Patient-specific reference model estimation for orthognathic surgical planning.

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Biomedical Engineering

Background:

• Accurate reference bony shape models are essential for effective orthognathic surgical planning.
• Current methods for deriving these models have limitations, including potential distortions and compromised precision due to overlooking nonlinear relationships.

Purpose of the Study:

• To develop and validate a novel self-supervised learning framework for estimating patient-specific reference bony shape models.
• To address the limitations of existing methods in accurately capturing the intricate nonlinear relationships in craniofacial structures.

Main Methods:

• A self-supervised learning framework utilizing a deep query network was developed.
• The network estimates similarity scores between patient midface and normal subject data in a high-dimensional space.
• High-dimensional features are aggregated and projected back into 3D structures to generate a patient-specific reference model.

Main Results:

• The framework was trained on 51 normal subjects and tested on 30 patients.
• Performance evaluation showed a mean Chamfer distance error of 2.25 mm and an average surface distance error of 2.30 mm.
• The method demonstrated superior accuracy in reference model estimation compared to existing approaches.

Conclusions:

• The proposed method effectively leverages correlations in a high-dimensional space to generate accurate patient-specific reference models.
• Both qualitative and quantitative analyses confirm the superiority of this approach over current state-of-the-art methods.
• This framework offers a significant advancement for orthognathic surgical planning.