Material characterization of human middle ear using machine-learning-based surrogate models

Arash Ebrahimian,Hossein Mohammadi,Nima Maftoon

DOI: https://doi.org/10.1016/j.jmbbm.2024.106478

IF: 4.042

2024-02-28

Journal of the Mechanical Behavior of Biomedical Materials

Abstract:This study aims to introduce a novel non-invasive method for rapid material characterization of middle-ear structures, taking into consideration the invaluable insights provided by the mechanical properties of ear tissues. Valuable insights into various ear pathologies can be gleaned from the mechanical properties of ear tissues, yet conventional techniques for assessing these properties often entail invasive procedures that preclude their use on living patients. In this study, in the first step, we developed machine-learning models of the middle ear to predict its responses with a significantly lower computational cost in comparison to finite-element models. Leveraging findings from prior research, we focused on the most influential model parameters: the Young's modulus and thickness of the tympanic membrane and the Young's modulus of the stapedial annular ligament. The eXtreme Gradient Boosting (XGBoost) method was implemented for creating the machine-learning models. Subsequently, we combined the created machine-learning models with Bayesian optimization (BoTorch) for fast and efficient estimation of the Young's moduli of the tympanic membrane and the stapedial annular ligament. We demonstrate that the resultant surrogate models can fairly represent the vibrational responses of the umbo, stapes footplate, and vibration patterns of the tympanic membrane at most frequencies. Also, our proposed material characterization approach successfully estimated the Young's moduli of the tympanic membrane and stapedial annular ligament (separately and simultaneously) with values of mean absolute percentage error of less than 7%. The remarkable accuracy achieved through the proposed material characterization method underscores its potential for eventual clinical applications of estimating mechanical properties of the middle-ear structures for diagnostic purposes.

engineering, biomedical,materials science, biomaterials