Introduction and Motivation
In recent years, frequency domain optical coherence tomography (FD-OCT),1, 2, 3, 4, 5, 6 also referred to as spectral or Fourier domain OCT, has proven to combine high speed and ultrahigh resolution imaging.7, 8, 9 High-speed, frequency-domain-based optical coherence tomography ophthalmic imaging systems are commonly designed to operate around , which allows for several thousand depth scans per second for three-dimensional (3-D) in vivo imaging of the human retina with retinal pathologies.10, 11 In FD-OCT, the interference pattern in the optical frequency spectrum is analyzed to provide both high-sensitivity and high-speed, depth-resolved measurements of the human retina. Time domain (TD) standard12, 13, 14 and ultrahigh resolution OCT15, 16, 17 have already demonstrated their clinical value as a noncontact diagnostic tool for the evaluation of retinal diseases by providing virtual sections of the retina at high axial resolution. While these techniques have achieved notable successes, they are limited by relatively low signal-to-noise ratios (SNR) and therefore long data acquisition times that make them unsuitable for 3-D retinal imaging. By contrast, high-speed FD-OCT, either in its time-encoded form with high-speed spectrally swept sources and a single detector,4, 6 or spatially encoded using broadband sources and a spectrometer, is widely expected to revolutionize clinical OCT-based retinal imaging. At present, standard resolution ophthalmic OCT sources commonly operate using light sources with up to bandwidth and axial resolution.12, 13, 14 Ultrahigh resolution ophthalmic OCT systems operate in the wavelength range, which falls almost completely outside the sensitivity profile of the human eye to visible light but within the transparency region below the water absorption peak or due to limited availability of appropriate light sources. Although OCT systems can resolve all major retinal layers, image quality can be severely degraded by abnormal opaque, scattering intraocular media like corneal haze or cataracts. In clinical OCT imaging, cataracts present a significant barrier to the derivation of high-resolution images of the retina. Cataracts degrade the transparency of the lens and remain a major cause of vision loss in both the developed and developing world.18 Most cataracts are age related, but secondary lens opacities can arise as result of other ocular diseases.
OCT-based retinal imaging is mostly limited by the water absorption19 of transparent structures such as the cornea, lens, and vitreous that focus light onto the retina. The absorption characteristics of water features two usable regions, separated by an absorption peak at ; the lower wavelength region covers the visible and the near-range infrared up to . The second window is narrower and restricted to of bandwidth above . Due to the residual average absorption, the allowed power levels at the cornea according to ANSI or ICNRP standards20, 21 increase at longer wavelengths by about a factor of four. Despite this drawback and the need for InGaAs spectrometer technology,22 the major advantage of imaging at is that light scattering is significantly reduced (proportional to ) as demonstrated in measurements from excised tissue samples as based on optical tissue parameters.19
Due to lower tissue scattering at longer wavelengths, OCT imaging at has already shown deeper penetration into tissue,19, 23, 24, 25 allowing for visualization of structures beneath the retinal pigment epithelium (RPE) as well as delineation of the choroidal structure. The ability to image at these depths with high resolution has major potential for early diagnosis of retinal disorders, such as neovascularization in age-related macular degeneration in which the earliest signs of pathology are likely to occur at the RPE-choroidal interface. Given the reduced scattering at longer wavelengths, we anticipate that images with this degree of penetration and quality can also be obtained in patients with clinically significant lens opacities. With the recent availability of novel broad bandwidth light sources and InGaAs-based CCD detectors, and widely tunable light sources for imaging at , high-speed FD-OCT has the potential to become a clinically viable tool to increase our knowledge of the pathogenesis of pathologies affecting the outer retinal layers as well as providing the capacity for the early detection of diseases and the objective monitoring of any therapeutic interventions. Current light source technology favors broadband emitting sources, necessary to achieve high axial resolutions, over spectrally swept ones because of the exploitation of nonlinear broadening effects not available in tunable narrow linewidth emitters. Furthermore, ongoing camera and source development should help to improve sampling depth and acquisition speed. Time domain and spatially encoded FD-OCT can exploit these light sources. The sensitivity advantage of FD-OCT allows for high-speed 3-D scanning but involves challenges due to the imaging geometry of the broadband, high-speed spectrometer.
