Orbit 2™ ProjectAuthor: JM Miller Modified: Support: StatusSeeking support through CDMRP Key CollaboratorsOther Labs[Your related project here?] |
Eye movements are critical for rapid and coordinated direction of the visual apparatus. Limitations of movement, and defects in binocular coordination severely impair vision. If the eyes are not aligned, a patient will have diplopia (see double) or will suppress vision in one of the eyes, reducing the field of view. If eyes do not move properly, visual information gathering is impaired. No less important than their function as sense organs, the eyes are the most important part of the face for social communication. Classic experiments show that observers of a face repeatedly examine the eyes. Eye movements send clear messages about a person's attention. Making and breaking eye contact is important in conversational turn taking. Gaze evokes emotion, causing arousal and a sense of immediacy in gaze recipients. A person who avoids the direct gaze of others may be seen as embarrassed, sad, disgusted, or deceptive.
Strabismus, misalignment of the eyes or limitation of eye movement, is prevalent in the US and usually treated surgically, with imperfect results. Battlefield trauma, often involving major damage to the orbits, challenges the reconstructive skills of surgeons, and frustrates software developed to aid less dramatic eye muscle disorders.
Several computer models of eye movement mechanics have been developed, but only Millers Orbit™ 1.8 Gaze Mechanics Simulation has achieved wide acceptance by scientists and clinicians. Orbit 1.8 has been useful in laboratory studies of eye movement control, and in the clinic, where it is used for diagnosing and planning treatments for complex strabismus disorders. Unfortunately, Orbit was written in an old, hard-coded style, which makes it prohibitively difficult to update as required to reflect new research findings and to apply to new classes of disorders.
But the solution is not to create another eye mechanics model to replace Orbit 1.8 – it would suffer from the same basic limitations – but to create a specialized Biomechanics Modeling Environment (BME), in which one could create, maintain and operate models, altering and updating them with relative ease. The goal of this project is to create an such a specialized BME, and to build within it Orbit 2, a new model of extraocular biomechanics. Programmed mostly in Java, Orbit 2 would run on any modern hardware platform or via a Web browser. With its ability to edit and create new models, BME should be general enough to accommodate developments in our understanding of eye movement biomechanics and control, and to aid in the diagnosis and treatment of complex traumatic and other disorders.
Figure 1: Scan paths. Shown this picture of a Russian girl, observers look mostly at the eyes (from A L Yarbus, 1965, Eye Movements and Vision). |
Eye movements are critical for rapid and coordinated direction of the visual apparatus. Limitations of movement, and defects in binocular coordination severely impair vision. No less important than their function as sense organs, the eyes are the most important part of the face for social communication. Classic experiments show that observers of a face (even in a photograph; see Figure 1) repeatedly examine the eyes. Eye movements send clear messages about a person's attention (Breed, 1972, Kelly, 1978, Loftus, Loftus & Messo, 1987), which reveals what he or she currently considers behaviorally relevant (Rutter, 1984), which in turn is a powerful clue to intent. Making and breaking eye contact is important in conversational turn taking (Argyle, 1988). The dynamics of eye movement, such as staring versus glancing, also conveys information. Gaze evokes emotion, causing arousal and a sense of immediacy in gaze recipients (Andersen, Guerrero, Buller & Jorgensen, 1998). A person who avoids the direct gaze of others is seen as embarrassed, sad, or disgusted (eg: Burgoon, Coker & Coker, 1987), and as deceptive (eg, Bond et al 1992).
The life of a victim of facial trauma may be saved, and acceptable cosmesis may be achieved, but without reconstruction of natural eye movement, the quality of that life in interaction with others will be compromised.
The prevalence of strabismus, misalignment of the eyes or limitation of eye movement, is 3-5% (Ingram, Walker, Wilson, Arnold & Dally, 1986), the vast majority of which is simple esotropia (crossed eyes), exotropia (wall eyes) or isolated muscle weaknesses. Despite the simplicity of such cases relative to the orbital injuries that can result from explosion or projectile impact, only 50-80% of cases of esotropia (Keenan & Willshaw, 1992, Lipton & Willshaw, 1995) and 69% of cases of exotropia (Keech, Scott & Christensen, 1987) are corrected with a single operation. Even re-operations have disappointing success (Harley, 1980, Kittleman & Mazow, 1986). There is a nationally recognized need for innovation in diagnosis and treatment of strabismus (NEAC, 1999).
 |
|
Figure 2: Knapps 1861 Ophthalmotrope. |
Since eye movements were first studied, students have recognized the need to model their mechanics. The ophthalmotropes of Ruete (1845), Wundt (1862), and others are clever and beautiful (see, eg, Figure 2), but extraocular modeling did not gain serious scientific and clinical utility prior to the seminal work of Robinson (1975). Their scientific utility has been striking. Early attempts by Miller & Robinson to compute binocular alignment from first principles made it obvious that the paths of EOMs, gaze dependence of these paths, and the mechanical role of supporting orbital tissues, were of critical importance in determining EOM actions, and were also totally unknown. We developed orbital MRI methods to obtaining these data, and these methods have now found clinical utility for determining eye muscle strength and contractility. Our MRI studies showed surprising stability of normal rectus EOM paths as the eyes moved about, and this finding led quickly to the discovery of EOM pulleys, which have drastically altered our notions of how the brain controls the eyes.
