Artificial Intelligence-Based Modeling Can Predict Face Shape Based on Underlying Craniomaxillofacial Bone.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• 3D Reconstruction

Background:

• Craniomaxillofacial reconstruction is complex due to 3D anatomy.
• Bilateral injuries lack mirroring techniques, necessitating pre-injury data.
• AI shows promise in facial feature identification and simulation.

Purpose of the Study:

• To hypothesize and test if AI can predict 3D face shape from underlying bone anatomy.
• To develop an AI workflow for subject-specific facial shape estimation.
• To assess the utility of AI-predicted facial geometry in surgical planning and forensics.

Main Methods:

• Developed an automated image processing and AI workflow using a modified 3D UNet.
• Trained models on 5 publicly available computed tomography (CT) imaging datasets.
• Incorporated bone geometry and metadata (BMI, age) for prediction.

Main Results:

• AI models provided visually reasonable predictions of facial tissue contour and geometry.
• Achieved a validation Dice score of 0.89 on combined datasets.
• Attained a Dice score of 0.925 when trained on the HNSCC dataset.

Conclusions:

• AI-driven 3D face shape prediction is feasible using bone geometry.
• This method can assist in preoperative craniomaxillofacial surgical planning.
• Automated face shape prediction has potential applications in forensic science.