Correspondence attention for facial appearance simulation.

Area of Science:

• Biomedical Engineering
• Computer Science
• Medical Imaging

Background:

• Accurate simulation of facial changes is crucial for orthognathic surgical planning in patients with jaw deformities.
• Traditional biomechanics-based methods (e.g., Finite Element Method - FEM) are labor-intensive and computationally inefficient.
• Current deep learning methods lack accuracy due to insufficient modeling of the physical relationship between facial soft tissue and bony structures.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for predicting facial soft tissue changes following orthognathic surgery.
• To address the limitations of existing methods by incorporating the physical interplay between bone and soft tissue.
• To improve the computational efficiency of facial change simulation in surgical planning.

Main Methods:

• Proposed an Attentive Correspondence assisted Movement Transformation network (ACMT-Net) for predicting facial changes.
• Utilized a point-to-point attentive correspondence matrix to correlate soft tissue alterations with bony movement.
• Introduced a contrastive loss with k-Nearest Neighbors (k-NN) based clustering for efficient self-supervised pre-training of the ACMT-Net.

Main Results:

• The ACMT-Net demonstrated significantly improved computational efficiency compared to state-of-the-art FEM-based methods.
• Achieved comparable accuracy in predicting facial changes in patients with jaw deformities.
• Validated the model's effectiveness on patient data, highlighting its practical applicability.

Conclusions:

• The ACMT-Net offers a robust and efficient alternative to traditional methods for simulating facial changes in orthognathic surgery.
• The proposed method enhances prediction accuracy by explicitly modeling the soft tissue-bone relationship.
• This deep learning approach holds promise for improving surgical planning and patient outcomes.