PET Quantification-Preserving Registration for Longitudinal Whole-Body PSMA PET/CT

Fryderyk Kögl; Anna Reithmeir; Daniel M. Lang; Emily Chan; Rickmer Braren; Julia A. Schnabel

cs.CVOct 15, 2026

PET Quantification-Preserving Registration for Longitudinal Whole-Body PSMA PET/CT

Authors: Fryderyk Kögl, Anna Reithmeir, Daniel M. Lang, Emily Chan, Rickmer Braren, Julia A. Schnabel

Organizations: Helmholtz Munich · Technical University of Munich · Munich Center for Machine Learning · University Medical Center Hamburg-Eppendorf ·

Abstract

We adapt the diffeomorphic coarse-to-fine LapIRN architecture for longitudinal whole-body PSMA PET/CT registration in the Learn2Reg 2026 PSMAReg challenge. The method combines affine pre-registration, multi-modal deformable registration, and test-time instance optimization. We further introduce PET biomarker preservation losses for metabolic tumor volume (MTV) and total lesion glycolysis (TLG), including Jacobian-based constraints, and a per-structure rigidity regularization for skeletal structures. The submitted configuration achieved 81.0% DSC and 6.17 mm HD95, with MTV and TLG errors of 3.3% and 3.3%, respectively, on our evaluation protocol. Code is available at https://github.com/koegl/psmareg.

Figures & tables

Figure 1: Overview of the proposed registration pipeline. The moving (follow-up) and fixed (baseline) PET/CT scans are first coarsely aligned by an affine pre-registration, and the resulting pair is then passed to the coarse-to-fine LapIRN network, which predicts a dense deformation field (green grid). Finally, after the LapIRN stage, we refine this field per pair by instance optimization (IO) and resample the moving scan with it into the fixed space (warped PET/CT). Both stages are driven by the same objective, a weighted sum of three groups of terms: registration accuracy, PET quantification, and deformation regularization (Sec. 2.3 ).

Panel of Figure 2 — Figure 2: Registration accuracy and biomarker preservation peak at different checkpoints. Validation CT Dice (left) continues to improve after MTV and TLG bias (right) reach their minima. The red dashed line indicates the checkpoint selected using the challenge composite ranking score.

Method	DSC (%) $\uparrow$	HD95 (mm) $\downarrow$	MTV error (%) $\downarrow$	TLG error (%) $\downarrow$	NDV (ppm) $\downarrow$	Time (s) $\downarrow$
Initial	14.8 ∗	30.9 ∗	–	–	–	–
Affine	52.9 ∗	11.7 ∗	5.47	4.70	–	19.1 $\pm$ 0.920
ConvexAdam	63.0 ∗	8.41 ∗	5.10	4.15	100	22.8 $\pm$ 1.06
NiftyReg	73.3 ∗	7.09 ∗	4.46	8.68 ∗	1525 ∗	49.7 $\pm$ 0.480
Ours (no IO)	76.5	6.65	3.22	5.08	0.00	19.5 $\pm$ 1.02
Ours (container)	81.0	6.17	3.29	3.31	864	89.8 $\pm$ 2.69

Table 1: Our method against the baselines on the official challenge validation set (

n=20

cases), evaluated against the surrogate labels of Sec. 3.1 . Every cell is a mean over the cases, runtime included: DSC and the MTV/TLG errors in %, NDV in ppm, and time in seconds per image pair on the hardware of Sec. 2.7 . Bold marks the best value in each column. ∗ marks metrics on which ours (container) is significantly better than the baseline in that row (paired Wilcoxon signed-rank, Holm-corrected within each metric,

p<0.05

); runtime is not tested. Dashes mark metrics undefined for a baseline.

Medical Image Registration

Medical image registration aligns scans across time, patients, or imaging modalities to track anatomy and support diagnosis, treatment, and image-guided procedures.

See all 96 papers

Registration is a Representation Problem: Solving Whole-Body PET/CT Registration
Oct 13, 2026Neel Dey, Adrian V. DalcaMedical Image Registration Deformable Image Registration
A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital · Department of Radiology, Harvard Medical School · Computer Science and AI Lab, Massachusetts Institute of Technology
CT-Guided Spatially-varying Regularization for Voxel-Wise Deformable Whole-Body PET Registration
Apr 24, 2026Xiangcen Wu, Ruohua Chen, Sichun Li +4Medical Image Registration PET/CT Imaging
Institute of Medical Technology, Peking University Health Science Center, Beijing, China · Ren Ji Hospital, Shanghai Jiao Tong University, Shanghai, China · University of Oxford, Oxford, United Kingdom
Is Deformable Image Registration Ready for Brain Metastasis Reirradiation Dose Accumulation? A Longitudinal MRI Benchmark of Registration Accuracy
Aug 27, 2026Hengjie Liu, Manju Sharma, Xinyi Fu +2Deformable Image Registration Medical Image Benchmarks
Department of Radiation Oncology, University of California, San Francisco, San Francisco, CA, USA

Oct 13, 2026J/K move · Enter open · S save

Registration is a Representation Problem: Solving Whole-Body PET/CT Registration

Neel Dey, Adrian V. Dalca

A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital · Department of Radiology, Harvard Medical School · Computer Science and AI Lab, Massachusetts Institute of Technology

Medical Image Registration Deformable Image Registration Longitudinal Medical Image Analysis

Open paper AskPDF arXiv

Raw image intensities are a poor surrogate for similarity across images in correspondence problems. Further, while auxiliary segmentation labels can provide semantic identification and improve matching, they can assign only coarse identities to fine-grained anatomy and often cover only a subset of structures, particularly in whole-body settings. We present an approach that instead establishes correspondence using complementary representations of anatomy and demonstrate its utility for longitudinal whole-body PET/CT registration in the Learn2Reg 2026 PSMAReg Challenge. We use our 3D foundation model, anatomix, to derive fine-grained, per-voxel semantic representations that are consistent across volumes and further employ off-the-shelf segmentation models to obtain coarse anatomical identities. Together, these representations drive a GPU-accelerated iterative registration solver (FireANTs) in a hybrid multistage framework: the first stage estimates alignment from dense multi-channel features, after which a second refines the deformation using both anatomix- and segmentation-based losses. Lastly, because registration should not introduce non-biological tumor volume changes across time, we develop lesion-specific, component-wise volume preservation schemes. Without any task-specific training, our system placed first on the PSMAReg Challenge leaderboard. Code and interactive visualizations are available at https://www.neeldey.com/reps-for-reg.

Abstract

Medical Image Registration

Explore similar work

Registration is a Representation Problem: Solving Whole-Body PET/CT Registration

CT-Guided Spatially-varying Regularization for Voxel-Wise Deformable Whole-Body PET Registration

Is Deformable Image Registration Ready for Brain Metastasis Reirradiation Dose Accumulation? A Longitudinal MRI Benchmark of Registration Accuracy