Remarkable progress has been made in understanding central oculomotor and visuomotor processes by assuming that the motoneurons and everything peripheral to them constituted a simple, slavish “oculomotor plant” (Keller & Robinson, 1971, 1972; Robinson, 1981; Skavenski & Robinson, 1973). This assumption also impeded progress in some respects because it led investigators to look only to central processes to understand the complex rotational kinematics exhibited by the eye moving in its various modes. Plant articulations began to receive more attention when Miller and colleagues first proposed the modern notion of extraocular muscle pulleys (Miller, 1989), and obtained the first functional evidence of their existence (Miller, Demer, & Rosenbaum, 1993). Studies followed, describing the anatomy, histology and ultrastructure of pulley tissues (Clark, Miller, & Demer, 1997; Clark, Miller, & Demer, 2000; Demer, 2002; Demer, Miller, Poukens, Vinters, & Glasgow, 1995; Kono, Poukens, & Demer, 2002), the innervation of SM contained in these tissues (Demer, Poukens, Miller, & Micevych, 1997), kinematic consequences (Miller, Pavlovski, & Shamaeva, 1999; Quaia & Optican, 1998; Raphan, 1998), and clinical implications (Clark, Isenberg, Rosenbaum, & Demer, 1999; Clark, Miller, & Demer, 1998; Clark, Miller, Rosenbaum, & Demer, 1998; Demer, Miller, & Poukens, 1996) of these long-neglected extraocular structures.

The discovery of EOM pulleys was surprising: how could physiologically significant gross anatomic structures have gone unnoticed in a part of the body as well studied as the orbit? The answer is that components of the pulleys had been noticed before, by Tenon (1806), Baudens (see Koornneef, 1977b), Sappey (1888), and others, although these fragmentary descriptions, supported by only ad-hoc functional conjectures, were largely forgotten and did not lead to the modern concept of connective tissue pulleys that stabilize posterior EOM paths relative to the orbital wall (Miller, 1989).

Even in the absence of predictions from modeling (Miller, 1999; Miller et al., 1999), pulleys would not have escaped attention if they were discrete, sharply delineated structures. However, they are distributed structures, characterized only by relative condensations of SM, elastin, and collagen, arrayed over the three-dimensional orbital volume. We wondered, therefore, if there might be other distributed extraocular connective tissue structures, still awaiting description.

Several prominent architectural features of extraocular connective tissues are classically recognized: (1) Tenon’s capsule, which consists of a thin fibrous membrane covering the globe from the limbus to the optic nerve, reflected sleeves that surround the EOMs as they penetrate the capsule, and various connections among sleeves and from sleeves to the orbital walls; (2) the superior transverse ligament of Whitnall, which is a thickening of the sheath of the levator palpebrae superioris (ie, the upper eyelid lifting) muscle that extends laterally and medially to insert in the orbital walls just behind the superior orbital rim; (3) Lockwood’s inferior ligament, which is a 5 to 8 mm wide thickening of Tenon’s sleeves between where the inferior rectus (IR) and inferior oblique (IO) muscles cross, that extends to insert laterally and medially into the canthal tendons and orbital walls. Lockwood’s ligament is sometimes said to send lateral and medial fibers up to Whitnall’s ligament, so that the two ligaments and their extensions form a ring in the anterior orbit (Dutton & Waldrop, 1994; Fink, 1948; Koornneef, 1977a; Warwick, 1976). These tissues are usually assigned the general role of supporting and protecting the eyeball (eg, von Noorden, 1990), which is to say that their specific functions are largely unknown.

Leo Koornneef initiated the study of connective tissue biomechanics when he observed that localized entrapment of a muscle in a “blowout” fracture of the orbital floor caused generalized disturbances of ocular motility, suspected that all the EOMs were tied together in complex ways, and set about describing extraocular connective tissues in detail. He used thick (60 - 140 µm) sections to preserve spatial relationships in the slice plane (Koornneef, 1974), and very thick (5 mm) sections, to visualize in each section architectural features perpendicular to the slice plane (Koornneef, 1977a). He presented stacks of sections, schematic drawings and artist’s renderings, delineating EOM sheaths, musculo-orbital connections and intermuscular connections, but perhaps his most important contribution was to show that orbital connective tissue structures are highly stereotyped across human subjects, strongly suggesting that the matrix of extraocular connective tissue is evolutionarily conserved because of its specific biomechanical functions (Koornneef, 1991). Recently, Dutton & Waldrop (1994) published studies similar to Koornneef’s using thick (150 µm) sections, manual tracing of sectional features and artist’s renderings.

