Maxillofacial bone movements-aware dual graph convolution approach for postoperative facial appearance prediction

Area of Science:

• Computer-aided surgery
• Medical imaging
• Computational geometry

Background:

• Accurate postoperative facial appearance prediction is crucial for orthognathic surgery planning and patient communication.
• Conventional biomechanical methods are computationally intensive and time-consuming.
• Current deep learning methods have limitations in capturing facial surface details and topology due to independent regional processing.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for predicting postoperative facial appearance.
• To overcome the limitations of existing deep learning methods in perceiving facial surface details and topology.
• To improve the robustness and reduce distortion in postoperative displacement predictions.

Main Methods:

• A novel dual graph convolution model utilizing both Euclidean and geodesic spaces.
• Learning on two graphs constructed from facial meshes to consider surface geometry.
• Transferring bone movements to facial movements in dual spaces.
• Implementing a coarse-to-fine prediction strategy for enhanced robustness.

Main Results:

• The proposed dual graph convolution model demonstrates superior performance compared to state-of-the-art methods.
• Qualitative and quantitative experiments on clinical data validate the model's effectiveness.
• The method successfully addresses limitations in perceiving facial surface details and topology.

Conclusions:

• The dual graph convolution model offers a significant advancement in postoperative facial appearance prediction for orthognathic surgery.
• This approach enhances computational efficiency and predictive accuracy.
• The coarse-to-fine strategy contributes to more robust and less distorted predictions.