Fully Automated AI-Based Digital Workflow for Mirroring of Healthy and Defective Craniofacial Models

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Biomedical Engineering

Background:

• Accurate craniofacial reconstruction relies on precise segmentation and mirroring of anatomical structures.
• Current manual methods are time-consuming and prone to operator variability, impacting surgical planning.
• Developing automated solutions is crucial for improving efficiency and reproducibility in craniofacial defect repair.

Purpose of the Study:

• To develop and validate a fully automated digital pipeline for craniofacial reconstruction.
• To integrate deep learning-based segmentation with algorithmic mirroring for enhanced precision.
• To establish a standardized, time-efficient workflow for surgical planning.

Main Methods:

• Utilized 388 cranial CT scans to train a 3D nnU-Net model for skull and mandible segmentation.
• Employed a Principal Component Analysis-Iterative Closest Point (PCA-ICP) algorithm for symmetry plane computation and mirroring.
• Quantified accuracy using Dice Similarity Coefficient (DSC), Mean Surface Distance (MSD), Hausdorff Distance (HD), and angular deviation compared to expert references.

Main Results:

• The nnU-Net model achieved high segmentation accuracy (mandible DSC: 0.956, skull DSC: 0.965).
• Automated mirroring demonstrated minimal angular deviation from expert planes (mandible: 1.32° ± 0.71°, skull: 1.75° ± 0.84° in defect cases).
• The workflow showed robust performance and accuracy, unaffected by the presence of defects.

Conclusions:

• The developed automated digital pipeline offers a standardized, reproducible, and time-efficient solution for craniofacial reconstruction.
• This approach enhances precision in segmentation and mirroring, improving surgical planning.
• The validated workflow shows significant clinical applicability for treating craniofacial defects.