Effects of reconstruction parameters on the image quality and quantification of PET images from PET/MRI and PET/CT systems

Abstract

Aim: To study how reconstruction parameters affect the positron emission tomography (PET) image quality and quantitative results for the different lesion to background radioactivity ratios in three different PET systems. Introduction: Multimodality imaging that combines magnetic resonance imaging (MRI) or computed tomography (CT) with a PET system can produce medical images containing both functional and anatomical information. The most used PET reconstruction algorithm in clinical systems is Ordered Subset Expectation Maximization (OSEM) with time-of-flight (TOF) and point spread function (PSF). In the OSEM algorithm, the image noise increases as the number of iterations increases. Thus, the reconstruction needs to be stopped before a complete convergence can be achieved. The Bayesian Penalized Likelihood (BPL) reconstruction algorithm (‘’Q-clear’’) has been newly introduced to reconstruct PET images, which applies a penalty method for image noise suppression so that the iterations can continue to full convergence. The image quality and noise suppression in the BPL can be controlled by the noise penalty factor (β). BPL algorithms are shown to improve signal-to-noise in PET images. Method: A NEMA IQ phantom was scanned on GE Signa PET/MRI, GE Discovery MI PET/CT, and Siemens Biograph mCT PET/CT system with 2:1, 4:1, and 10:1 sphere-to-background radioactivity concentration ratios of the 2-[¹⁸F]FDG solution. Acquired list-mode data were used to reconstruct PET images with either OSEM or Q-clear algorithms, with and without TOF. The number of iterations and β-values were varied, while the matrix size, number of subsets, and filter size were kept constants for all reconstructions. After reconstruction, the images were evaluated and compared using the NEMA analysis tools available for each system, using automatic localisation of the region-of-interests (ROI). Contrast recovery (CR) and background variability (BV) values were determined for each ROI in all reconstructed PET images to assess the image quality and quantification accuracy. Result: Results showed that CR increased with increased sphere size from 10 mm to 22 mm in diameter and activity concentration ratios (sphere to background) from 2:1 to 10:1. CR and BV decreased gradually in reconstructed images with increased β-values for the smallest sphere, i.e., 10 mm in diameter. Increased number of iterations in OSEM algorithm showed to increase BV with low significant variation of CR. The comparison between reconstruction algorithms showed higher CR values and lower BV values with Q-clear than with OSEM. Reconstructed PET images with TOF showed higher CR and lower BV than reconstructions without TOF for both algorithms. The optimal reconstruction parameters were for GE-Signa and Discovery MI systems a β-value between 150 and 350 for TOF Q-clear, and three iterations, 16 subsets, 5 mm FWHM Gaussian filter for TOF OSEM. For the Biograph mCT system, the optimal reconstruction parameters were two iterations, 21 subsets, and 5 mm FWHM Gaussian filter for OSEM algorithm with TOF. Conclusion: PET images acquired on GE Discovery MI PET/CT and reconstructed with the Q-clear algorithm provided the best image quality and quantitative accuracy for the smallest sphere.