A Parameterization Framework for Image-Based Resolution Modeling of Spatially-Variant PSF Deformation in Dedicated...

Spatially-variant (SV) point spread function (PSF) deformation in limited-angle PET systems can significantly degrade lesion localization, shape fidelity, and quantitative accuracy, particularly in dual-panel breast PET (BPET) scanners. While image-based resolution modeling (IRM) has been shown to improve image quality, scanner implementation of SV-IRM remains a major challenge, as it is not practical to measure hundreds of PSFs, as used in previous simulation studies. In this work, we propose a parameterization framework to characterize SV PSF deformation using a limited number of measured PSFs. Each PSF kernel is modeled using a Gaussian mixture model (GMM), and the spatial variation of the GMM parameters is described using polynomial fitting across the field of view. This approach enables the generation of a dense PSF lookup table from a modest number of measurements, allowing practical implementation of SV-IRM in a real BPET system. The proposed method was evaluated using multiple physical phantoms, including hot-rod, single-lesion, and NEMA NU 4 phantoms. Results demonstrate that SV-IRM improves lesion localization, restores elongated structures, and achieves higher contrast recovery compared to both no IRM and spatially-invariant IRM approaches. A moderate increase in background noise was observed, reflecting the inherent trade-off between resolution recovery and noise amplification. This work provides a practical strategy for incorporating spatially-variant resolution modeling into limited-angle PET systems and demonstrates its potential to improve quantitative imaging performance in dedicated breast PET applications.