
to the RC’s. Visual grading analysis by the physicians confrmed
the results. Diagnostic confdence of BSREM-reconstructed (β:
550) patient studies at acquisition times of 60 s was graded
similar to PSF-OSEM studies at 90 s. In both phantom and
patient studies, the physicians preferred β-values in the interval
550-800. Conclusion: Optimized β-values in BSREM image
reconstructions might lead to improved activity quantifcation
and image quality in 68Ga-DOTOTAC PET/CT examinations or be
utilized to lower scan times or administered activities compared
to standard PSF-OSEM image reconstruction.
References: none
Original languageEnglish
Article numberOP-0883
JournalEuropean Journal of Nuclear Medicine and Molecular Imaging
Issue numberSuppl. 1
Pages (from-to)305
Number of pages1
Publication statusPublished - 2021
EventAnnual Congress of the European Association of Nuclear Medicine 2021 -
Duration: 20. Oct 202123. Oct 2021


ConferenceAnnual Congress of the European Association of Nuclear Medicine 2021

Cite this