Wide-field whole eye OCT system with demonstration of quantitative retinal curvature estimation

Current conventional clinical OCT systems image either only the anterior or the posterior eye during a single acquisition. This localized imaging limits conventional OCT’s use for characterizing global ocular morphometry and biometry, which requires knowledge of spatial relationships across the entire eye. We developed a “whole eye” optical coherence tomography system that simultaneously acquires volumes with a wide field-of-view for both the anterior chamber (14 x 14 mm) and retina (55°) using a single source and detector. This system was used to measure retinal curvature in a pilot population and compared against curvature of the same eyes measured with magnetic resonance imaging. © 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
Optical coherence tomography (OCT) has revolutionized clinical observation of the eye since its inception [1] and is an indispensable part of the modern ophthalmic practice. Part of OCT's appeal is its ability to rapidly provide non-contact, high resolution, three-dimensional (3D) anatomical information about the imaged eye [2][3][4][5]. If these 3D OCT representations of the eye covered both the anterior and posterior eye, then it should be possible to use OCT in lieu of a variety of other imaging devices to characterize ocular biometry and morphometry [6][7][8][9]. Ocular biometry and morphometry are important tasks in clinical practice such as for modern cataract surgery planning (currently using multi-modal partial coherence interferometry and keratometry [10,11]) or for research applications such as for correlating myopia and eye shape (currently using magnetic resonance imaging [12,13]). In this latter research application, MRI has been used to show that pathologic myopia is correlated with asymmetries in posterior eye shape; translation of these findings to routine clinical use would be substantially aided if OCT could be used in lieu of MRI. Early efforts to examine ocular shape with OCT have been performed (e.g. Shinohara, et al. [14], Ohno-Matsui, et al. [15], among others); however the shape analysis has been restricted to qualitative descriptors of the morphometry with limited comparison to MRI due to disparities in the separate and distinct qualitative descriptors used for each modality. Quantitative analysis of retinal or ocular shape using OCT has been limited owing to multiple optical distortions inherent to posterior segment OCT that alter the image of the posterior eye such that the image is not at 1:1 copy of anatomical morphometry; knowledge of the scanning system and the whole eye is required to remove these distortions to recreate the actual shape of the imaged posterior eye to allow for quantitative analysis from OCT [16][17][18][19] Though "whole eye OCT" has been demonstrated for ocular biometry and morphometry in animal models with relatively small eyes [20,21], current-research and clinical OCT systems image only the anterior or posterior of the human eye fully during a single acquisition. This localized imaging limits conventional OCT's use in human subjects for characterizing global ocular morphometry and biometry which requires knowledge of spatial relationships across the entire eye.
The limited localized imaging in human eyes is due to technical and optical constraints which limit OCT's ability to image the human eye as a whole. One constraint is the depth or axial range of OCT. As an interferometric imaging technique, the detected signal in OCT is a result of the interference of path length matched light from the sample and reference arms. In a time domain OCT system, the reference arm could be stepped across the entire approximately 23 mm depth of the eye so that sample reflectors from the anterior eye all the way to the posterior eye would be within a coherence length of the reference arm to produce a signal for each sample reflector at each step. However, moving the reference arm across this large distance came at the expense of increased scan time (hence increasing motion artifacts) and decreased sensitivity compared to more modern Fourier domain (FD) systems [22][23][24]. Fourier domain systems are much faster with better sensitivity but still suffer from sensitivity roll-off, wherein reflectors spatially distant from the corresponding reference arm position (higher frequencies in Fourier space) are attenuated. The image depth is further halved by the presence of the complex conjugate artifact in FD processing. For the spectrometer designs currently employed by the majority of clinical FD OCT systems, the axial range of these systems is typically limited to approximately 2 mm -an order of magnitude smaller than the axial length of the human eye.
