Uncertainty quantification and global sensitivity analysis of a patient-specific mandibular finite element model using Latin hypercube sampling and sparse polynomial chaos expansion.

Most mandibular finite element models are deterministic and overlook the influence of biological and modeling variability. This study proposes an automated, uncertainty-aware computational workflow for a patient-specific mandibular FE model using Latin Hypercube Sampling and sparse Polynomial chaos expansion. We investigated 24 input parameters related to masticatory muscle loading, material properties, and geometric discretization over 2400 simulations. Global and regional sensitivity analyses were performed using both Spearman rank correlations and variance-based Sobol indices to identify the primary drivers of von Mises stress. Within this modeling framework, muscle force magnitudes and articular disc stiffness are the dominant sources of stress variability, whereas high-resolution bone property classification had a negligible influence within the explored ranges. Statistical convergence analysis demonstrated that contact-sensitive condylar regions require significantly larger sample sizes (N> 2000) for stable sensitivity estimates compared to global metrics. This methodological benchmark identifies critical modeling parameters and provides a scalable pipeline for future patient-specific biomechanical studies and cohort-level simulations.