Related Experiment Video
Updated: May 27, 2025
08:18
Three-Dimensional Reconstruction of Orbital Fractures
Published on: May 16, 2025
108
3D Deep Learning for Virtual Orbital Defect Reconstruction: A Precise and Automated Approach.
Fangfang Yu1, Chang Liu1, Chenglan Zhong2
1State Key Laboratory of Oral Diseases & National Center for Stomatology & National Clinical Research Center for Oral Diseases & Department of Oral and Maxillofacial Surgery, West China Hospital of Stomatology.
The Journal of Craniofacial Surgery
|February 17, 2025
Summary
A new 3D U-Net+++ model precisely reconstructs orbital defects, offering a fast and automated solution for both unilateral and bilateral cases. This AI-driven approach significantly improves virtual surgical planning for complex orbital fractures.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Computer-Aided Surgery
Background:
- Accurate virtual orbital reconstruction is vital for preoperative planning.
- Traditional methods like mirroring are inadequate for complex orbital defects and are inefficient.
- Existing techniques struggle with defects crossing the midline.
Purpose of the Study:
- To introduce a modified 3D U-Net+++ architecture for precise and automated orbital defect reconstruction.
- To overcome limitations of traditional methods in handling complex and bilateral orbital defects.
- To enhance the accuracy and efficiency of virtual surgical planning.
Main Methods:
- Developed and trained a modified 3D U-Net+++ model on 300 synthetic orbital defects from CT scans.
- Validated the model on 15 clinical cases of orbital fractures.
- Evaluated performance using quantitative metrics (HD95, ASSD, Surface DSC, PSNR, SSIM) and surgeon assessments (Likert scale).
Main Results:
- Achieved high accuracy on synthetic data (HD95 < 2.0 mm, Surface DSC > 0.94) and outperformed other networks.
- Clinical validation showed excellent surgeon ratings (>4/5) for structural integrity, edge consistency, and morphology.
- Demonstrated high precision for clinical defects (HD95 ~ 2.5 mm, Surface DSC > 0.91) with rapid processing (~10 seconds/case).
Conclusions:
- The modified 3D U-Net+++ provides a precise and highly automated solution for orbital defect reconstruction.
- This method is particularly effective for challenging bilateral and trans-midline orbital defects.
- The approach promises significant improvements in clinical practice and preoperative planning for orbital surgery.
