Adaptive Optics Lab

Introduction

The human retina is extremely thin and delicate, much like a piece of wet tissue paper. Yet its 1/100th of an inch thickness supports a microcosm of diverse cells organized in discrete layers, each playing a critical role in how we see. The ability of these retinal cells to function properly is unfortunately adversely impacted by disease, which can ultimately lead to blindness. Observing these cells through the natural pupil of the eye to see if they are healthy or sick is extremely difficult because optical defects in the cornea and crystalline lens blur the retinal image. To complicate matters more, reflections from cells at many different depths create a host of superimposed images. At Indiana we are developing a special camera that is equipped with adaptive optics (AO) and optical coherence tomography (OCT) that can remove the blur and disentangle the many images. The camera is designed to image microstructures the size of single cells in the living human retina. The high-axial resolution of OCT combined with the high-transverse resolution of AO provides a powerful imaging tool whose image quality can surpass either methodology performing alone. First images of the living human retina were recently collected with the camera. Early results suggest that an AO-OCT camera is significantly more effective at disentangling images than other state-of-art retina cameras. In the coming years, we will build on these previous achievements with new improvements that will allow us to see further into the microcosm of cells that compose the human retina.

This project is funded by the NSF Center for Adaptive Optics that is headquartered at UC-Santa Cruz. The project was specifically tailored to maximize the Center’s impact in high-resolution retina imaging by diversifying the Center’s investment in the development of three complimentary adaptive optics cameras: conventional flood-illumination (Rochester), confocal scanning laser ophthalmoscope (cSLO) (Houston), and OCT (Indiana). Collectively these cameras cover the major imaging modalities currently available to patients visiting an eyecare clinic. Individually, the Indiana camera may hold the highest promise for detecting the faintest reflections in the retina and sectioning the retina with the narrowest slices. It also requires the most development. When the retina is viewed through the pupil of the eye, high camera performance is necessitated by the inherent low contrast of most cells in the retina and the many superimposed reflections that originate from cells at adjacent depths in the thick retina. The need for very high spatial resolution is vividly illustrated by Figure 1 that shows a series of point spreads for various combinations of AO and major camera architectures as well as two commercial instruments (without AO). A scaled histological cross section of the human retina is shown on the left for comparison. The two AO-OCT point spreads are exceedingly small in size and are significantly smaller than those of the other camera architectures. Most importantly, the AO-OCT point spreads are at least as small as many of the cell nuclei shown in the retina cross-section suggesting that these cells can be resolved and could be detected with sufficient signal to noise.

Description of the Indiana AO-OCT Retina Camera

An optical layout and photo of the Indiana AO-OCT retina camera¹ (CCD-based) are shown in Figure 2. The camera is based on a free-space Michelson interferometer design and consists of three independent sub-systems to perform AO, 1-D OCT axial scanning, and 2-D OCT imaging.

AO System: The centerpiece of the AO system is a 37 actuator Xinetics mirror and a Shack-Hartmann wavefront sensor employing a 17x17 lenslet array. The closed-loop control operates at up to 22 wavefront corrections per second, which is sufficient for capturing most temporal fluctuations in the eye’s wave aberrations (Hofer et al., 2001).

1-D OCT Axial Scanning: The distance between the patient’s retina and the retina camera does not remain fixed due to involuntary head and eye movements, fundus pulsations (Fercher, 1984; Schmetterer et al., 1995), and drifts and microfluctuations in accommodation (which alter the OPL of the eye). Accurate positioning of the coherence gate in the retina requires tracking and compensating for these retina movements relative to the camera. This is accomplished with a 1-D OCT that performs up to 20 A-scans per sec, each traversing the full thickness of the retina, and custom software that tracks the retina in real time. 1-D OCT tracking on several subjects revealed a peak-to-valley retina motion ranging from 11 to 30 microns RMS over 5-sec intervals. This is sufficient to provide meaningful optical sectioning of the ~200 micron thick retina.

2-D OCT Imaging: En face slices of the retina were obtained using a four-step phase shift reconstruction method, analogous to that employed by phase-shift interferometry cameras (Bordenave et al., 2002; Dresel et al., 1992). The system is capable of collecting four images in less than 7 msec, although longer exposures are typically used to increase signal to noise. The short exposures significantly reduced eye motion artifacts, but did not completely eliminate it. Each of the four images was followed by a 1 msec delay to allow l/4 movement of the piezo-electric mirror.

