Abstract
Cardiac PET imaging suffers from motion artifacts due to long acquisition times and physiological motion. As a proof of concept, we investigate dense 3D cardiac motion estimation by bi-planar fusion of two orthogonally fine-tuned 2D RAFT (Recurrent All-Pairs Field Transforms) models on a simulated 4D XCAT phantom. We fine-tune a coronal-view model (Y-view) to predict (dx,dz) and a sagittal-view model (X-view) to predict (dy,dz), then deterministically fuse their outputs into a volumetric flow field (dx,dy,dz). On the held-out evaluation pair from the phantom, the Y- and X-view models achieve 0.148 and 0.204 px EPE, respectively, while the fused field achieves 0.178 px EPE (0.27 mm), 96.7% sub-pixel accuracy, and 88.6% directional accuracy. The two dz estimates are correlated (r=0.880) and have similar signed bias (-0.0091 and -0.0085 voxels), and averaging reduces dz MAE to 0.0504 voxels. Relative to the evaluated Farneback and unadapted pretrained-RAFT baselines, fusion reduces 3D EPE by 74% and 66%, respectively. These results support the within-phantom feasibility of parameter-free bi-planar fusion of two 2D RAFT models, while generalization to independent anatomies, realistic PET statistics, and clinical data remains to be established. Cardiac PET Optical Flow Domain Adaptation Motion Correction XCAT Phantom
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/RIME_021.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=gAQHAJ9zAY
BibTex
@InProceedings{CorNub_DomainAdapted_MICCAISAT2026,
author = { Correa, Nubia AND Tan, Ryan AND Zou, Sarah J. AND Chinn, Garry AND Levin, Craig S.},
title = { { Domain-Adapted Bi-Planar RAFT for Dense 3D Cardiac Motion Estimation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17277},
month = {pending},
page = {pending}
}