Physical parameter estimation from porcine ex vivo vocal fold dynamics in an inverse problem framework

Abstract

This study presents a framework for a direct comparison of experimental vocal fold dynamics data to a numerical two-mass-model (2MM) by solving the corresponding inverse problem of which parameters lead to similar model behavior. The introduced 2MM features improvements such as a variable stiffness and a modified collision force. A set of physiologically sensible degrees of freedom is presented, and three optimization algorithms are compared on synthetic vocal fold trajectories. Finally, a total of 288 high-speed video recordings of six excised porcine larynges were optimized to validate the proposed framework. Particular focus lay on the subglottal pressure, as the experimental subglottal pressure is directly comparable to the model subglottal pressure. Fundamental frequency, amplitude and objective function values were also investigated. The employed 2MM is able to replicate the behavior of the porcine vocal folds very well. The model trajectories’ fundamental frequency matches the one of the experimental trajectories in 98.6 % of the recordings. The relative error of the model trajectory amplitudes is on average 9.5 %. The experiments feature a mean subglottal pressure of 10.16 (SD = 2.31) cmH 2O ; in the model, it was on average 7.61 (SD = 2.40) cmH 2O. A tendency of the model to underestimate the subglottal pressure is found, but the model is capable of inferring trends in the subglottal pressure. The average absolute error between the subglottal pressure in the model and the experiment is 2.90 (SD = 1.80) cmH 2O or 27.5 %. A detailed analysis of the factors affecting the accuracy in matching the subglottal pressure is presented.