Surrogate-Assisted Multi-Objective Optimization of a Case-Specific Mandibular Advancement Device for Obstructive Sleep Apnea Using Integrated FEA and CFD: A Computational Proof of Concept

Abstract

Obstructive sleep apnea (OSA) is commonly treated using mandibular advancement devices (MADs), but designs that improve upperairway patency may also increase appliance loading and alter force transfer to dentoalveolar structures. This study developed a casespecific computational framework for MAD design and multi-objective optimization by integrating parametric computer-aided design,
finite element analysis (FEA), computational fluid dynamics (CFD), design-of-experiments sampling, surrogate modeling, and Pareto
optimization.
A three-dimensional anatomical model was reconstructed from a publicly available cone-beam computed tomography volume. A two-piece
MAD was parameterized using six design variables: mandibular advancement, vertical opening, guiding-block thickness, splint thickness,
sagittal mandibular rotation, and material elastic modulus. A total of 120 geometrically feasible configurations were evaluated using FEA
and CFD. Separate XGBoost and feedforward artificial neural-network surrogate models were trained for the simulated biomechanical and
aerodynamic responses. The validated models for maximum MAD von Mises stress, airway pressure drop, and minimum cross-sectional
airway area were coupled with the non-dominated sorting genetic algorithm II (NSGA-II). Tooth stress, periodontal-ligament strain, tooth
displacement, and MAD deformation were retained as secondary outputs and did not affect Pareto ranking.
The selected surrogate models achieved external-test coefficients of determination ranging from 0.91 to 0.96. The optimization identified
53 unique Pareto-optimal configurations. The selected compromise design comprised a mandibular advancement of 6.4 mm, vertical
opening of 3.2 mm, guiding-block thickness of 4.1 mm, splint thickness of 2.6 mm, sagittal rotation of 10.8°, and material elastic modulus
of 1250 MPa. Relative to a central reference configuration, the compromise design reduced maximum MAD von Mises stress by 19.4%,
airway pressure drop by 16.3%, and 99th-percentile airflow velocity by 6.3%, while increasing the minimum cross-sectional airway area by
8.3%. High-fidelity verification yielded surrogate-prediction errors of 1.8%, 1.9%, and 1.3% for MAD stress, pressure drop, and minimum
cross-sectional airway area, respectively.
The proposed framework enables computationally efficient screening of competing appliance-level structural and aerodynamic objectives
before physical prototyping. However, the findings represent a single-case, anatomy-based computational proof of concept. Because the
selected public CBCT case was not clinically characterized for OSA and lacked polysomnographic and treatment-response data, the results
should not be interpreted as patient-specific predictions of disease severity, therapeutic response, or clinical outcome.