ONH-PINN: A Physics-Informed Neural Network for the Discovery of In Vivo Biomechanical Properties of the Optic Nerve Head

Abstract

Purpose : (1) To develop ONH-PINN (Optic Nerve Head Physics-Informed Neural Network), an efficient and user-friendly solution to extract in vivo biomechanical properties of optic nerve head (ONH) tissues in a clinical setting. (2) To verify its numerical accuracy with synthetic intraocular pressure (IOP)-induced ONH deformations.

Methods : Our approach is tailored to extract in vivo biomechanical properties of ONH connective tissues, specifically those of the peripapillary sclera (PPS) and of the lamina cribrosa (LC), based on full-field IOP-induced deformations. These deformations can be obtained from in vivo biomechanical tests during optical coherence tomography imaging of the ONH (Girard et al., J R Soc Interface. 2013; 10:20130459). To verify this method, we generated synthetic IOP-induced ONH deformations with predefined biomechanical properties for the PPS and LC using the finite element method (FEM). Both tissues were modeled as nearly incompressible hyperelastic materials, each characterized by their respective shear stiffness parameters (c_PPS and c_LC). In contrast to the gold standard for biomechanical properties discovery (i.e. inverse FEM), ONH-PINN eliminates the need for creating a meshed representation of the ONH morphology. Instead, a randomly sampled set of points (i.e. 3D point cloud) suffices as input. By using the synthetic IOP-induced ONH deformations as training data, along with the governing equations of structural mechanics (leveraging our knowledge of ONH biomechanics), we determined the unknown material properties, c_PPS and c_LC (see Figure 1).

Results : From the given synthetic ONH deformations, ONH-PINN accurately rediscovered the biomechanical properties of ONH connective tissues, showcasing a high precision with a percentage error of 0.4% for c_PPS and 2.1% for c_LC, respectively.

Conclusions : ONH-PINN has the capability to derive in vivo biomechanical properties of ONH tissues from deformations induced by IOP, acquired through in vivo biomechanical tests. This method, distinguished by its user-friendly nature, holds promise for advancing clinical assessments of ONH biomechanics, which could be useful for the management of glaucoma.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

Figure 1. Schematic of ONH-PINN. Biomechanical properties of ONH connective tissues are recovered from the ONH morphology, synthetic training data, and knowledge of ONH biomechanics.