However, there have been few systematic attempts to characterize distributed extraocular structures in terms of histologic composition or mechanical properties. Traditional histologic methods are inadequate because slices thick enough to resist mechanical distortion by the microtome knife and subsequent processing are too thick to be uniformly penetrated by histochemicals. Modern tomographic methods fall far short of the spatial resolution and chemical discrimination needed to distinguish varieties of connective tissue. Biomechanical properties of orbital tissues, measured in identified sub-structures and small tissue samples (Collins, O'Meara, & Scott, 1975; Collins, Scott, & O'Meara, 1969; Stager, 1996) are essential to quantitative modeling, but such studies are unlikely to find new distributed structures: cutting apart soft, elastic structures distorts geometric relationships in ways that may make it impossible to infer mechanical functions.

An imaging approach would be suitable to characterize the overall, or architectural, features of the extraocular space, but unfortunately imaging cannot directly give mechanical properties. However, imaging can give strong hints about mechanical properties, and direction for subsequent mechanical measurements, by showing the types and densities of constituent tissues. As with the pulleys, we expect connective tissue functions to be disclosed by patterns of histologic specialization (Demer et al., 1995). Accordingly, we have developed methods for producing three-dimensional (3D) images of extraocular tissues based on histologic data.

Collagen is a major supportive protein of skin, tendon, bone, cartilage and connective tissue, and is the predominant component of extraocular connective tissue. Existing descriptions of extraocular tissue architecture relevant to oculomotility are mainly descriptions of condensations of collagen. It would be particularly interesting, however, to characterize the distributions of the more specialized types of connective tissue: elastin and SM. Elastin, a microscopic non-cellular fibril, has the remarkable property of resisting straightening from, and promptly returning to, its crumpled resting configuration. Smooth, or nonstriated, muscle acts under control of the autonomic nervous system to alter the length and stiffness of tissues of which it is a component. Koornneef (1977c), for example, conjectured that tension in the connective tissue system might be influenced by distributed extraocular SM cells, altering its role in the performance of eye movements.

The classical description of human extraocular SM divides it into two main parts (Duke-Elder & Wybar, 1961; Kestenbaum, 1963; Warwick, 1976). The peribulbar part, called the capsulopalpebral muscle of Hesser, is said to be an incomplete (absent on the lateral side) ring of SM in the anterior orbit that includes the superior palpebral muscle of Müller, which arises from the inferior or global aspect of the levator muscle and inserts in the upper eyelid, the inferior palpebral muscle of Müller, which arises from the inferior or orbital aspect of the IR where it crosses the IO and extends anteriorly to insert in the lower eyelid, and some fibers extending from around the lateral rectus (LR) and medial rectus (MR) muscles to the orbital walls (“check ligaments”). But the medial part of the capsulopalpebral muscle is described as “feeble”, and the inferior portion as “feeble, especially in its lateral part” (Warwick, 1976), suggesting that peribulbar SM is concentrated in the superior orbit. The second part of the orbital SM, called the orbital muscle of Müller, spans the inferior orbital fissure, and is thought to be vestigial in humans. Fibers are sometimes found to extend from the orbital muscle into the fascial suspensions of the LR and IR (Dutton & Waldrop, 1994).

Interest in EOM pulleys has motivated several microscopic studies of connective tissues surrounding the EOMs at the level of the globe equator. Histochemical studies found abundant SM and elastin in the orbital aspect of the MR pulley, as had been previously noted by Koornneef (1977b), and to a lesser extent in the LR and IR pulleys (Demer et al., 1995; Demer et al., 1997), and a prominent crescent of SM at the level of the globe equator, extending from the nasal border of the superior rectus muscle (SR) pulley, passing through the orbital aspect of the MR pulley, and terminating on the lateral border of the IR pulley (Demer, Oh, & Poukens, 2000; Kono et al., 2002). Lockwood (1886) noted “a lot of elastic connective tissue coupling the LR and MR to the globe and orbit”. A recent microscopic study of the fine structure of the orbital aspect of the human MR pulley revealed dense bands of collagen fibers alternating at right angles to each other interspersed with elastin fibrils and discrete bundles of SM inserting in the periorbita (Porter, Poukens, Baker, & Demer, 1996). SM and elastin unrelated to extraocular mechanics are found in the walls of arteries and large veins. Blood vessels are easily identified by their lumens under high magnification.

