Corneal digital twins move from simplified shapes to patient-specific simulations

The cornea does more than focus light. Its curved, layered structure must hold its shape against constant internal pressure, and how stiff or compliant that tissue is determines both the sharpness of a person’s vision and the safety of surgery performed on it. For two decades, researchers have turned to finite element modeling, a computational technique that divides a structure into thousands of small elements and calculates how each responds to force, to answer questions that clinical measurements alone cannot. A new review published in Frontiers in Bioengineering and Biotechnology traces how these corneal models evolved from crude idealized shapes into personalized digital replicas built directly from patient scans, and maps where they now influence surgical planning, disease management and device design.

The review, written by a team from Dalian University of Technology and Dalian Third People’s Hospital in China, systematically surveyed the modeling literature. The authors searched Scopus and Web of Science for studies published between January 2007 and May 2026, retrieving 955 documents from Scopus and 184 from Web of Science. After removing duplicates and screening, 243 studies remained, from which 86 papers were selected for detailed discussion based on methodological rigor, clinical relevance and recency. The selected work clusters heavily in the last five years, with 25 articles on refractive surgery procedures such as LASIK, PRK and SMILE, 39 on keratoconus and corneal cross-linking, three on transplantation and seven on orthokeratology lenses.

What has changed most dramatically is the geometry. Early models represented the cornea as rotationally symmetric surfaces derived from idealized spheres or paraboloids, an approach that could only capture average displacement under pressure and ignored the asymmetric transitions at the limbus, the ring where cornea meets sclera. Modern workflows instead extract thousands of spatial data points from clinical imaging, using optical coherence tomography for high axial resolution and Scheimpflug tomography systems such as Pentacam for full anterior-segment topography. Interpolation algorithms like non-uniform rational B-splines then reconstruct refined surfaces of the anterior cornea, posterior cornea and limbus. Iterative optimization now routinely matches clinical data to within 95 percent and keeps displacement errors below 0.001 millimeters, according to studies cited in the review.

But a scanned cornea is not a stress-free cornea. Clinical images capture the tissue in its pressurized equilibrium state under intraocular pressure, typically between 10 and 21 millimeters of mercury. Loading that deformed geometry directly into a simulation stacks geometric stiffness on top of material stiffness, artificially stiffening the model. The review highlights quantitative work showing this error inflates predicted stress by roughly 10 to 15 percent across 130 patient-specific models of healthy, keratoconic and post-LASIK corneas. The fix is the inverse finite element method: an algorithm iteratively back-calculates the clinically measured shape to the zero-pressure configuration, initializing the stress field so the simulated preoperative state physically matches what the patient’s images actually show.

Material descriptions have undergone a parallel transformation. Linear elastic models, which assume stiffness is constant regardless of deformation, cannot reproduce the cornea’s characteristic J-shaped stress-strain curve, in which the tissue is compliant at low strain and stiffens dramatically at high strain. Hyperelastic formulations such as Ogden and Mooney-Rivlin capture that strain-stiffening behavior. Because the corneal stroma is built from collagen lamellae arranged in orthogonal patterns, with circumferential fibers concentrated at the limbus, anisotropic fiber-reinforced models based on the Holzapfel-Gasser-Ogden framework have become the most widely adopted approach. Viscoelastic terms add time-dependent relaxation and creep, which matter for interpreting high-speed air-puff measurements and long-term outcomes after cross-linking treatment.

One striking illustration of why derivation method matters comes from a recent study the review cites: when the same anisotropic material model was fitted to uniaxial tensile tests versus inflation tests, the fiber nonlinearity parameter differed by nearly an order of magnitude, roughly 24 versus 223, even though both fits achieved correlation coefficients of 0.99. The review’s authors compiled a reference table of reported parameter ranges for linear elastic, hyperelastic, anisotropic and microstructural models, documenting whether each value came from tensile testing, inflation testing, biaxial testing or inverse finite element calibration, precisely so modelers do not transfer parameters across incompatible experimental contexts.

