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

Area of Science:

• Biomedical Engineering
• Computational Mechanics
• Medical Diagnostics

Background:

• Assessing middle-ear tissue mechanical properties is crucial for understanding ear pathologies.
• Current methods are often invasive, limiting their use in living patients.
• Accurate material characterization requires understanding parameters like Young's modulus and tissue thickness.

Purpose of the Study:

• To develop a novel, non-invasive method for rapid material characterization of middle-ear structures.
• To create computationally efficient machine-learning models for predicting middle-ear responses.
• To accurately estimate the mechanical properties (Young's moduli) of the tympanic membrane and stapedial annular ligament.

Main Methods:

• Developed machine-learning models (eXtreme Gradient Boosting - XGBoost) for the middle ear.
• Integrated machine-learning models with Bayesian optimization (BoTorch) for efficient parameter estimation.
• Focused on key parameters: Young's modulus and thickness of the tympanic membrane, and Young's modulus of the stapedial annular ligament.

Main Results:

• The developed surrogate models accurately represent vibrational responses of middle-ear structures across various frequencies.
• Successfully estimated Young's moduli for the tympanic membrane and stapedial annular ligament with <7% mean absolute percentage error.
• Demonstrated simultaneous and separate estimation capabilities for tissue mechanical properties.

Conclusions:

• The proposed non-invasive method offers a computationally efficient and accurate approach for middle-ear material characterization.
• The high accuracy suggests significant potential for clinical applications in diagnosing ear pathologies.
• This technique could revolutionize the assessment of middle-ear mechanical properties for diagnostic purposes.