Tissue imaging methods can be ordered from low distortion methods which, like MRI, provide a physiologically realistic view, albeit with low spatial resolution and poor tissue differentiation, through those which, like thick slices, provide moderate resolution, differentiation and distortions, to high resolution methods which, like stained thin sections, provide high spatial resolution and exquisite tissue differentiation, at the cost of significant spatial distortions. Instead of abandoning the thin slice histologic approach, as did Koornneef, or abandoning computer reconstruction, as did Dutton & Waldrop, we used computer-aided image processing to compensate for distortions in the slice plane caused by histologic processing, and 3D reconstruction techniques to restore the spatial relationships perpendicular to the slice plane. We correct the distortions in high-resolution slices by warping them to fit correlated, low distortion slices, guided by features that are visible in every slice.

In the present work, we distinguish several types of tissue (striated muscle, SM, collagen, and elastin) as well as familiar anatomic structures (EOMs, globe, optic nerve, blood vessels and nerves). Our results, presented as manipulable 3D objects, support and extend the results of Kono et al (2002), who used more conventional methods to study these tissues.

We used magnetic resonance images (MRI) of whole, and digital images of serially sectioned human cadaveric orbits. Specimens were identified only by code, and were accompanied only by aspects of the donor’s medical history that bore on orbital structure and histology, such as race, age, diseases with extraocular manifestations, and ocular surgeries.

Cadaveric materials were obtained from UCLA Medical Center and a tissue bank (IIAM, Scranton, PA). Study of cadaveric specimens was conducted in compliance with state and local law.

We report here results from 2 orbits: (1) Specimen “H5” was harvested at autopsy 20 hrs after death from complications of cardiac transplantation of a 44-year-old white male with Marfan syndrome. Marfan syndrome is caused by a mutation in the gene that codes for the glycoprotein fibrillin-1, which forms the core of elastic fibrils and bonds together SM cells. For this reason, and on the basis of previous histological examinations of Marfan’s orbits, we expected elastin to be abnormal, and so did not analyze its distribution. Although the fine structure of smooth muscle is probably abnormal in Marfan’s, its overall distribution probably is not (Oh, Clark, Velez, Rosenbaum, & Demer, 2002). MRI was done after exenteration. (2) Specimen “H7” was harvested from a 17-month-old male victim of “Sudden Infant Death Syndrome”, obtained from a tissue bank. The whole head was frozen, and MRI was done after thawing but prior to exenteration..

Three-dimensional (3D) reconstruction from slice data has become familiar in connection with tomography, confocal microscopy, and such projects as the Visible Human (National Library of Medicine), so it is easy to imagine that reconstruction problems have all been solved. However, these applications require only relatively straightforward reconstruction methods: registration of each slice with the next is unambiguous, shape distortions are small or non-existent and, in any case, neighboring slices are similarly distorted. In contrast, if we wish to utilize histochemical and immunohistochemical processing to reveal the fine structure and constituent tissues in a sample, we have a much more difficult reconstruction problem, because the requisite reagents and stains can effectively penetrate only thin sections. Consequently: (1) surrounding bone must be removed to avoid damage to the microtome knife, which tends to cause soft elastic tissues, such those of the orbit, to collapse, and spatial relationships to be lost; (2) imbedding compounds, used to support the specimen during thin sectioning, introduce distortions as they harden and shrink, especially when the tissue contains distinct compartments, such as the globe; (3) registration of sequential slices is lost when they are cut; (4) each slice may be uniquely and non-linearly distorted by cutting and subsequent processing.

As an orbit or other tissue passes from life, through the stages of histologic processing, it can be imaged with increasing resolution and tissue differentiation, at the cost of accumulating distortions (Figure 1). This tradeoff is the main problem of thin-slice reconstruction, which we have approached with a method of ordered warping with intrinsic fiducials. Briefly, we used non-linear 2-dimensional image warping algorithms to bring the images from a given processing stage into alignment with corresponding images from the preceding stage. Warping must be guided by reference points, or fiducials, for which we distinguished strong structures, the globe, optic nerve & EOMs (see Figure 2), which could be found in every slice, from the remaining weak structures, the distributed collagen, elastin and SM. We then identified and correlated the strong structures across tomographic, block-face and thin-slice images, and with these structures as references or fiducials, warped each block-face image into alignment with nearby tomographic images (tomographic image planes were sparser than block-face image planes), approximately correcting the block-face images for imbedding distortions. We repeated this procedure, except now warping thin-slice images to corrected block-face images, reducing the idiosyncratic distortions of the thin-slice images. Corrected thin-slice images were then aligned, and 3D orbits were reconstructed, fitting smooth surfaces to the strong structures, and using volumetric rendering for the weak structures, so as to visualize the distributions of collagen, SM and elastin with reference to the globe, optic nerve and EOMs. Details of the reconstruction procedure are given in Appendix A.

