Original scientific paper

https://doi.org/10.17559/TV-20260327003487

Research on Fracture Surface Search and Reconstruction Strategies for 3D Bone Fragment Stitching

Lei Yin ; College of Computer Science and Technoloy, Changchun University of Science and Technology, Changchun, Jilin, 130021, China
Weiwei Cao ; College of Computer Science and Technoloy, Changchun University of Science and Technology, Changchun, Jilin, 130021, China
Junqi Li ; College of Computer Science and Technoloy, Changchun University of Science and Technology, Changchun, Jilin, 130021, China
Weili Shi ; College of Computer Science and Technoloy, Changchun University of Science and Technology, Changchun, Jilin, 130021, China
Feng Qu ; College of Computer Science and Technoloy, Changchun University of Science and Technology, Changchun, Jilin, 130021, China; Zhongshan Institute of Changchun University of Science and Technology
Miao Yu ; College of Computer Science and Technoloy, Changchun University of Science and Technology, Changchun, Jilin, 130021, China
Zhengang Jiang ; College of Computer Science and Technoloy, Changchun University of Science and Technology, Changchun, Jilin, 130021, China *

* Corresponding author.

Abstract

This study presents an automated three-dimensional (3D) fracture stitching framework for virtual fracture reduction, aiming to improve the accuracy and robustness of preoperative planning. Existing fracture registration methods often rely on Iterative Closest Point (ICP), which is sensitive to inaccurate initialization and prone to local convergence. Moreover, surface noise, fragment defects, and irregular fracture morphologies can cause unstable feature extraction and unreliable matching. To address these challenges, we propose a fracture-surface-aware coarse-to-fine registration framework rather than a direct combination of standard PCA, FPFH, and ICP modules. First, PCA-guided directional filtering and curvature-guided region growing are used to localize fracture-relevant surfaces, reducing interference from non-contact bone regions. Second, FPFH descriptors are embedded into a PCA-constrained coarse registration scheme, providing a more stable initialization than stochastic SAC-IA-based matching. Finally, a Full-Bidirectional ICP strategy is introduced, which retains mutual nearest-neighbor correspondences and minimizes a symmetric bidirectional error to suppress density-driven drift. Experiments on real and simulated fracture datasets demonstrate that the proposed method achieves improved registration accuracy, stability, and efficiency, indicating its potential for virtual fracture reduction and clinical surgical planning.

Keywords

bone fragment stitching; FPFH; fracture-surface extraction; full-bidirectional ICP; point cloud registration; virtual fracture reduction

Hrčak ID:

348714

URI

https://hrcak.srce.hr/348714

Publication date:

30.6.2026.

Visits: 0 *