Impact of patient size on image quality of OSEM3D and BSREM reconstructions in [68Ga]Ga-DOTA-TATE PET/MR

Christina P W Cox; Tessa Brabander; Frederik A Verburg; Marcel Segbers

doi:10.1186/s40658-025-00777-w

Impact of patient size on image quality of OSEM3D and BSREM reconstructions in [⁶⁸Ga]Ga-DOTA-TATE PET/MR

EJNMMI Phys. 2025 Jul 1;12(1):60. doi: 10.1186/s40658-025-00777-w.

Authors

Christina P W Cox¹, Tessa Brabander², Frederik A Verburg², Marcel Segbers²

Affiliations

¹ Department of Radiology & Nuclear Medicine, Erasmus Medical Centre, Postbus 2040, 3000 CA, Rotterdam, The Netherlands. c.cox@erasmusmc.nl.
² Department of Radiology & Nuclear Medicine, Erasmus Medical Centre, Postbus 2040, 3000 CA, Rotterdam, The Netherlands.

Abstract

Background: Previous [⁶⁸Ga]Ga-DOTA-TATE PET/CT studies using ordered subset expectation maximization (OSEM3D) based reconstruction algorithms, demonstrated non-linear relations between body weight and image quality. Block Sequential Regularized Expectation Maximization (BSREM) algorithm reduces noise amplification during reconstruction. The impact of the reconstruction algorithm on the relation between image quality and patient size in [⁶⁸Ga]Ga-DOTA-TATE PET/MR may differ from PET/CT and OSEM3D. Therefore, the aim of this study is to investigate the relation between patient size and image quality in OSEM3D and BSREM [⁶⁸Ga]Ga-DOTA-TATE PET/MR reconstructions.

Methods: [⁶⁸Ga]Ga-DOTA-TATE PET/MR images of 55 patients were included. Images were reconstructed using OSEM3D (VUE Point FX SharpIR, 4 iterations, 28 subsets and 7 mm Gaussian filter) and BSREM (Q.Clear, β = 300). Liver signal-to-noise ratio (SNRliver) normalized for injected activity and acquisition time (SNRliver,norm) was measured to perform curve fitting with patient-dependent parameters using fixed, linear and non-linear fit models, followed by Akaike's corrected information criterion (AICc) model selection.

Results: BSREM mean SNRliver was significantly (p < 0.001) higher than OSEM3D mean SNRliver. Body mass, the best patient-dependent parameter for both algorithms, clarified 40% (linear model) and 53% (non-linear model) of the variability in SNRliver,norm for OSEM3D and 20% (linear model) and 21% (non-linear model) for BSREM. AICc preferred a non-linear model for OSEM3D and a linear model for BSREM.

Conclusion: The image quality predictor body weight is a weaker predictor for BSREM than for OSEM3D image quality in [⁶⁸Ga]Ga-DOTA-TATE PET/MR. Therefore, a linear dosage regimen based on body weight is preferable for BSREM, whereas a quadratic dosage regimen based on body weight is optimal for OSEM3D.

Keywords: BSREM; Block-sequential regularized expectation maximization; Body weight; Dosage regimen; Image quality; Neuroendocrine tumour; OSEM; Ordered subset expectation maximization; PET/MR; Patient size; [68Ga]Ga-DOTA-TATE.