Automatic Multi-View X-Ray/CT Registration Using Bone Substructure Contours

TL;DR

The paper tackles robust intraoperative X-ray/CT registration for orthopedic procedures by introducing a fully automated, multi-view approach that leverages semantic contours of femur substructures to reduce ICP ambiguity. It combines automatic segmentation of the diaphysis and condyles using nnU-Net with Skeleton Loss and a contour-based ICP that optimizes 2D–3D correspondences across three substructures, enhanced with residual-based reweighting. The authors provide a public dataset of five cadaveric femurs with CT and X-ray poses, and demonstrate sub-millimeter accuracy ($<1$ mm) on real data, dramatically outperforming a commercial, manually initialized method and showing strong robustness to initialization and contour noise. The method’s practical impact lies in its automation and accuracy, enabling easier integration with tracking systems for improved intraoperative navigation in CAS.

Abstract

Purpose: Accurate intraoperative X-ray/CT registration is essential for surgical navigation in orthopedic procedures. However, existing methods struggle with consistently achieving sub-millimeter accuracy, robustness under broad initial pose estimates or need manual key-point annotations. This work aims to address these challenges by proposing a novel multi-view X-ray/CT registration method for intraoperative bone registration. Methods: The proposed registration method consists of a multi-view, contour-based iterative closest point (ICP) optimization. Unlike previous methods, which attempt to match bone contours across the entire silhouette in both imaging modalities, we focus on matching specific subcategories of contours corresponding to bone substructures. This leads to reduced ambiguity in the ICP matches, resulting in a more robust and accurate registration solution. This approach requires only two X-ray images and operates fully automatically. Additionally, we contribute a dataset of 5 cadaveric specimens, including real X-ray images, X-ray image poses and the corresponding CT scans. Results: The proposed registration method is evaluated on real X-ray images using mean reprojection error (mRPD). The method consistently achieves sub-millimeter accuracy with a mRPD 0.67mm compared to 5.35mm by a commercial solution requiring manual intervention. Furthermore, the method offers improved practical applicability, being fully automatic. Conclusion: Our method offers a practical, accurate, and efficient solution for multi-view X-ray/CT registration in orthopedic surgeries, which can be easily combined with tracking systems. By improving registration accuracy and minimizing manual intervention, it enhances intraoperative navigation, contributing to more accurate and effective surgical outcomes in computer-assisted surgery (CAS).

Figures (18)

• Figure 1: Overview of the proposed registration pipeline. X-ray image poses are computed using a fiducial marker. Contours of the diaphysis (red), medial and lateral condyles (green and yellow resp.) are automatically extracted in the images. The registration is performed using a contour-based ICP with matches across the same classes of substructures
• Figure 2: Example of a training sample. (a): DRR. (b): contour labels of the diaphysis (red), medial and lateral condyles (green and yellow resp.). (c): overlay of both
• Figure 3: Example of bone contour predictions. Our model accurately segments the occluding contours of the bone substructures while distinguishing the medial condyle from the lateral condyle, even under challenging viewpoints, as shown in image (c). Some images with strong noise can lead to false predictions as seen in image (d)
• Figure 4: Evaluation of different registration methods on our femur dataset via mRPD. A method successfully registered an X-ray image if the final mRPD was less than one millimeter (red line)
• Figure 5: Robustness to random initialization of Zhang's method without substructures (top row) and the proposed method with substructures (bottom row). The abscissae represent initial parameter values, with each graph showing one sampled pose parameter. Each registration corresponds to six crosses with the same ordinate (mRPD) across graphs, representing the six initial pose dimensions