Representative thin slices, prior to warping, are shown in Figure 3, at the level of the globe equator (A-C) and near the back of the globe (D-F). The Masson’s Trichrome (MT) series (A & D) clearly shows (in the original images, if not in the screen-resolution figure herein) all 6 EOMs, the levator palpabrae, the lacrimal gland, and many arteries and veins, along with the sclera and a complex network of surrounding collagenous fascial sheaths. The delicate contents of the globe can be seen to have been substantially distorted by retinal detachment and vitreous collapse caused by processing, but these structures were of no interest to us in this project.

The SM α-actin (SMAA) series (Figure 3B & E) shows only SM, which is found distributed in the collagenous connective tissue, and in the walls of arteries and large veins (the lacrimal gland is also stained, presumably because of its myoepithelial cell content (Warwick, 1976)); only distributed SM is of interest to us. The lacrimal gland was easily identified and digitally removed from subsequent reconstructions so that it would not obscure structures of interest. Most blood vessels could be identified by their lumens and removed, however, some smaller vessels may have been missed and so would contribute to the SM in our reconstructions. Note in Figure 3 the prominent band of SM (between the red arrows) extending from the orbital face of the MR to the IR, and the absence of any non-vascular SM in Figure 3, 3.6 mm posterior.

At the resolution of the MT and SMAA series, the Elastin van Giessen (EVG) series (Figure 3, C & F) shows little elastin. However, at higher magnification the tiny elastin fibrils are plainly visible against the orange counterstain (Figure 4). The pattern of elastin becomes apparent in the 3D reconstructions, below.

Viewed from the front (Figure 5), both H5 and H7 show that the globe and EOMs are ringed with SM; rotating each object shows that the SM ring does not extend through the depth of the orbit, but is concentrated near the globe equator. This incomplete ring corresponds to the capsulopalpebral muscle of Hesser.

Because specimen H5 was affected by Marfan syndrome, which disturbs the fine structure of SM (Oh et al., 2002), we processed H5 for the gross SM distribution, showing all regions (excepting the anterior orbit) in which our antibody-based stain detected significant amounts of SM. This required “enlarging” the lightest deposits of SM, which would otherwise have been invisible at the resolution of the reconstruction. Consequently, H5 over-represents the total quantity of SM. The H7 reconstruction, in contrast, was designed to fairly visualize the quantity of SM visible in the thin slices and, consequently, does not show the lightest deposits. Apart from these differences, the similarity of SM distributions in H5 and H7 supports our assumption that Marfan’s does not affect the overall SM distribution. With H7, we were particularly successful in correcting processing distortions: three well-aligned SM “strings” are visible in the supramedial quadrant of the anterior orbit: from top to bottom, these are the supraorbital artery, supraorbital vein and ophthalmic artery. In H5 this arterial SM is too distorted to identify. The excellent alignment of H7 reveals some of the fine structure of collagen, particularly in frontal view. However the spatial resolution of this reconstruction (260 pixels/cm) was insufficient to show the fine collagen detail visible in thin slices.

Removing collagen from the reconstruction clearly shows the SM and elastin architecture (Figure 6). Rotate the SM object (Figure 6A) to the medial side (so that the three arteries mentioned above are seen at the top of the frame) to see that the SM band extends superiorly to the superior margin of the MR. Rotate the elastin object (Figure 6B) to the medial side to see the elastin band’s similar superior extent. Comparing the medial aspects of the two objects of Figure 6, we see that both SM and elastin bands have a width of roughly 5–8 mm, and extend from the region between the IR and IO around the globe to the superior margin of the MR, mainly on the MR’s orbital face. But whereas the SM band follows an equatorial course, with its anterior edge falling roughly at the junction of the MR (shown as brick red in the figures) with its tendon (not visible because it is non-muscular), the anterior edge of the elastin band lies about 2 mm posterior to the globe equator. Rotate each object of Figure 6 so that the underside of the globe is visible to see that the concentrations of SM and elastin extend between the IR and the IO muscles, with elastin extending considerably farther into the posterior orbit.

