3DDX: Bone Surface Reconstruction from a Single Standard-Geometry Radiograph via Dual-Face Depth Estimation

TL;DR

This work tackles 2D-3D reconstruction from a single X-ray by introducing 3DDX, which jointly estimates front- and back-face depth maps to recover full bone surfaces with absolute scale. It advances multi-depth-map supervision by proposing the center-aligned scale-invariant loss (CASI) and leverages the fixed radiography geometry to reconstruct 3D shapes, followed by statistical-shape-model-based completion. Evaluated on a large clinical dataset (600 patients, 2651 X-ray images), 3DDX substantially reduces surface reconstruction error compared with single-face baselines and yields high-quality 3D bone surfaces, especially when combined with shape completion. The approach demonstrates strong potential for clinical impact in low-dose bone assessment and broad deployment, including settings with limited access to CT or advanced imaging resources.

Abstract

Radiography is widely used in orthopedics for its affordability and low radiation exposure. 3D reconstruction from a single radiograph, so-called 2D-3D reconstruction, offers the possibility of various clinical applications, but achieving clinically viable accuracy and computational efficiency is still an unsolved challenge. Unlike other areas in computer vision, X-ray imaging's unique properties, such as ray penetration and fixed geometry, have not been fully exploited. We propose a novel approach that simultaneously learns multiple depth maps (front- and back-surface of multiple bones) derived from the X-ray image to computed tomography registration. The proposed method not only leverages the fixed geometry characteristic of X-ray imaging but also enhances the precision of the reconstruction of the whole surface. Our study involved 600 CT and 2651 X-ray images (4 to 5 posed X-ray images per patient), demonstrating our method's superiority over traditional approaches with a surface reconstruction error reduction from 4.78 mm to 1.96 mm. This significant accuracy improvement and enhanced computational efficiency suggest our approach's potential for clinical application.

Figures (2)

• Figure 1: Overview of the proposed method. (a) The dual-face depth estimation using depth maps estimation model $G_{d}$ and a bone segmentation model $G_{s}$ to mask the invalid region. (b) 3D surface reconstruction from the estimated depth maps using X-ray geometry to produce initial 3D estimation. (c) 3D shape completion using bone statistical shape model fitting $G_{c}$.
• Figure 2: Visualization of the reconstructed and completed 3D shapes. (a) Comparison between conventional ( 256$\mathcal{L}_{\textit{si}}^{\textit{indep}}$), the proposed ( 256$\mathcal{L}_{\textit{casi}}^{\textit{indep}}$), and the proposed++( 1024$\mathcal{L}_{\textit{casi}}^{\textit{indep}}$†) methods. (b) Visualization of two representative estimated samples (blocked region and slightly diseased bone) using the proposed method. (c) A combined visualization of the target bones.