Abstract
Accurate 3D vertebral shape reconstruction from intraoperative X-rays can support surgical navigation without requiring intraoperative CT. However, biplanar projections impose only weak constraints on shape, making the problem ill-posed and existing methods susceptible to anatomically implausible outputs, particularly when dealing with single-view settings or projections containing limited shape information. We present SPVR, a 2D-to-3D reconstruction framework that recovers individual vertebral shapes from posteroanterior and lateral simulated X-rays. In contrast to previous approaches, SPVR incorporates a learned, vertebral-class-conditioned explicit shape prior that regularizes reconstructions toward anatomically plausible shapes, adding particular benefit in outlier cases where image evidence is limited. Incorporating the shape prior here improves the Dice score by 0.027 in the dual-view setting and by 0.070 in the single-view setting. SPVR was evaluated on the public VerSe dataset and on an independent cohort containing pathological cases. Across both dual- and single-view settings, it consistently outperforms four published approaches, achieving ≥ 0.012 higher Dice scores and ≥ 0.227 mm lower 95th-percentile Hausdorff distances compared to the runner-up. For distance-based metrics, the performance gain relative to the runner-up is substantially more pronounced on challenging pathological cases.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_012.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=ZYvOpBD90y
BibTex
@InProceedings{deMar_SPVR_MICCAISAT2026,
author = { de Lange, Marijn AND Wang, Jianing AND Liu, Han AND Gao, Riqiang AND Regensburger, Alois AND Chabin, Guillaume AND Grbic, Sasa},
title = { { SPVR: Explicit Shape-Prior-Guided 3D Vertebral Reconstruction from Orthogonal Projections } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17259},
month = {pending},
page = {pending}
}