6-DoF dental pose estimation for AR-assisted craniofacial surgery.

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Accurate patient-to-virtual anatomy registration is crucial for augmented reality (AR) in craniofacial surgery.
• Traditional marker-based methods are limited in dynamic surgical settings, and soft tissue landmarks lack reliability.
• A novel markerless approach is needed for robust 6D skull pose estimation.

Purpose of the Study:

• To develop and validate a markerless framework for precise 6D skull pose estimation in craniofacial surgery using upper teeth as rigid landmarks.
• To enable accurate augmented reality (AR) integration for improved surgical navigation.
• To overcome the limitations of traditional registration methods.

Main Methods:

• A pipeline utilizing a fine-tuned Segment Anything Model 2 (SAM 2) for segmenting teeth from monocular images.
• Training a pose estimation model to predict 6D skull pose directly from segmented tooth masks.
• Implementing a patient-specific training strategy using synthetic masks derived from intraoral scans for rapid subject adaptation.

Main Results:

• The proposed framework was evaluated on a dataset of 159 images from eight healthy subjects.
• High performance was demonstrated across various 6D pose estimation metrics, confirming the framework's effectiveness.
• The method successfully validated the use of upper teeth as reliable rigid landmarks for skull pose.

Conclusions:

• Leveraging patient-specific synthetic data eliminates the need for extensive real-world annotations, reducing development time and cost.
• The approach prevents model overfitting, leading to a more robust and generalizable solution for surgical navigation.
• This non-invasive technique offers a promising alternative for AR-assisted craniofacial surgery.