By model we mean a hybrid homeomorphic biomechanical model. Parts of a homeomorphic model, and interactions among those parts, have the same form (functional & geometric) as corresponding parts and interactions in the modeled system. Creating natural correspondences between model and physiology first facilitates development and validation of submodels by independent specialists, and subsequently, comprehension and use of the model by medically trained personnel. A homeomorphic model can treat arbitrary new situations, if they can be expressed in the models terms. Successful predictions validate the model. Unsuccessful predictions tend to specify the physiologic research needed to improve the model.
A model able to accurately reflect a wide range of mechanical and innervational injuries and repairs cannot be built from homogeneous, generalized elements, without foundering on convergence failures and requiring impractical computer power. If a complex model is to be accurate, stable and practical it must be able to simulate significant functional and cosmetic subtleties without squandering resources on aspects that are unimportant. It must be a hybrid model, each part of which is a submodel of a type suited to the particular subsystem and its important features. The model must be appropriate, at every level, to the corresponding physiology. A hybrid model of the orbit might be composed of B-Splines or Finite Elements for bone, Musculoskeletal Strands (Pai, 2002) for muscles and connective tissue bands, Disembodied Calculations (black boxes) for vestibular and brain systems, and possibly other model types. It follows that a workable biomechanical model could not result from first building an overall geometric model, and later adding material properties.
Biomechanical models can be dynamic, in the sense of calculating effects that depend on time derivatives of position (eg, effects of inertia & viscosity), or static, in the sense of calculating only positional equilibria (static force balance). Dynamic models provide more information & realism, and much theory has been developed to support such models. However, in a practical model, the computational cost of dynamic calculations must be balanced against their utility.
A Biomechanical Modeling Environment (BME) is necessary to support communication among submodels, and editing and customization of varying submodel types. In a binocular orbital model, communication will need to be mechanical and electrical, and both local (eg, direct mechanical coupling) and remote (eg, innervation and functional electrical stimulation). Editing must support parametric (eg, stiffnesses), functional (eg, the type of nonlinearity) and topologic (the arrangement & connectivity of submodels) alterations to the model, recognizing that such a model will be altered and refined at all levels throughout its useful life. Special fitting capabilities can also be provided to conveniently adjust biomechanical parameters (but not functional forms or topology), geometry and surface properties, subject to continuity constraints, to fit the model to individual cases.
It is essential that a model intended for wide use over a period of time that justifies the effort of creating it be created and maintained within a specialized BME. Hard coded models (such as Orbit 1.8) quickly encounter situations that would seem to lie in their domains, but which they cannot handle (eg, certain complex eye muscle surgeries). Advances in the theory and data underlying a model (eg, active pulleys) may make it obsolete. Hard-coded models must be substantially rewritten in these situations (Orbit™ went through several versions, 3 of which – v1.0, v1.5, and v1.8 – were widely distributed), and eventually the code tends to become unmaintainable. By shifting focus away from the models themselves, to an environment for creating, maintaining, and using models, it becomes possible to create models that are flexible enough to be responsive to changing requirements.
Within the proposed BME, we plan to create an advanced eye model that will reflect some intraocular biomechanics (related to the near response and its linkage to vergence, and essentially all that is known about extraocular biomechanics, along with suitable characterizations of brainstem innervation. The first model of extraocular biomechanics (Robinson, 1975) became possible when the innervation-length-tension surface of extraocular muscle was measured in humans (Robinson, O'Meara, Scott & Collins, 1969). Thirty years later, despite considerable efforts, there does not exist a 3-dimensional homeomorphic dynamic eye model, such as the one we propose to create (Quaia & Optican, 2003).