In this study, we present the first comparison of 3-D FD-OCT to a comparable OCT system built upon the spectrometer-based operation scheme and demonstrate the clinical feasibility and performance advantages of using this novel OCT method to image retinal and choroidal structure in patients with cataract.
To enable 3-D in vivo retinal imaging at , an FD-OCT system has been developed employing an all-mirror-based, high-resolution, high-speed spectrometer using an InGaAs, line array camera (Sensors Unlimited, Inc., Princeton, New Jersey) with pixel pitch, capable of similar to a previously presented setup22 operating at . The source operated at with a bandwidth of ( ASE module, NP-Photonics, Inc., Tucson, Arizona) emitting more than (ex fiber) and was fiber-optically linked to a 20/80 beamsplitter serving the sample arm and the reference arm, respectively (Fig. 1 ), enabling axial resolution. The reference arm incorporated a free-space pathway and dispersion compensation material to balance fiber length mismatch and the dispersion of the human eye. The device was interfaced to a patient module (OCT-2, provided by Zeiss Meditec, Inc., Dublin, California), modified for operation at to enable clinical retinal imaging at this wavelength. The spectrometer itself was designed around a flat, reflective grating, blazed for and completed with a gold-coated spherical collimation and focusing mirror, with 25 and diameter and and radius of curvature, respectively, distributing and concentrating the light along the linear pixel array. The all-reflective approach was chosen to utilize readily available components and operates in a very compact geometry, maximizing mechanical stability and minimizing internal chromatic aberrations. Electronic data were transferred by a standard CameraLink connection and acquired by a frame grabber that was connected to an external digital signal processor responsible for timing and control of the scanning galvanometric mirrors. The typical B-mode scan dimensions were 256 depth positions (512 sample points in the frequency domain) and 512 depth scans, enabling up to . The alternative system was built on the same platform (patient module, electronics and post processing) and used a lens and reflective grating based spectrometer, designed for bandwidth together with a (Atmel Aviiva) line camera and a broadband Titanium:Sapphire laser (Femtolasers Integral, Vienna, Austria) emitting with bandwidth centered at to achieve axial resolution,10 acquiring 1024 depth-scans per cross section.
The exposure time per sample line was set to at and at the system, to operate at comparable performance. Both systems were set up in parallel to allow for fast switching between the instruments and operated under optimal imaging conditions with sensitivity that was measured by comparison of the noise level to a previous calibration measurement at the same camera settings with a neutral filter of known density. The depth-dependent sensitivity loss that is common to spatially encoded OCT was found to be greater with the broadband spectrometer ( at half of the scanning range at and at ). The scanning ranges were similar, despite the different spectral ranges, detector resolutions (2048 versus 512 elements), and spectral utilization of the detectors width ( for the device and for the device). Aside from the typical effect of finite pixel size, this was mainly caused by limitations of the imaging geometry and cross talk between adjacent channels that increased the depth-dependent signal loss at higher pixel numbers and increased detector length and bandwidth. The effective axial resolutions for the systems at were found to be and at . Measurements were taken close to the zero delay position—with the proximal part of the retina, the inner limiting membrane (ILM) oriented toward the higher SNR at the zero delay to allow for optimal and equal imaging conditions of both instruments and to avoid preference of one of the systems. After spectral resampling and Fourier transformation, the cross-sectional slices were shifted according to utilizing their cross-correlation to compensate motion artifacts along the slow scanning axis.
