Computational models of biological systems can provide detailed information that is not possible to obtain experimentally. Such models are often created and analyzed even though the geometry and material properties are not well understood. This approach introduces modeling errors of undetermined magnitude. An alternative approach is to use parametric variation of all model parameters to determine the validity of modeling assumptions and to identify the most and least influential model parameters.

The purpose of this study was to test modeling assumptions commonly applied in phonation modeling and to determine the most influential parameters of the vocal folds. A systematic analysis of the vocal fold structure was performed using finite element models of the vocal fold. Parametric variation of model parameters was a core component of this approach.

Simplifying assumptions commonly applied to vocal fold structural models were examined for their effect on vibratory response. These assumptions included the incompressibility and planar displacement assumptions. The incompressibility assumption was found to introduce only minor changes in modal frequency while the accuracy of planar displacement assumption was found to depend upon the ratio of longitudinal to transverse stiffness. Compatibility conditions for the proper implementation of these assumptions are provided.

The geometric and material parameters of vocal fold models were ranked according to their respective influence on the modal frequencies of the vocal folds. This was accomplished by simultaneous variation of all model parameters based on ranges defined for each parameter. Several different methods of sampling and analysis were used to rank the vocal fold parameters. Sample sizes ranged from 30 to 475. More than 2000 unique finite element models of the vocal folds were analyzed in the course of this study. A set of five dominant parameters were consistently identified as exerting the most influence on modal frequencies of the vocal folds. Deformed-state modal analysis was also performed to determine if these parameters were influential under loaded conditions. The same five parameters were again found to dominate the results of loaded modal analysis. A statistical model based on just seven physical parameters was able to account for over 93% of model response.

Modeling assumptions were examined in detail, and the most influential vocal fold model parameters were identified. Several non-influential parameters were also identified. These results have great import for guiding future experimental studies designed to measure vocal fold tissue properties, and suggest that these parameters should be given careful consideration in future numerical models of phonation.

The general concept of generating large numbers of models via sampling techniques is recommended in future studies involving computational models of biological systems. Statistical analysis may then be performed, providing generalized information on trends, model sensitivities, and the validity of modeling assumptions. This approach is encouraged in future studies involving the computational modeling of biological systems.