Prior to the development of computational models of extraocular biomechanics (Miller & Robinson, 1984, Robinson, 1975), it seemed sufficient to describe orbital anatomy in terms of origins, insertions, and cross sections of EOMs, measured in cadavers. But attempts to calculate binocular alignment from first principles made it obvious that the paths of EOMs, gaze dependence of these paths, and the mechanical role of supporting orbital tissues, were of critical importance in determining EOM actions. For example, computational simulations of EOM transposition surgery failed because the models assumed that EOM bellies could sideslip in the orbit to follow transposed insertions (Miller, 1988). Early radiographic studies in monkeys (Miller & Robins, 1987) and humans (Simonsz, Harting, de Waal & Verbeeten, 1985) then suggested that the bellies of rectus EOMs have paths that are stable relative to the orbit despite changes in gaze. Miller employed magnetic resonance imaging (MRI) to demonstrate stability of normal rectus EOM belly paths throughout the oculomotor range (Miller, 1989). Only the anterior rectus tendon paths moved relative to the orbit. These studies also revealed that during contraction EOM cross-sectional area increases, with the point of maximum cross section shifting toward the EOM's origin. This showed how non-invasive MRI could be used to demonstrate the contractile state of EOMs.
 |
| Figure 3: Orbit 1.8™ Gaze Mechanics Simulation on a laptop computer, with Ruetes (1857) ophthalmotrope in the background. Screen shows windows for setting eye parameters, a graphical representation of the resulting eyes, and a binocular alignment chart comparing simulated and measured eye positions. |
Several computer models of eye movement mechanics have been developed (egs: Clement, 1985, Robinson, 1975, Simonsz & Spekreijse, 1996), but only the Orbit™ 1.8 Gaze Mechanics Simulation (Miller, Pavlovski & Shamaeva, 1999; see Figure 3) has achieved wide acceptance (approximately 100 licensees in 12 countries) by scientists and clinicians in roughly equal numbers. A recently developed model, called SEE++ includes a partial implementation of Orbit 1.8. Orbit includes submodels of muscle force (including elastic and contractile force components), muscle paths (including muscle length), Herings Law, which determines innervations to an eye moving passively or under cover, Sherringtons Law of reciprocal innervation, which relates innervations of paired antagonistic muscles.
There has been considerable development over the past 20 years in understanding the basic mechanical arrangements by which the extraocular muscles (EOMs) rotate the eyes. During the preceding 20 years, eye movement measurements had been used to infer brainstem mechanisms, based on classical assumptions about extraocular mechanics, so that much of that work has now to be reconsidered. General lessons have also been learned: [1] Fundamental revisions can be required in our understanding of apparently simple and heavily studied systems when they are looked at with new tools. [2] A strong biomechanical model will reveal gaps and errors in our physiological understanding, and will, unless it is a very flexible model, quickly render itself obsolete.
Advances in battlefield medicine and body armor have dramatically reduced lethality of combat injuries, but have produced a growing population of war veterans with disconcerting eye and facial injuries, who will have difficulty returning to civilian society without reconstructive procedures and prosthetics that lie beyond the state of the art. Complete facial rehabilitation that achieves animation of regrown or transplanted tissues and continuity with native tissues requires fundamental work in many disciplines and lies well in the future. In contrast, restoration of coordinated eye movement would restore to normality a central feature of the face, and can be effectively addressed in a relatively short time with comparatively modest effort.
In cases where vision remains, restoration of normally coordinated eye movements would improve basic functioning by improving the ability to scan the environment for visual information, restoring single vision, and enlarging the visual field. Even if vision is lost and the eyeball is replaced with a (non-seeing) prosthetic, the communicative functions of gaze would be restored. The result would be an improvement in the social functioning of the patient, and in the comfort of those he or she interacts with.
Ongoing work to repair or regrow hard and soft facial tissues would be synergistic with our proposed modeling effort, which will calculate optimal attachments of muscles and connective tissues for achieving normal eye movements, even if reconstructed orbits only approximated normality. Extraocular muscles (EOMs) have unique contractile and other properties (Porter & Baker, 1996), not duplicated by muscles that might be harvested to replace them. Artificial contractile elements are similarly unlikely, at least initially, to replicate the properties of EOMs. In both cases, intuitions of eye muscle surgeons developed from experience making small adjustments to EOMs would not be sufficient to achieve good outcomes. The biomechanical model we propose to develop will be able to do so. Thus, it would be expected to encourage development of these replacement technologies.
Experience with Orbit 1.8 (see above) has taught us that a hard-coded model of high complexity is difficult to maintain, and soon becomes obsolete. Here, we propose to focus on the environment (BME) and its resources (MPs and data-derived eye parameter sets), with which we will develop, maintain, and apply new ocular models. The proposed modeling environment will be useful for developing and maintaining biomechanical models more advanced than the one we will actually construct under the proposed project (eg: a model that can calculate orbital bone fractures resulting from blunt globe trauma), and models of other body systems (eg: hands), but we will not pursue those applications here.