Recall that it was necessary to set a density threshold, below which a tissue was not shown in the reconstructions. Thus, the voids apparent in the SM and elastin bands of Figure 6 are actually regions of low tissue density.

Rotate each object of Figure 6 to view the lateral orbit, where smaller amounts of SM, and traces of elastin, can be seen to extend to the LR.

Our tissues were processed for SM and elastin as far into the posterior orbit as significant densities were seen. Anteriorly, as we see in Figure 6A and B, both tissues dissipate gradually, except in the inferior orbit, where the elastin band, particularly, ends abruptly. We re-examined the histologic slides in the elastin series and determined that elastin actually continues anteriorly to the inferior tarsal plate. Classical descriptions of the inferior palpebral muscle lead us to expect SM to continue anteriorly, as well, but we found only scattered cells anterior to those shown in the reconstructions, until we approached the inferior tarsal plate, where a broad band of SM ran posterolaterally to insert on the orbital wall.

High-resolution reconstructions of the region of the MR (Figure 7) show the anterior-posterior extents of SM and elastin more clearly, and reveal that, whereas SM is found in the space between the MR and the orbital wall, elastin tends to surround the MR itself more closely (note the elastin that can be seen on the global side of the MR, through the translucent globe).

High-resolution reconstructions of the region in the inferior orbit where the IR and IO cross show that both the SM and elastin bands enter and substantially terminate in the region of intersection (Figure 8). The small amount of SM visible in the inferolateral quadrant of Figure 6A appears to be separate from the infero-medial band, and is almost absent in Sample H5 (Figure 8A).

We have described circumferential distributions of SM and elastin in the equatorial orbit, extending from the superior margin of the MR to the crossing of the IR and IO. In the inferior orbit, elastin, but not SM, continues anteriorly. We returned to the histologically prepared tissue slices to assess whether the orientations of SM cells and fascicles (bundles) followed the overall distribution of SM.

In the inferior orbit, anterior to the IR–IO crossing, SM fascicles tended to have an anteroposterior orientation, with individual cells oriented in various directions. More posteriorly, in the equatorial region, SM fascicles and cells were more circumferentially organized, with some cells running radially toward the medial orbital wall. Thus, in the equatorial region, individual cells and fascicles tend to be aligned with the overall distribution of SM, suggesting that this muscle could modulate the separation between the MR pulley and the crossing of the IR and IO.

In summary, we used a method of ordered warping with intrinsic fiducials to reconstruct the 3D architecture of histochemically identified extraocular connective tissues, and thereby identified substantial bands of SM and elastin extending from the region between the IR–IO crossing to the MR pulley. We propose that these two roughly coincident tissue bands compose a single functional structure, and call it the inframedial peribulbar muscle (IMPM).

We are impressed by the high density of both SM and elastin in the IMPM, compared to the paucity of these tissues elsewhere nearby, and by the fact that the IMPM extends between two previously-identified connective tissue structures: the stout MR pulley and the well-known condensation of connective tissues at the junction of the IR and IO, which we have elsewhere proposed functions as a double-pulley (Demer, 2002).

As with EOM pulleys, the IMPM has not gone completely unnoticed, but apart from the earlier work of our group (Kono et al., 2002), we know of no description of the heavy concentration of elastin in this region, and it is fair to say that classical descriptions do not suggest that this region contains the most substantial component of the peribulbar SM. Our findings (with the caveat that they are derived from only two samples) do not confirm the classically described continuities of peribulbar SM with the superior and inferior palpebral muscles (Duke-Elder & Wybar, 1961).

The present study and that of Kono et al. (2002) both drew histological data from the same set, but where we used mainly graphical methods of data analyses and presentation, Kono et al used mainly numerical methods, yielding two quite different and largely independent analyses. Still, there are no substantial inconsistencies between the 2 studies, which tends to validate the methods of both, and each offers unique findings, which shows some of the relative strengths of the two approaches.

Both studies agree that the structure we have called the IMPM is the most significant equatorial intermuscular connection, and we are in essential agreement on its dimensions, although the current study makes clear that it does not have a simple shape, as can be seen particularly in the H7 elastin distribution (Figure 6). The H7 reconstruction also shows that the SM and elastin distributions only partly overlap, with elastin extending more posteriorly at the MR, and so coursing anteriorly as it extends to the IR–IO crossing.

