PT - JOURNAL ARTICLE AU - Georgios SOULTANIDIS AU - Nicolas Karakatsanis AU - Yang Yang AU - Philip Robson AU - Zahi Fayad TI - PET reconstruction strategies for motion correction in coronary PET/MR imaging DP - 2019 May 01 TA - Journal of Nuclear Medicine PG - 2025--2025 VI - 60 IP - supplement 1 4099 - http://jnm.snmjournals.org/content/60/supplement_1/2025.short 4100 - http://jnm.snmjournals.org/content/60/supplement_1/2025.full SO - J Nucl Med2019 May 01; 60 AB - 2025Introduction: Recently, Sodium Fluoride (18F-NaF) PET imaging of the coronary arteries enabled the visualization and quantification of microcalcifications. Initial studies have indicated the necessity of a motion correction method to compensate two asynchronous, yet rhythmic motions; the cardiac and the respiratory. In this study, our aim is to systematically evaluate the performance of two well-known methods for motion-corrected PET reconstruction, reconstruct-transform-average (RTA) and motion-corrected image reconstruction (MCIR), and compare them against a novel and computationally efficient clustered MCIR (cMCIR) method. Our comparisons were conducted in the setting of cardio-respiratory motion correction for coronary PET imaging, using anthropomorphic PET/MR simulations as a common validation platform. Methods: The RTA method involves the independent reconstruction of each gate followed by transformation to a reference gate, according to the estimated motion fields, and averaging. The MCIR method incorporates the motion fields within the reconstruction to allow the direct reconstruction of motion-corrected PET images. However, the complexity of cardiorespiratory motion requires a high number of gates, which places high demands on RTA, in terms of low-count gates, and MCIR, in terms of computational complexity. We propose a novel motion correction approach involving the clustering of the total number of cardiac and respiratory gates into subsets of balanced counts followed by the parallel application of the MCIR approach for each subset and averaging of the resulting motion-corrected images according to the total scan duration of each subset. For the comparative evaluation of the three above approaches, we utilized the anthropomorphic XCAT computational phantom and simulated a 30-minute PET acquisition of a patient with cardiorespiratory motion. Additionally, we simulated three coronary plaques of 2 x 5 mm and a lesion-to-background ratio (LBR) of 6. The XCAT PET and attenuation data were re-binned to 16 gates, (4 cardiac x 4 respiratory combinations). Motion fields were calculated from the binned XCAT attenuation maps with the hierarchical adaptive local affine registration algorithm. For the PET acquisition simulation as well as the application of the three motion correction strategies we used the libraries of the Software for Tomographic Image Reconstruction (STIR). For the cMCIR method, we clustered the data into 4 balanced subsets, each comprised of 4 gates. All reconstructions were based on the ordered subsets expectation maximization (OSEM) algorithm for 3 iterations with 21 subsets per iteration. Afterwards, we visually segmented each plaque region of interest, extracted the mean Standardized Uptake Value (SUVmean) of each plaque ROI and evaluated the LBR and contrast to noise ratio (CNR) using as a background reference region the heart left ventricle. Results: The LBR was highest for MCIR, while RTA and cMCIR performed similarly. For lesion 1, the contrast recovery was at 78% for MCIR, 72% for RTA and 73% for cMCIR, against the motionless static acquisition. For CNR, the highest values were given by RTA and MCIR, with approximately 20% losses for lesion 1 and 60% for lesions 2 and 3. The cMCIR yielded approximately 7% higher losses than RTA. Conclusions: MCIR and RTA motion correction methods performed similarly, in terms of contrast recovery and CNR of coronary lesions, with MCIR achieving consistently higher scores. The introduced cMCIR approach demonstrated a comparable performance to that of the MCIR and RTA methods. All three motion correction methods significantly improved coronary lesions detectability, relative to no motion correction, with the cMCIR method offering the additional benefit of more efficient utilization of available computational resources in clinic.