Results

As a first step towards validating the en face OCT system and to avoid complications arising from human eyes (e.g. motion artifacts), through gated images were obtained on in vitro bovine and goldfish samples. Figure 3 shows a cross-sectional slice (x-z) through a stack of en face (x-y) reconstructions of an in vitro bovine fundus. The dark region on top is saline solution, which as anticipated reflected little light; the middle gray band is believed to be the retina; the dark elliptical structures lying immediately below the retinal surface are cross sections through small blood vessels; and the bright band deeper in the tissue is suggestive of the highly reflective retinal pigment epithelium layer and choriod. The bright band starts at 365 microns below the retinal surface. The resized image reveals blood vessels (dark patches) that are circular as would be predicted by histology. Note the relatively sharp edges of the 54 micron diameter blood vessels that is indicative of the 14 micron axial resolution. The natural appearance of the images including sharp yet curved boundaries between the saline solution, retina, and RPE as well as the circular appearance of the vessels provides strong suggestive evidence that the coherence gate is indeed stepping accurately through the tissue sample. As the CCD camera remained focused on the blood vessel for all reconstructions, spatial resolution was poor through much of the image. Yoking the focal plane to the coherence gate will remove this limitation.

Because we had little control over the sacrifice time of the bovine eye, we next performed a similar experiment on two goldfish eyes with special attention on minimizing the time delay between sacrifice and image collection. Figure 4 (right) shows a cross-sectional slice (x-z) through a stack of en face (x-y) coherence gated reconstructions. On the left is a histological sample collected on the same eye at approximately the same physical location. Note the reasonable correlation in retinal and choriodal thickness between histology and coherence gated tomographs. Features in the tomograph are reasonably well corroborated by measurements in the histology. Interestingly, the fresher fundus is found to have a noticeably more transparent retina that exceeds the sensitivity of our current OCT camera. We are puzzled by this finding as we have measured the sensitivity of our camera to be 76 dB, which based on published OCT measurements of the human retina should be sufficient to at least begin detecting some of the retinal layers.

Armed with these results on in vitro eyes, we next performed through gating on one subject. The AO-OCT camera was focused on the photoreceptor layer using an incoherent light source. The coherence gate was stepped through the retina in 10 micron steps with the 1-D OCT measuring the axial retina location prior to each image collection. The AO system provided a fixed correction across a 6.8 mm pupil at the eye with retinal interferograms collected through a smaller 6 mm pupil. Wavefront aberrations were reduced from approximately 0.3 to < 0.1 microns RMS. Interferograms were collected of a 0.5 deg patch of retina at 6 deg eccentricity (inferior). Each reconstruction was obtained from a sequence of four 4 msec images that approached one half the pixel well capacity of the retinal CCD. 42 reconstructions were collected in total. Figure 5 shows a cross-sectional slice (x-z) through the stack of 41 reconstructions that clearly reveals bright reflections at what is likely the inner limiting membrane, retinal pigment epithelium, and choriod. Although the dynamic range in the image is less than 25 dB, the through gating results indicate this range is sufficient for capturing high spatial resolution en face images of several bright tissue layers. Figure 5 also shows five small sections of en face reconstructions obtained from different depths in the volume data. Note the variation in pattern across the five images. Although not confirmative, it is suggestive of the presence of retinal structure as the pattern of noise (i.e. speckle) should be similar in all five. Better results are expected as experimental protocol and camera performance improves.

Footnote: Coherence gating in biomedical imaging systems can be realized through numerous approaches. In year one, Indiana systematically analyzed these for the specific application of imaging single cells in the living human retina with a camera outfitted with adaptive optics. High temporal (100 Hz) and spatial motion (10 microns RMS) of the retina in addition to the practical need to yoke the coherence gate (i.e. optical path length matchpoint) to the focus plane (so that adaptive optics can be effective) were severe application-specific constraints. Point scanning OCT designs – although the most abundant and mature – had the most difficulty meeting these demands using OCT technology at that time. A flood-illuminated design that employed a CCD array, although substantially less developed than other approaches, was determined the best overall gating approach for our specific application.