There is further agreement that the remaining three quadrants of the capsulopalpebral muscle contain little SM. The H7 reconstruction clarifies that smooth muscle in the superior quadrants is mostly vascular, and SM in the infero-medial quadrant tends to follow Lockwood’s ligament to the inner canthus, rather than contribute to an intermuscular connection.

It would not be wrong to say that the IMPM is a specialized part of Lockwood’s ligament, keeping in mind that the IMPM’s elastin and SM components are feeble lateral to the IR, where Lockwood’s is well defined, and well defined to the superior margin of the MR, where only some authors consider Lockwood’s to extend. But it seems preferable to describe the IMPM as a distinct musculo-elastic structure, joining the MR pulley with the coupled pulleys of the IR and IO.

Demer and colleagues (1997) have demonstrated that extraocular SM in the vicinity of the pulleys receives norepinepherine innervation from the superior cervical ganglion, and nitric oxide innervation from the pterigopalatine ganglion. That is, there is support for excitatory and inhibitory control of the SM band we have described. What might contraction of the IMPM do?

If its SM cells were loosely coupled or irregularly arrayed, the overall effect of contraction might simply be increased stiffness of this component of the suspension of the MR, IR and IO pulleys, and increased stability of pulleys themselves (Demer, 2000). Alternatively, if contraction resulted in reduction of the distance between MR and IR pulleys, there could be complex effects on binocular alignment. There is evidence that the MR pulley is firmly anchored to the orbital wall, whereas the IR pulley is not (Kono et al., 2002). Using the Orbit™ 1.8 simulation of extraocular biomechanics (Miller, 1999; Miller et al., 1999), it is possible to show the effect on binocular alignment of medial movement of the IR pulley caused by contraction of the IMPM (Figure 9).

IMPM contraction is therefore expected to cause the vertical gaze contingent changes in horizontal alignment that clinicians refer to as a “V-pattern”. Conversely, it can be shown that laxity of the IMPM should cause an “A-pattern” (exotropia in downgaze). It has recently been demonstrated that in convergence, the rectus pulley array extorts around the orbital axis, with the IR pulley moving nasally (Demer, Kono, & Wright, 2003). The IMPM is well-positioned to effect this nasal shift of the IR pulley in convergence, and to assist the orbital layer of the IO in effecting an inferior shift of the LR pulley. Thus, it is possible that innervation to IMPM SM is normally modulated to refine alignment, that defects in it could produce strabismus, and that pharmacologic interventions could be used to treat strabismus on these dimensions.

In some cases of blunt trauma to the periorbital region, avulsions of the rectus EOMs, called “flap tears”, are observed, most commonly in the IR, and sometimes in the MR (Ludwig & Brown, 2001). In the “narrowing” type of flap tear, it is always the lateral edge of the IR or the superior edge of the MR that is torn free. It is possible that the IMPM, which attaches to the medial edge of the IR and the inferior edge of the MR, stabilizes and protects these parts of the EOMs from this type of trauma.

Details of our reconstruction procedure follow:

In an alert subject, it would be possible to collect MR or CT images during voluntary fixation, showing physiologic muscle paths, muscle cross-sections, and globe positions (all functions of EOM innervation), as well as some of the main connective tissue structures (which may move because of EOM activity or changes in SM tone). Alert-subject tomography introduces essentially no distortion, but scan times are limited to the periods over which stable fixation can be maintained, giving modest spatial resolution in the scan plane of about 200 – 800 µm, and poor spatial resolution perpendicular to the scan plane of 2 – 3 mm. Bone, fat, muscle, and connective tissue, can usually be differentiated, but types of connective tissue cannot.

In the present study, we began with cadaveric orbits, scanned in quasi-coronal planes by MRI with bone intact (specimen 7) or after exenteration en bloc with periorbita intact (specimen 5), using 3” phased array surface coils, a T1-weighted pulse sequence and multiple excitations in a 1.5 T scanner (Signa, General Electric, Milwaukee WI), achieving pixel resolutions of 156 or 195 µm. In the absence of normal innervation patterns, EOM paths, EOM cross-sections, and the positions of dependent structures would be somewhat abnormal, and other abnormalities may have resulted from post-mortem changes. However, such cadaveric tomographic images are free of tissue processing distortions.

