Background : Respiratory masks are essential for protection against airborne contaminants in both medical and industrial environments. However, prolonged use often leads to discomfort and pressure sores. This issue, typically caused by poor fit, compromises user comfort, protection efficiency, and compliance with safety guidelines.
Objective : This study aims to predict facial deformation and pressure distribution during mask usage, based on a limited set of annotated biomechanical data, to ensure optimal fit.
Method : We developed a semi-supervised graph neural network that models facial geometries as graph structures. The proposed framework utilizes 45 labeled and 120 unlabeled facial datasets, employing a variational graph autoencoder constrained by Hertzian contact theory. It also integrates an XGBoost-based module for deformation zone classification.
Results : Our approach achieves a deformation RMSE of 0.164 mm (R2 = 0.9896) and a pressure RMSE of 0.0492 kPa (R2 = 0.9517), representing a 34.27% improvement over Random Forest, 20.62% over PointNet++, and 10.01% over TPSNET. Five-fold cross-validation confirms robust generalization with minimal overfitting and inference time under 2 seconds.
Conclusion : This study presents a real-time personalized respiratory mask fitting model capable of accurately predicting facial deformation and contact pressure. The approach demonstrates strong generalization from limited labeled data, enhancing comfort, safety, and compliance in medical and industrial applications.
| Date | 22 Dec 2025 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Luc Duong (Supervisor) & Bahe Hachem (Co-supervisor) |
|---|
Mlika, E. (Author),
Duong (Supervisor) & Hachem (Co-supervisor),
22 Dec 2025Student thesis: Master's thesis › Master in Engineering: Information Technology Engineering