Eigeneyes: A statistical model of whole-eye geometry and its applications

Modeling the three-dimensional geometry of the optical surfaces of the human eye has important applications, including the study of myopia and presbyopia, as well as the planning of cataract surgery. In this paper, we present the Eigeneyes, a statistical shape model of the complete three-dimensional geometry of the human eye based on principal component analysis (PCA). The model was constructed from 572 Optical Coherence Tomography eye scans in which the cornea, crystalline lens, and retina were visible. Eigeneyes captures geometric variability across ocular structures using a small number of coefficients that represent global modes of variation. Eigeneyes achieved a statistically significantly lower reconstruction error than other state-of-the-art methods while using fewer parameters. Furthermore, the Eigeneye coefficients were modeled within a multivariate probabilistic framework, enabling the generation of realistic synthetic eyes conditioned on age and/or axial length.