Characterization and Initial Optimization of Reconstruction Performance for Detector Crosstalk of Ultrahigh Performance Brain PET Imager.

The NeuroEXPLORER (NX) is a next-generation, ultrahigh-performance brain positron emission tomography/computed tomography (PET/CT) system designed for both high spatial resolution and sensitivity. To meet high-sensitivity (HS) requirements, the system uses 2-cm-long scintillation crystals, while the novel U-shaped depth-of-interaction (DOI) encoding and small crystal elements enable high resolution. However, using long, small crystals increases signal crosstalk between crystals arising from Compton scattering, photoelectron escape, and incomplete optical isolation, which can degrade spatial resolution. Although the NX electronics detect crosstalk between microblocks and support real-time recovery using measured energy, accurate identification of the first interaction point is not always possible. Incorporating crosstalk events increases coincidence sensitivity by approximately 50% but compromises spatial resolution. Given such a large fraction of detector crosstalk, evaluating its impact on reconstruction performance in the NX is imperative. In this work, we evaluated image resolution in three reconstruction modes: (HS; all events), higher resolution (HR; noncrosstalk events), and lower resolution (LR; crosstalk events), using real phantoms and human studies with multiple tracers. Guided by the resolution characterization of the HYDRA-OLG brain phantom, we introduced a two-Gaussian mixture point spread function (2G-PSF) for HS reconstruction to improve contrast while suppressing noise. At matched iterations, HR yielded the highest image contrast, HS showed slightly reduced contrast, and LR exhibited the largest partial volume effect, as expected. Modeling crosstalk with the 2G-PSF for HS increased contrast toward HR levels without sacrificing sensitivity. Overall, these results provide a practical strategy for including crosstalk events in high-resolution PET systems while mitigating their impact on spatial resolution.