The study was approved by the Ethics Committee of the City of Vienna, Austria, and conducted according to the tenets of the Declaration of Helsinki (Final Revision: 52nd WMA General Assembly, Edinburgh, Scotland, 2000). Written and informed consent was obtained from all study participants after detailed explanation of the aims and risks of the study. Twenty-six eyes of 14 patients were imaged. Patients were known to have cataracts and pathological conditions affecting the RPE and adjacent structures. Cataract grades were determined by an experienced ophthalmologist using the LOCS III grading scheme.26 All pupils were dilated to using topical Tropicamide 1% (Mydriaticum, Agepha) to provide sufficient parafoveal access to the retina at the different imaging conditions. While of broadband radiation at can easily be seen by the subject, light at was imperceptible even though it was delivered at higher power of up to to compensate for water absorption losses in the vitreous. However, nonlinear effects in the light generation of the amplified spontaneous emission (ASE) source emitted a small portion additionally emitted of incident light at the margin of the visible light that could be used as a dim fixation light during the scan, alternative to external fixation of the other eye. All eyes were examined with the system, immediately after the FD-OCT measurements.
Results and Discussion
Retinas were imaged with different scanning patterns, altering the transversal sampling range and orientation (horizontal or vertical) with both systems. The raster scans typically spanned regions about 1 to 3 times the foveal diameter ( of visual field, equivalent to ) in a square field. The system sampled in a grid, while the system resembled a grid, permitting to simultaneously produce high-density scans within the full volume. Compared to normal subjects, the wide image sensitivity range of almost (at absolute sensitivity in the usable image depth range, calibrated to a test measurement with an attenuated test sample) was reduced to less than ( abs.) in eyes with advanced cataract where retinal reconstruction was impossible at in the TD and FD systems. Fixation was more challenging using the faint visible portion of light emitted form the ASE source of the device compared with the bright fixation spot in the system. In some cases, external fixation with the other eye, in the or even in both systems, had to be used.
With cataract patients, the system provided a significantly higher SNR, which facilitated the investigation of intraretinal structures, despite optical scattering in the beam path. Figure 2 shows a comparison of 3-D rendered retinal volume with virtual sections for the same region of interest in a normal healthy subject without cataract. From these images, it is clear that the depth-dependent intensity characteristics at the two wavelengths are not equal, as seen with time domain and time-encoded frequency domain OCT systems in previous studies.19, 24 While producing comparable peak signals within the depth scans, the individual deeper layers have different relative intensity, in addition to the different wavelength-dependent penetration. However, the degrees of backscattering at the level of the RPE are comparable in the two measurements, indicating a low influence of wavelength-independent system parameters (i.e., FD-OCT typical depth-dependent sensitivity loss for of retinal thickness is similar). As described in a previous article,19 multiple wavelength-dependent effects play a role when comparing the intensity values, including backscattering decrease and increase of specific layers and the lower depth-dependent dispersion at . The lower transverse sampling (512 depth-scans/frame at versus 1024 depth-scans/frame at ) and axial resolution of the system reduce the contrast of subtle details of intraretinal structure. Nevertheless, for retinal layers beneath the RPE such as the choriocapillaris and superficial parts of the choroid, the visualization of the anatomical microstructure is improved in respect to the case, and individual lumina of the choroidal vessels can be visualized. Similar differences in the two imaging modalities can be noticed in patients with retinal diseases. A patient with age-related macular degeneration (AMD) and cataract could be imaged with the OCT system (Fig. 3 ), while the system failed to identify more than the RPE. Despite the present cataract, the outer retinal structure and inner choroidal structure can clearly be visualized, although the lower resolution and sampling interval reduces the overall image quality and ability to resolve details.
Deeper penetration into the choroid despite cataract is demonstrated in Fig. 4 , where the RPE is used as a boundary for segmentation of the choroid. Selection of a choroidal slice in the 3-D dataset and successive integration along the depth results in visualization of the choroidal vessel system beneath the retina, shown for the normal subject [Fig. 4a] cataract patient with AMD [Fig. 4b]. The altered distribution of vessel reflectivity in the central region and relative lower signal in its elliptic circumference coincides with the abnormal retinal structure seen in Fig. 3. It has to be pointed out that the tomograms were acquired in an orientation that allows monitoring of all retinal layers in one measurement with the higher SNR at the ILM, rather than close to the RPE. To further enhance the signal in the choroid, the OCT zero-delay plane could be placed on the other side; this however would deteriorate the signal on the more proximal retinal structure. The latter would be acceptable in the normal case but could not be used to investigate the extended pathologic structure in Fig. 3.
