Non-Intrusive Reduced Order Modeling of Patient-Specific Cochlear Implantations

Area of Science:

• Biomedical Engineering
• Computational Science
• Medical Simulation

Background:

• Cochlear implants are vital for treating severe to profound hearing loss.
• Patient-specific simulations offer insights for surgical planning and post-operative analysis.
• Current simulations face significant computational challenges, limiting their clinical application.

Purpose of the Study:

• To develop a computationally efficient method for patient-specific cochlear implant simulations.
• To reduce the time required for generating and simulating patient-specific models.
• To enable rapid prediction of intracochlear voltage distribution for clinical use.

Main Methods:

• A non-intrusive reduced-order model (ROM) was developed using proper orthogonal decomposition and radial basis function interpolation.
• The ROM was trained on a dataset of 528 snapshots from 24 distinct cochlear models with varying electrical stimuli.
• The model utilizes five parameters: three for cochlear geometry and two for electrode array position and active electrode.

Main Results:

• The reduced-order model drastically decreased simulation time from approximately 1.5 hours to under one second.
• Accuracy was validated using a leave-one-out strategy, achieving a high precision.
• The relative error compared to finite element solutions was found to be 2.5%.

Conclusions:

• The non-intrusive reduced-order model accurately predicts 3D intracochlear voltage distribution for new patient geometries and implant positions.
• This demonstrates the feasibility of fast, patient-specific numerical simulations for cochlear implants.
• These simulations can support cochlear implant fitting, individualized sound coding strategies, and surgical decision-making.