The recent advent of sources with unique coherence length properties have overcome some of these depth limitations. Wavelength tunable vertical cavity surface emitting lasers (VCSEL) with long coherence lengths can have axial imaging ranges on the order of a meter [25,26]. Using a VCSEL source, "whole eye OCT" for ocular biometry length measurements has been demonstrated [27]. However, though this source overcame the interferometric limitations of imaging the whole eye, the demonstrated imaging system did not overcome the optical limitations of imaging over the length of the eye. In described VCSEL systems, the anterior segment can be fully visualized but only a tiny area on the retina was imaged from the initially telecentric light passing through the anterior eye optics. This imaging topology has also been used in a spectrometer based system that switched between reference arms to increase effective imaging depth but still with the same limitation in field of view of the posterior eye [28]. Producing useable OCT scans of both the anterior and posterior eye comparable to a dedicated OCT system for each region requires consideration of system depth of focus over the eye length as well as different scanning requirements for each region of the eye: telecentric for the anterior eye and collimated for the posterior eye which needs to pass first through the anterior eye optics.
Aside from combining two independent systems into a shared set of terminal optics [29,30], "whole eye OCT" with both anterior and posterior segment B scans comparable to dedicated OCT systems for each has been shown using a single swept source by either switching between imaging planes [31,32] or polarization multiplexing [33][34][35]. With polarization multiplexing, optical limitations are addressed by splitting the light into orthogonal polarization states and creating a dedicated anterior and posterior eye optical system channel for each polarization. In these early implementations, though, the field-ofview in the posterior eye only encompassed a small region around the fovea (approximately <20°) limiting their use for full ocular morphometry studies.
Building on the polarization encoded optical design and utilizing a coherence revival detection scheme, in this work we describe the development of the first whole eye OCT system with wide field of view OCT for both the anterior and posterior eye simultaneously.  Fig.  7 with the uncorrected image in Fig. 7(A) and the corrected image in Fig. 7(B). These corrections were also applied to the RPE segmentation. We fit a circle to the RPE segmentation for each dewarped radial and used the mean value of those circles as the retinal curvature estimation. We validated the above methods utilizing the eye phantom described in Section 2.2.2. For estimating the curvature of the eye phantom, we replaced the Polans eye model with the optics of the eye phantom within our ZEMAX model but kept all other methods the same.

Wide-fiel
To validate our curvature measurements in vivo, we estimated the subject's retinal curvature using MRI. The primary advantages of MRI are that it can image the subject's entire ocular globe and does not require dewarping of images acquired from an optically based, commercial imaging system. However, MRI has several drawbacks that limit its use for ocular imaging including cost, limited resolution, long acquisition time, and subject suitability.
We calculated retinal curvature measurements for MRI along the vitreous-eyewall interface ( Fig. 7(C)) [49]. For each subject, the MRI slice containing the optic nerve and the ocular lens was located. The outer vitreous boundary within this slice was segmented using active contours which generates a simple closed curve. To better match the OCT imaging scenario which is aligned to the subject's visual axis by fixation, the MRI volumes were rotated to align with the visual axis of the eye. Due to the low resolution of the MRI volumes, the visual axis was defined as 3.5° temporal to the optical axis of the eye which was identified as the line fitting manually identified corneal apex, anterior lens apex, and posterior lens pole within the MRI image [50]. After rotation, the anterior 120° of the ocular globe was removed to exclude anterior eye components from the posterior segment fitting [51]. The remaining posterior globe was then fit to a sphere by least squares to determine its radius of curvature.

Patient imaging
Informed consent was obtained from each subject under a Duke University Medical Center Institutional Review Board approved protocol prior to any imaging. The study was performed in accordance with HIPAA regulations.
Volunteers were imaged with both whole eye OCT and MRI. The OCT system described in Section 2.1 was mounted to a repurposed ophthalmic imaging stage providing translation, pitch and yaw. Subjects were seated, and a standard interface consisting of a chin rest and forehead rest was used to stabilize each subject while they were asked to fixate on the integrated fixation target reticle. Volumes, averaged B-scans, and repeated radials were then acquired with the following scan densities: rectangular volumes (2592 samples x 700 A-scans x 700 B-scans; depth x width x length; 6.38 seconds at 100 kHz), repeated B-scans in a single location (1200 A-scans x 100 repeated B-scans; 1.61 seconds), and radial volumes with repeated scans at each radial location (1000 A-scans x 8 repeated B-scans x 18 radials; 2.03 seconds). Scans with repeated B-scans were registered and averaged to improve image quality. Repeated radial volumes were utilized for ocular measurements described in Section 2.2.
