Parametric Eye Models
Jessica R. Crouch1 and Andrew W. Cherry
Department of Computer Science, OldDominionUniversity, Norfolk, VA23529
Background & Problem:
Patient specific model generationis a research area thatcould improve the utility of medical virtual reality simulators byallowing procedures to be practiced and evaluated in the context of real patients’ anatomies. Research has largely focused on building customized models from medical images. However,eye model generation based on clinical measurements is appropriate becausethe eye is accessible to direct observation and contains small structures that are not well resolved in images. Duringpatient examination, ophthalmologists routinely measure physical parameters of the eye, includinganterior chamber angle, cornea thickness and diameter, globe axial length, and pupil diameter. The goal of this work was to develop a modeling framework and associated software for producing patient-specific volumetric eye models based on a set of clinically measurable parameters.
Methods:
The geometry of onestructure in a multi-part anatomical model cannot be specifiedindependently from its neighbors because complex connections andshape dependencies exist between structures. In the eye example, the iris connects to the choroid just behind where the cornea and sclerameet, so from a modeling standpoint the iris’ position should be dependent on the location of the cornea/sclera boundary. The location of this boundary, in turn, depends on thediameters of the cornea and globe. Therefore the diameter of the cornea impacts the sclera and iris as well as the cornea. Suchdependency relationships motivated the development of a modeling approach that relies on ashape parameter network. The network is an acyclic, directed graph in which input parameters are assigned values by a user and derived parameters compute their values by evaluating functionswhose inputs are other parameters, and optionally, spatially varying object point coordinates. The shape parameters for most of the structures in the eye have a domain that spans an interval of spherical coordinates and a
range that spans an interval of world space (x,y,z)coordinates. Each structure is associated with a parameter
that computes the position of points in its volume, and some structures also have aparameter that provides object based coordinate frames. Each derived parameter has an evaluation function which can be defined using any appropriate equations. For the eye model, the evaluation functions include ellipsoids, Gaussian blending, cubic spline curves, and other non-linear interpolation functions.
Results:
The multi-part eye models generated are space-filling with non-overlapping interior structures, and include fiber orientation vectors for the cornea and certain other structures. The model supports generation of 3D finite element meshes as well as surface meshes [1]. The entire network for the eye model contains approximately 100 parameters; the subset involved in computing cornea shape is shown in Fig. 1. A visualization of some of the structures of a generated eye model is shown in Fig. 2.
Discussion:
The parameter network approach allows a wide variety of models to be generated by adjusting input parameter values. Because the parameter network capturesshape dependency relationships, oneevaluation function can be modified without invalidating theothers.The parameter network approach is extensible, in the sense that it can be employed with any type of solid shape model that defines an object based coordinate system.
References:
1. J Crouch, J Merriam, E Crouch. “Finite Element Model of Cornea Deformation.” Lecture Notes in Computer Science. Sep 2005. 3750: 591 – 598.