Abstract
Purpose :
To present a stochastic model of the corneal and ocular biometry in keratoconic eyes, based on previous work. This model is capable of generating an unlimited number of random, but realistic biometry sets, including the corneal elevation, intraocular distances and wavefronts, with the same statistical and epidemiological properties as the original keratoconic data it is based on.
Methods :
The data of 145 keratoconic eyes of 145 patients (aged 18 – 60 years) was recorded with an autorefractometer, Scheimpflug imaging (Oculus Pentacam), optical biometer (Haag–Streit Lenstar) and an aberrometer (Tracey iTrace), which lead to a set of 97 biometric parameters. In order to reduce this number to 18 parameters, the Zernike coefficients of corneal elevation were compressed using Principal Component Analysis. These data were subsequently fitted with a linear combination of multivariate Gaussians through an Expectation Maximization algorithm, from which it is possible to generate an unlimited number of random biometry sets with the same distributions as the original data. These biometry sets can then be used to calculate the associated wavefronts and other ocular parameters. Equality between the original keratoconic data and the synthetic data was assessed using "two one-sided" tests.
Results :
In order to verify the accuracy of the wavefront calculations, the wavefronts derived from the measured biometry were compared to the originally measured wavefronts and found significantly equal (two one-sided t test, p < 0.05). Next the biometry of 1000 synthetic eyes were generated by the stochastic model, followed by ray tracing to obtain the associated wavefronts. These synthetic data were found significantly equal to the originally measured data (two one-sided t test, p < 0.05), thus making them statistically indistinguishable.
Conclusions :
To the best of our knowledge this is the first eye model dedicated specifically to keratoconus. It produces synthetic biometry data of eyes that is indistinguishable from actual measurements. This model may be interesting for visual optics researchers that do not have access to actual biometry data.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.