Optimization of reconstruction in quantitative brain PET images: Benefits from PSF modeling and correction of edge artifacts
Emilie Verrecchia‐Ramos,Merwan Ginet,Olivier Morel,Marc Engels‐Deutsch,Sinan Ben Mahmoud,Paul Retif
DOI: https://doi.org/10.1002/mp.17419
IF: 4.506
2024-09-19
Medical Physics
Abstract:Background Modern PET reconstruction algorithms incorporate point‐spread‐function (PSF) correction to mitigate partial volume effect. However, PSF correction can introduce edge artifacts that lead to quantification errors. Consequently, current international guidelines advise against using PSF correction in brain PET reconstruction. Purpose We aimed to investigate PSF‐induced quantification errors in recent digital PET systems and identify conditions that mitigate them. This study utilized brain PET imaging with alginate‐based realistic phantoms, simulating lesion‐to‐background activity ratios of 10:1 and 2:1, with eleven reconstruction parameter sets. Methods Phantoms were prepared using a commercial anthropomorphic head phantom and two homemade inserts. Each insert contained a homogeneous 18F‐FDG alginate background with hot spheres of varying diameter (3, 4, 6, 8, 10, 12, and 15 mm). PET imaging was conducted on a digital PET‐CT system Biograph Vision 600 (Siemens), with a 10 min scan duration. Imaging was performed with and without PSF correction, with 2, 4, 6, 12, 18, or 24 iterations in reconstruction, and with or without additional Gaussian postfiltering. We assessed the recovery coefficient (RC), contrast recovery coefficient (CRC), variability, and CRC‐to‐variability ratios for each sphere size and reconstruction parameter set. Results PSF‐corrected images of the 10:1 spheres exhibited a nonmonotonic CRC‐to‐sphere diameter relationship due to edge artifacts overshoot in the 10 mm‐diameter sphere. In contrast, PSF images of the 2:1 spheres showed a monotonically increasing relationship. Non‐PSF images of both phantoms showed an expected monotonically increasing CRC‐to‐sphere diameter relationship but with lower CRC values compared to PSF images. The nonmonotonic relationship observed with 10:1 spheres was mitigated by applying a 3‐mm FWHM Gaussian postfiltering. For both phantoms, reconstructions with 6 iterations, PSF correction, and additional 3‐mm FWHM Gaussian postfiltering demonstrated the highest CRC‐to‐variability ratios. Conclusions Our findings indicate that Gaussian postfiltering suppresses PSF artifacts. This parameter set corrected the nonmonotonic CRC‐to‐sphere diameter relationship and improved the CRC‐to‐variability ratio compared to non‐PSF reconstructions. Therefore, to enhance lesion detectability without compromising quantification accuracy, PSF correction coupled with Gaussian postfiltering should be used in brain PET.
radiology, nuclear medicine & medical imaging