Reduced order modelling of air puff test for corneal material characterisation

TL;DR

This paper develops a reduced order model for the air puff test on the cornea, significantly decreasing simulation time and integrating neural networks to enhance accuracy and understanding of fluid-structure interactions.

Contribution

It introduces a fast reduced order model for the air puff test and proposes neural network frameworks to improve model accuracy and interpretability of corneal biomechanical responses.

Findings

Reduced simulation time from 48 hours to 12 minutes.

Neural network accurately models pressure and deformation distributions.

Hybrid models enhance intraocular pressure estimation.

Abstract

Models of the fluid-structure interaction (FSI) model for the air puff test were analysed. Using Abaqus, the air puff test is applied to eyes with varying biomechanical parameters, such as material properties, corneal thickness, and radius. A reduced order model of the air puff (a turbulent impinging jet) has been acquired to decrease simulation time from 48 hours for the FSI model to approximately 12 minutes for the finite element analysis (FEA) model alone. To further accelerate simulations and improve model accuracy, Physics-Informed Neural Networks (PINNs) will be integrated with the reduced-order model. This hybrid approach will help expand the model to a larger dataset, enhancing intraocular pressure (IOP) estimation accuracy and the corneal material properties algorithm through inverse FEA. Additionally, a neural network (NN) framework with embedded Gaussian-modulated waveforms is proposed to model the pressure and deformation distributions on the corneal surface as functions of spatial and temporal parameters. By learning the relationship between corneal biomechanical inputs such as Corneal Central Thickness (CCT), Intraocular Pressure (IOP), and baseline properties (Mu), and the governing coefficients of pressure and deformation, the network accurately reconstructs the result that matches well with the high-fidelity CFD data. This approach can quickly capture the distribution of pressure and deformation. It can also provide insights into the distinct spatial and temporal dynamics of pressure and deformation, giving a more comprehensive understanding of fluid-structure interaction phenomena in the air puff test.