Fully automated AI-based digital workflow for orbital fracture mirroring: A multicenter approach

Area of Science:

• Medical imaging
• Artificial intelligence in medicine
• Surgical planning

Background:

• Orbital reconstruction demands precise anatomical restoration for function and aesthetics.
• Current virtual surgical planning relies on manual, operator-dependent segmentation and mirroring.

Purpose of the Study:

• To develop and validate a fully automated, AI-based workflow for orbital segmentation and mirroring.
• To improve efficiency and reproducibility in orbital reconstruction surgery.

Main Methods:

• Trained a 3D nnU-Net model on 502 multi-institutional CT scans for automated orbital segmentation.
• Employed Principal Component Analysis and Iterative Closest Point registration for algorithmic mirroring.
• Benchmarked automated results against manual segmentation by clinical experts using Dice Similarity Coefficient, Mean Surface Distance, and Hausdorff Distance.

Main Results:

• AI-driven segmentation achieved high accuracy (mean Dice Similarity Coefficient of 0.936), particularly in non-fractured orbits (up to 0.941).
• Automated mirroring demonstrated minimal angular deviation from manual planes (fractured: 0.95° ± 0.65°; non-fractured: 0.72° ± 0.53°).
• Consistent performance across international datasets (Germany, Romania, USA) indicates robustness.

Conclusions:

• The automated AI workflow provides robust and clinically applicable segmentation and mirroring for orbital reconstruction.
• This technology enables standardized, efficient, and reproducible surgical planning.
• Potential to significantly improve functional and aesthetic outcomes in patients undergoing orbital reconstruction.