Abstract
BACKGROUND: Head motion during dynamic positron emission tomography (dPET) compromises pharmacokinetic modeling, especially in prolonged acquisitions where motion artifacts invalidate quantitative analysis. Although motion correction methods are well-established for static PET, their application in dynamic imaging remains limited. This study evaluates the efficacy of a data-driven dynamic head motion correction (dHMC) algorithm in preserving the reliability of kinetic parameter, using dynamic (68) Ga-PSMA-11 PET of intracranial tumors as a challenging model due to the tracer’s inability to cross the blood–brain barrier, unlike the diffusely distributed (18)F-FDG used in prior studies. METHODS: Sixteen patients with suspected glioma based on preoperative imaging underwent 40-min dynamic (68) Ga-PSMA-11 PET head scans. Head motion was categorized into minor and high-motion groups based on maximum displacement (> 3 mm) or time-weighted mean displacement (> 2 mm) within the tumor region. The performance of dHMC was evaluated in tumor and reference organs (parotid and lacrimal glands) based on fitting metrics, including the coefficient of determination (R(2)) and Akaike information criterion (AIC), and quantitative kinetic parameters. Time-activity curves (TACs) and key kinetic parameters (k(2) and k(3)) were generated using an irreversible two-tissue compartment model (2T3k). RESULTS: dHMC robustly corrected motion-induced inaccuracies across motion levels, enabling analysis in 43.75% of previously non-analyzable cases. The algorithm yielded three major improvements without introducing bias: First, it significantly enhanced pharmacokinetic modeling fidelity, evidenced by smoother TACs, along with increased R(2) values of 8.35%, 4.54%, and 30.9% in tumor, parotid, and lacrimal glands, respectively, and reduced AIC values by 6.95%, 5.33%, and 13.32%. Second, it improved the accuracy and homogeneity of k(2) and k(3) estimates in reference organs, indicated by significantly reduced data dispersion. Third, it brought the quantitative parameters of the parotid gland closer to established reference standards. All improvements were most pronounced in high-motion cases. CONCLUSION: The dHMC algorithm enhances fitting performance, parametric accuracy, and inter-frame consistency in dynamic (68) Ga-PSMA-11 PET, validating its broad applicability across varying motion scenarios and underscoring the essential role of motion correction in maintaining quantitative accuracy in dPET. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s40658-026-00843-x.