Multi-Quantifying Maxillofacial Traits via a Demographic Parity-Based AI Model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oral and Maxillofacial Surgery

Background:

• Accurate quantification of maxillofacial traits is crucial for surgical planning and outcomes.
• Existing artificial intelligence (AI) models struggle with demographic disparities in data, limiting their generalization.
• Maxillary alveolar basal bone quantification involves multiple indices vital for aesthetic and functional assessments.

Purpose of the Study:

• To develop a demographic parity strategy for AI-based multi-quantification of maxillofacial indices.
• To enhance the generalization ability of AI models for analyzing cone-beam computed tomography (CBCT) images.
• To address performance disparities in AI models related to sensitive attributes like sex.

Main Methods:

• Collected 4,000 CBCT sagittal images of maxillary alveolar basal bone.
• Developed a deep learning model with shared parameters and multiple regression heads.
• Identified sex as a sensitive attribute, subdivided data, trained submodels, and ensembled them for final generalization.

Main Results:

• Initial AI model showed underperformance in quantifying major basal bone indices.
• The ensembled model, trained on male and female submodels, achieved equal performance across genders.
• The final model demonstrated low error, high consistency, and strong correlation, similar to clinicians but faster.

Conclusions:

• Demographic parity strategy significantly improves AI model generalization for maxillofacial trait quantification.
• This approach is effective even for highly variable traits, benefiting appearance-conscious maxillofacial surgery.
• The developed AI model offers a reliable and efficient tool for clinical application in maxillofacial surgery.