Two optical safety standards were used during the study. Initial subjects were imaged with the ANSI Standard for Safe Use of Lasers (ANSI Z136.1-2014). Optical power was measured to be 1.7 mW prior to the eye which is below the standard's 8 hour limit for 1045 nm light. The power was split through the use of fiber polarization controllers with 0.7 mW focused onto the retina and 1.0 mW focused onto the anterior segment. Later subjects were imaged under the newer ANSI Standard for Light Hazard Protection for Ophthalmic Instruments (ANSI Z80. . This standard allows for increased optical power exposure for imaging under scanning beam and one hour exposure limit assumptions, both conditions met during our imaging sessions. Using this standard, optical power was measured to be 3. 6 Fig. 8 image Group surfac apex a surfac As described plane using a done anytime designed to i practice but w a summed vo the anterior se Figure 8 In addition scans at a sin corneal apex resulting in a frequencies an chamber port (Fig. 2 whole OCT re g. 9 using both mber and retina 1, the volume tation with ant the corneal ap be seen as a g een the fovea a scan direction ventional temp n to rectangula gle location, o were present clipped, squar nd therefore de ion of the ima A)) which was ollowing the m urface to be 12 nt resolution of s well with the T volumetric rende ber and view of r ye of normal subje t in both volumes, the inferior region of papilledema sub 0. 36-2016. ectangular volu h ANSI standa a from the mac s were render terior chamber pex are present gray disk in th and the nasal e of the volume oral-nasal dire ar volumes, we or both, time pe in both. Thes re-like periodic epths. In Fig. 1  ean R c as difference tion (p = Figure 12 and MRI e shape in 3). While than our ). This is eased ICP s d e l d d

Discussion
While there have been previously developed OCT systems capable of imaging the anterior and posterior regions of the human eye [25,[28][29][30][31][32][33][34][35], none thus far can simultaneously image both the anterior chamber and the retina with large fields of view using a single source and single detector. Previous systems were capable of imaging the entire anterior chamber or anterior segment but have been limited in their ability to image the retina. Single source systems that image both regions simultaneously have achieved only ~18° FOV [33][34][35] on the retina with some limited to only ~1° [25]. Systems that switch between imaging planes (thus increasing acquisition time) have been limited to 25° FOV or less [31,32]. In contrast, the system presented here images the retina with a 55° FOV and is comparable to current generation clinical imaging systems that are dedicated to imaging only the retina. Simultaneously acquiring the anterior and posterior regions of the eye allows correction of optical distortions present within the posterior imaging plane using only one set of data [18] with retinal curvature being only one potential morphometric measurement.
Creating a system that images both regions simultaneously comes with trade-offs, however. To maximize the area in which we could measure retinal curvature, during our design we optimized for retinal field-of-view. However, to accommodate the physical presence of the cube beamsplitters in the imaging optics (required for imaging the anterior chamber), 55° retinal FOV became our upper limit which is significantly smaller than the ultra-wide FOV retinal-only OCT systems that achieve between 80 and 100° FOV [53,54]. The last of the cube beamsplitters utilized a wire grid polarizer and introduced an additional compromise. An advantage of these splitters is their wide acceptance angle for a broad wavelength range over the conventional dielectric cube splitters used elsewhere in the system, ± 30° versus ± 2.5°. However, one significant drawback is that their reflection efficiency is only 80% compared to greater than 99% for the dielectric cubes, and thus we limited its use to only where we needed large angles of incidence.
We utilized linear polarization multiplexing to generate two distinct optical paths in the sample arm and the coherence revival effect present within both lasers to frequency multiplex both imaging planes into a single OCT scan. However, it should be noted that these techniques are not dependent on one another. With changes to either the sample arm path lengths or the addition of a second reference arm, one could use a laser with a narrower instantaneous line width [25] and the polarization multiplexing topology described in Section 2.1 to achieve similar results. It should be noted that because the anterior chamber and retinal imaging planes and corresponding imaging paths were polarization multiplexed, this allowed us to independently control their respective path lengths. Following initial alignment and calibration the reference arm and anterior chamber imaging path remained static but the retinal arm Diopter control optics of the sample arm were located on a highly repeatable, linear, motorized stage. This allowed us to account for subjects with different ocular lengths without the need of adjusting the reference arm or recalibrating the system between imaging sessions.