Musculoskeletal Strands: Dr Pai is developing a new primitive for neuromuscular modeling, which does not suffer from the limited realism of lines-of-force, or the computational burden of finite elements. This new primitive, the ``Strand element, efficiently represents the behavior of thin elastic solids using the theory of Cosserat rods (Pai, 2002; Rubin, 2000). Under the proposed project, we will develop the theory of Strands to implement activation-contraction properties suitable for dynamic modeling of extraocular muscles (EOMs). Strand templates will be developed so that, eg, muscle elements can be warped graphically and with respect to their contractile and elastic properties, to fit empirical data, such as those from MRI. We will develop new, more efficient implementations of Strands so that it will be possible to represent an EOM with multiple Strands, each corresponding to a muscle layer (Demer, 2002) or compartment. Future extensions of these methods might support EOM modeling at the level of motor units to simulate the complex neuromuscular interactions that appear to characterize the oculomotor system (Miller et al., 2002).
Eyeball Model: A particularly important military application that our software will support is evaluation and treatment blunt impact injuries. One common injury is the "orbital blowout fracture", in which the eye is hit by a object and transmits impact force sideways (by means of the globes elasticity and hydrostatic pressure) to the fragile bones of the nasal wall and floor of the orbit, which consequently explode. We will use simple models of globe deformation, based on pre-computed elastostatic Green's functions (James & Pai, 1999) or modal deformation models (James and Pai 2002) to prepare our environment for future modeling of these phenomena.
Orbital Bones: Extraocular muscles couple to rigid orbital bones, having complex shapes. We will represent bone using B-Spline Surface Patches, and develop algorithms for fitting these general surfaces to orbital MRI data and immunohistochemically stained tissue sections. (Our BME would allow this simple bone model to be replaced with one that could simulate fracture).
Contact Handling: We will develop new methods for detecting and handling contact among Musculoskeletal Strands representing muscles and connective tissue bands, an elastically deformable globe, and B-Spline surfaces representing orbital walls. This is expected to be a difficult problem to solve.
Our first set of tasks will be to develop the Parametric Interface, the associated Member Export function, and the Execution System, using a known, fixed simulation (Orbit 1.8 model code) to drive the system. The result will be a minimal system that produces known results.
The useful, but not essential, automated Parameter Fitter will then be developed, if time allows. The Parameter Fitter would automatically set model parameters to optimize a chosen criterion, eg, to minimize the squared differences between simulated eye positions and positions measured in the clinic (see accompanying Orbit 1.8 Gaze Mechanics Simulation Users Manual).
 |
 |
|
Figure 5: Lancaster Red-Green test is commonly used to measure binocular alignment in the clinic. Above: Colored filters allow the fixing eye to see only the red target, which is positioned by the examiner, and the following eye to see only the green target, which is moved by the patient so it appears to lie on top of the red target. With nothing else for the eyes to align on, they relax to exhibit their basic misalignments. Right: Scripted simulation of this test begins (A) with the position and description of the fixing eye, computes the innervations required to hold it on the red target (B), the innervations, which according to Herings law, are provided by the brain to the following eye (C), and finally, the position of the following eye. Each ellipse in the flowchart corresponds to an execution of a monocular model. The sequence implements a pair of eyes performing a specific task. |
We will then prepare the system for the new Extraocular Biomechanical Model and the Biomechanical Construction Kit (BMCK) by implementing Command Export and Model Containers, and extending the Runtime Environment to support multiple models. As part of this task, we will characterize properties common to all of the types of biomechanical model we will support, particularly regarding communication among models. In the case of a binocular alignment simulation (see Figure 5), eg, the brain model sends innervations to the eye model (mirroring the physiology), and the eye model sends eye positions to the brain model (using the eye model to calculate the brains efference copy, that is, the transform of its innervations into eye position space).
Next, we will create the code interfaces and general libraries of the Biomechanical Framework, and install the new Modeling Primitives, testing by creating a new simulation equivalent to that of Orbit 1.8. It will be useful that the Orbit 1.8 core code will have already been integrated into BME, making it possible to run automated tests to verify the correctness of the Framework.
Finally, we will develop BMCK. This is the final phase of BME development, because only at this point will we have identified the elements of the Model Framework that must be supported by BMCK. If necessary, we will postpone development of BMCK, since it will be possible, if burdensome, to create and test models without it.
A 0th-level prototype Eye Model, incorporating new Model Primitives, will be created midway through the project for testing the new BME. Although this prototype will intentionally represent no physiology beyond that in Orbit 1.8, it will already offer new capabilities, including dynamic simulation and enhanced graphical interaction, based on Strand capabilities.
We will fit the eyeball and bone surfaced developed above to fit orbital MRI and histochemical slice data.
Once satisfied that we can move beyond Orbit 1.8, we will prototype a new Ocular Biomechanics Model incorporating Coordinated Active Pulleys, Weak Differential Active Pulleys, and a more realistic representation of extraocular connective tissues, including such smooth muscle and elastin structures as the Inferomedial Peribulbar muscle (Miller et al., 2003).