Bone was thinned or removed mechanically (specimen 7) and residua decalcified (Demer et al., 2000), staining with fluorscein to improve contrast of the embedded tissue while allowing the subsequently cut thin slices to be washed clean, and embedding in paraffin. Each block was then mounted on a microtome for sectioning perpendicular to the orbital axis.

The planed block face was then digitally photographed at 200µm intervals at a spatial resolution of 520 pixels/cm and color resolution of 24 bits/pixel using a Lumina digital camera (Leaf Systems), yielding a series of block-face images, which had modest tissue differentiation and certain overall distortions associated with exenteration and embedding, but no distortions associated with cutting or processing individual thin slices.

Each block-face image was compared to the nearest cadaveric tomographic image. Adobe Photoshop^TM 6.0 (Adobe Systems, San Jose CA) was used to correct the block-face image for linear distortions using the “scale” tool, and for non-linear embedding distortions that affected surrounding tissue using the “liquefy” warping tool. The most prominent non-linear distortion was an invagination of the globe caused by shrinkage of the embedding compound used to fill it. This stage yielded block-face images at 200 µm intervals corrected for exenteration and embedding distortions.

Twenty 10 µm slices were cut between each imaged block-face, and were subsequently stained and mounted: the first section in each series of 10 was stained with Mason’s trichrome stain (which shows muscle and collagen), the second with EVG (for elastin), and the third, with a stain constructed from an antibody to SM α−actin linked to a blue chromogen (Demer et al., 1997). Two more sections were saved for possible future use, and the rest were discarded.

Stained sections were digitally photographed at a spatial resolution of 520 pixels/cm and color resolution of 24 bits/pixel using the Lumina camera. This resolution allowed each slice to be captured as a single image file of manageable size, and provided adequate resolution to resolve collagen and SM, but not elastin. For sample H7 the EVG series was therefore also imaged through a microscope at 10,400 pixels/cm. These high-resolution EVG images were assembled into montages using the 520 pixels/cm images as templates. The resulting files were processed using Photoshop to increase contrast between the black-brown stained elastin fibrils and surrounding tissues, so that the elastin survived down-sampling to 520 pixels/cm.

This processing stage yielded 3 series of thin-slice images at 100µm intervals with high spatial resolution and tissue differentiation, but with non-linear distortions that were unique to each slice.

Using Morph^TM 2.5 (Gryphon Software, San Diego CA), we then chose reference points, or fiducials, on the strong structures visible in all slices, and warped each thin-slice image to its neighboring corrected block-face image, thereby correcting for non-linear distortions introduced by microtome slicing, and histologic processing.

Photoshop was then used to isolate and extract the desired soft tissues. Level adjustment was used to correct for color variation between slices. Background tinting was removed using color filtration. Images were then edited manually using selection tools to remove any remaining tissues other than those of interest.

Custom software (“TSR” for Thin-Slice Reconstruction) was developed to combine the 3D surfaces derived from the strong structures (globe, optic nerve and EOMs), with overlapping volumetric data from the weak structures (distributed collagen, elastin and SM) in the 3 series of stained thin slices. Photoshop files were imported into TSR, and color correction filters were applied to compensate for differences in staining effectiveness across slices of a given series. Contours were traced to outline the strong structures. Slices were translated into register using an automatic method that maximized the cross-correlation of pixels in adjacent slices, or manually when the automatic method failed. For strong structures, surfaces were constructed to “skin” correlated contours across slices. For weak structures, pixels were thickened to voxels (volume elements) and TSR filters were applied to smooth them within and between slices, as necessary, to reasonably represent the raw slice data, while minimizing distracting artifacts of reconstruction. Spatial resolution was limited by computer memory, but was generally sufficient to show the detail preserved in the reconstructions. Surfaces representing strong structures and volumetric data representing weak structures were then adjusted for color and transparency to maximize visibility of structures of interest, and rendered to enhance three-dimensionality. The resulting composite object was viewed from 180 different angles (360° in 20° horizontal intervals and 180° in 18° vertical intervals), and the snapshots transferred to Quicktime^TM VR Authoring Studio 1.0 (Apple Computer, Cupertino CA) to produce the final QTVR objects.

This work was supported by the Smith-Kettlewell Eye Research Institute, National Eye Institute consortium grant EY-08313 to Joseph L. Demer and Joel M. Miller, and National Eye Institute grant EY-13443 to Joel M Miller. Commercial relationships: JMM owns Eidactics, which markets Orbit™ 1.8 software.