The clinical payoff is most visible in refractive surgery. In patient-specific simulations of LASIK performed under physiological intraocular pressure of 15 millimeters of mercury, each additional 50 micrometers of residual corneal thickness was associated with approximately a 12 percent reduction in postoperative peak von Mises stress, a relationship that was strongly linear. Comparative finite element studies summarized in the review found that PRK, which ablates the surface without creating a flap, raised postoperative peak stress by about 28 percent over baseline, while LASIK raised it by about 53 percent. SMILE, which removes a lenticule of tissue through a small incision while preserving the stiff anterior cap that bears an estimated 40 to 50 percent of the mechanical load, showed lower displacement and stress than LASIK under identical conditions. One patient-specific study incorporating individualized material properties and intraocular pressure for the first time reported a mean postoperative dioptric error of plus 0.40 plus or minus 0.30 diopters, while finding that SMILE induced the largest increases in residual stromal bed stress, 68.2 percent, and strain, 48.8 percent, with PRK the least invasive of the three.

Keratoconus, a progressive thinning disorder affecting roughly 1 in 2,000 people, has driven much of the multiscale modeling work. Advanced models now represent the cornea as a discrete network of collagen fibrils and cross-links, then upscale microstructural force balance to continuum equations, allowing researchers to simulate how cross-link rupture and fiber disorganization produce the cone-shaped protrusion characteristic of the disease. Simulations indicate the regional stiffness of a keratoconic cone is reduced by 41 to 60 percent relative to healthy tissue. The fiber dispersion parameter, which quantifies how randomly collagen fibers are oriented, stratifies cleanly with disease severity in experimentally calibrated models, rising from 0.05 to 0.15 in healthy corneas to above 0.35 in advanced cases. For corneal cross-linking treatment, modeled deepening of the treatment from 200 to 300 micrometers enhanced corneal flattening by 24 to 34 percent and reduced higher-order aberrations by 22 to 31 percent in one patient-specific simulation, and a customized light-dose protocol was predicted to flatten the cornea more than a standard protocol, with predicted changes correlating moderately with clinical measurements.

The modeling toolkit has also reached transplantation and orthokeratology. Finite element studies of penetrating keratoplasty showed that sutures generate extreme asymmetric stress at the limbus, a mechanism linked to postoperative irregular astigmatism. Simulations of descemet stripping endothelial keratoplasty identified air-injection pressures that hold the graft in place without damaging endothelial cells. For orthokeratology lenses, which reshape the cornea overnight to correct myopia, fluid-structure interaction models revealed that the reverse-geometry lens creates central positive pressure and mid-peripheral negative suction through tear-film hydrodynamics, driving epithelial migration and stromal reshaping. Coupled oxygen-diffusion and viscoelastic models predict diurnal thickness fluctuations and cumulative stiffness changes after long-term lens wear.

Significant bottlenecks remain before these models become routine clinical tools. Most existing simulations of air-puff tonometry treat the cornea as a pure solid mechanics problem, ignoring the damping effect of the aqueous humor inside the eye, which causes pure structural models to overestimate corneal stiffness and diverge from clinical deformation measurements. High-precision patient-specific simulations involving anisotropic materials and complex surgical procedures can take hours to days of computation, far too slow for surgical navigation. Artificial intelligence is emerging as the workaround: machine learning models now predict air-puff pressure distributions with correlation above 0.98 against computational fluid dynamics, and a neural network approach identifies anisotropic material parameters from clinical deformation data in under one second with a mean error of 3.5 percent, replacing hundreds of inverse iterations. The review’s authors argue these accelerated pipelines are what will carry corneal digital twins from specialized research settings into everyday ophthalmic practice, supporting individualized surgical design, refined cross-linking parameters and improved graft assessment.

Subject of Research: Finite element modeling of corneal biomechanics for surgical planning and disease management

Article Title: A critical review of the progress and application of finite element models in corneal biomechanics

Article References: Zhang, S., Lan, Y., Wang, B., Cheng, W., Lyu, Y., Zhang, L., & Jin, L. (2026). A critical review of the progress and application of finite element models in corneal biomechanics. Frontiers in Bioengineering and Biotechnology, 14, Article 1850437. https://doi.org/10.3389/fbioe.2026.1850437

Image Credits: AI Generated

DOI: 10.3389/fbioe.2026.1850437

Keywords: corneal biomechanics, finite element modeling, refractive surgery, keratoconus, corneal cross-linking, LASIK, SMILE, digital twin, corneal transplantation, orthokeratology, inverse finite element method, machine learning