Abstract
We present X2V, a method for recovering deformable respiratory motion from a single intraoperative X-ray in previously unseen patients, without per-patient models or optimization. Existing single-view methods are either patient-specific or, when trained across patients, constrain motion with a linear subspace or direct regression. X2V constrains the deformation to a respiratory motion manifold learned from population 4D-CT and parameterized in the space of stationary velocity fields, whose exponential yields smooth, invertible deformations. At inference, a lightweight encoder maps the input X-ray onto the learned motion manifold in a single forward pass, without the target volume or test-time optimization. Under leave-one-case-out evaluation on the DIR-Lab COPD cohort, X2V recovers 16.25 mm of respiratory motion down to a residual 11.12 mm from a single projection, improving on the four-view LiftReg baseline by 22% while producing fold-free deformations, where LiftReg folds on every case. The same manifold, reused unchanged, transfers to a second cohort with a different acquisition protocol, where X2V again improves on four-view LiftReg (4.70 vs 6.23 mm).
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/TIA_020.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{ManSid_X2V_MICCAISAT2026,
author = { Mangipudi, Siddharth AND Akash, Naren AND Narayanan, P. J.},
title = { { X2V: Deformable 2D/3D Registration from Intraoperative X-ray to Volume via Motion Manifold Learning } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17269},
month = {pending},
page = {pending}
}