Longitudinal analysis of PSMA PET/CT scans requires precise alignment across time points, but registration can distort the quantitative PET signal that is essential for tracking disease. A new preprint describes a registration method designed to preserve this quantification while handling the whole-body, multi-modal nature of the scans.
The approach adapts the diffeomorphic coarse-to-fine LapIRN architecture, a well-established framework for deformable image registration. The method first applies affine pre-registration to roughly align the images, then performs multi-modal deformable registration to refine the alignment. The work is submitted to the Learn2Reg 2026 PSMAReg challenge, which focuses on this specific clinical task.
By keeping the deformation diffeomorphic and integrating multi-modal inputs, the method aims to maintain the integrity of PET uptake values during alignment. This is critical for reliable longitudinal comparisons in prostate cancer imaging, where changes in PSMA expression over time inform treatment decisions. The preprint reports the method's design and its intended use in the challenge, though full results are not detailed in the abstract.