Abstract
2D X-Ray imaging provides real-time guidance during minimally invasive interventions. Nonetheless, it lacks depth and soft-tissue contrast. Clinicians must therefore mentally register it with preoperative 3D Computed Tomography (CT), increasing cognitive overload. LXPose, a recent two-stage CNN framework, automates this 3D/2D reg
istration in real time (20ms) through a single forward pass, bypassing the much slower (7s) state-of-the-art optimization methods. However, its accuracy remains limited by the domain gap between synthetic training projections and real intraoperative X-Rays. Here, we propose to build on this method and present LXPose++, which introduces two targeted improvements: masking the CT to retain only rigid bones, preventing the networks from aligning anatomies that deform intraoperatively; and integrating a Convolutional Block Attention Module (CBAM) to refine feature extraction. On two datasets spanning different organs, these improvements nearly halve the projection error of LXPose with no added inference cost. Crucially, LXPose++ reaches the accuracy of state-of the-art optimization methods, while being 40 times faster, a step toward
online clinical applications. Code and models are availa ble here.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COLAS_009.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=lZ3wkLeq4R
BibTex
@InProceedings{FacFed_LXPose_MICCAISAT2026,
author = { Facente, Federica AND Wei, Wen AND Kattel, Manasi AND Delingette, Hervé AND Ayache, Nicholas AND Berthet-Rayne, Pierre AND Billot, Benjamin},
title = { { LXPose++: an Attention-Based Framework for Robust Real-Time CT to X-Ray Registration } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS pending},
month = {pending},
page = {pending}
}