While polarization was used to create two imaging planes, this topology was not suitable for conventional polarization sensitive OCT (PS-OCT) [55]. Here each tissue image plane was interrogated with only a single linear polarization, S-polarization for the anterior chamber and P-polarization for the posterior segment, providing only partially the information needed to compute the Jones matrix [55]. Additionally, this system utilized only a single detector and not a polarization diversity detection scheme [56][57][58]; however, there may be some benefit to doing so. Each polarization multiplexed channel would have its own detector allowing for more reference power being utilized without saturating the detector, and any frequency information overlap between the two polarization states would be better separated.
Imaging both the anterior chamber and posterior segment simultaneously offers a wealth of information, however, generating individual ocular optical models required several assumptions. One limitation of the system was that it primarily imaged the anterior chamber and not the entire anterior segment including the full crystalline lens. Figure 10(B) does show an acrylic intraocular lens present within the image but this was an exception rather than the rule. While having the entire lens would be undoubtedly beneficial, differences in modeled optical distortion due to the lens between individuals is minimal due to the retinal imaging beam pivoting through, instead of across, the lens. This allows the individual beam to be focused onto the retina and have little angular change to the beam's chief ray. An exception to this is comparing phakic and pseudophakic eyes, and for pseudophakic eyes, our optical model included an acrylic intraocular lens in place of the gradient index lens of the Polans eye model [38].
It should be noted that current clinical biometers utilize a single averaged group index to estimate eye length [59,60]. In our model we included individualized optical parameters such as corneal curvature, thickness, a modeled lens, and group indices of the cornea, lens, and aqueous/vitreous [43,44]. Because the optical properties of the crystalline lens index and shape remain active areas of research [61][62][63][64] and due to individual and age dependent gradient refractive index and dynamic shape changes in non-mydriatic eyes, this may be a source of variability in the measurements that would affect any system imaging through the lens.
The important parameters that affected retinal shape are ocular axial length and system-toeye distance. Our group as well as others have previously shown that adjusting the position between the subject and system results in a variation of retinal shape [18,53,65]. Estimating subject distance along the optical axis with our whole eye system was relatively straight forward. We compared the distance between the calibration depth of the anterior chamber imaging path to the depth of the corneal apex within a given volume. These distances were entirely in air and a group index of refraction was assumed to be n g-air = 1 allowing the optical path length difference between the two measurements to directly measure the physical length. Lateral shift or rotation away from the optical axis can also cause changes in the retinal fieldof-view. This was mitigated by having the subject fixate on the integrated target and aligning the system such that the corneal saturation artifact was centered. Lack of compliance by the subject or misalignment by the system operator would introduce error to the optical model. This error could be further reduced in the future through automated pupil tracking [65].
MRI is currently the standard used to measure posterior eye shape [12,13,49,66] but is limited by cost, resolution, and accessibility. In the small pilot study shown in Section 3.2, retinal curvature as measured only with whole eye OCT based parameters was found to not be statistically significantly different from MRI; a larger, powered study is needed to definitely demonstrate the biometric equivalence of the two platforms. If the measurements are shown to be equivalent, this could open areas for research in subjects with high myopia [12,13,66] or subjects with elevated intracranial pressure where the posterior eye is being pushed in toward the vitreous [52]. In addition to these potential applications for retinal shape and because the system images the entire anterior chamber as well as the retina, whole eye OCT offers the potential for other metrics which could entirely be captured within a single volume as well. Other metrics from a single whole eye volume could include corneal shape and thickness [7,9], anterior chamber angle and depth [67], and ocular axial length [27,28,32,35] some of which we performed as part of the optical distortion correction shown in Section 2.2.

Conclusion
We have developed and demonstrated a high speed OCT system capable of truly simultaneous imaging of both the anterior and posterior eye with sufficient field of view to visualize the full anterior chamber width, macula, optic nerve, and retina to the arcades within a single acquisition. This has important implications both for clinical and research ocular imaging as well as for ocular biometric applications.

Funding
We acknowledge support from the National Institutes of Health (R01-EY024312).