In Fig. 5 , 3-D OCT imaging at (top) and (bottom) of a patient with an epiretinal membrane and macular pucker is depicted. The imaging sensitivity has been significantly reduced by the presence of moderately advanced cataract (LOCS: N04 NC4) when using the OCT system. The effect of the cataract varied as a function of the scan angle in some cases, markedly reducing contrast in the tomogram and effectively obscuring any details of retinal structure. When the same retina was imaged at , sufficient imaging sensitivity allowed perceiving the details of the intraretinal structure. The increase in image pixelation and speckle arises in the image from lower transverse sampling and resolution.
Differences in the light scattering properties of the cataract can have a significant impact on the quality of the image. In Fig. 6 , the lens opacity would be expected to have a greater light scattering effect because of the presence of significant posterior subcapsular opacity (LOCS III: NO1.5 NC1.5 C2.2 P3.0). The image obtained at has sufficient signal strength to allow for 3-D imaging. However, the image obtained at provides considerably more detail of retinal structure and reveals puckering of the retinal surface with preservation of the integrity of the deeper retinal layers.
In this study, we have demonstrated that the performance of FD-OCT systems is significantly affected by light scattering induced by cataract. By contrast, light scattering with OCT systems is noticeably less and provides better delineation of retinal structure. It should be emphasized that the present study was designed to evaluate the principle of retinal imaging at longer wavelengths in patients with cataract and retinal pathologies and therefore does not provide a detailed analysis of the effects of different types of lens opacity on OCT images. A priori, we would expect less of an effect with nuclear opacities (e.g., nuclear sclerosis) compared with the greater scattering effects of cortical or subcapsular opacities. 27 A more detailed analysis of this type is the subject of further investigation.
The better performance in cataract patients of the OCT system suggests that it has a place in the clinical assessment of retinal structure. For the analysis of diseases that affect the outer retina and choroid, the great depth penetration that can be achieved with this modality suggests that it will be of value in the early diagnosis of similar conditions and in monitoring the effects of treatment. For example, some of the earliest signs of AMD can be observed at the level of Bruch’s membrane and adjacent choriocapillaries.28
Diabetic retinopathy is also associated with a change in the choriodal vascularization and with increased deposition of laminar deposits in Bruch’s membrane that are not readily apparent with routine clinical examination techniques.29 It is possible that imaging at longer wavelengths may also be beneficial in the assessment of glaucomatous optic neuropathy, possibly allowing analysis of the structural changes within the lamina cribrosa that have, clinically, been restricted to the surface layers of this structure.24, 30 Longer wavelengths may also be useful in the assessment of anterior segment structures. OCT can monitor changes in corneal thickness with great accuracy and may therefore be valuable in the assessment of laser ablative techniques used in refractive surgery where light sources might also replace sources operating at .31
Last, since the longer wavelengths do not stimulate the retina, they can be used to probe the effects of retinal stimulation by other, shorter and more visible wavelengths. We have already demonstrated that OCT can be used to detect physiological correlates of neuronal activity32, 33 within the retina, and we see the use of these longer wavelengths as a way of separating the processes of stimulation and neuronal recording at the same point in the retina.
In summary, using light sources, we have observed significant improvements in the quality of OCT images obtained in eyes with turbid media. Combined with the advantage of greater depth penetration into the choroid, we anticipate that this imaging modality will become a valuable clinical tool in early assessment, staging, and therapeutic monitoring of retinal disease.
Support for this project was provided by the following institutions: Cardiff University, FP6-IST-NMP-2 STREPT (017128), FWF Y 159-PAT, the Christian Doppler Society, NP Photonics (Tucson, Arizona), FEMTOLASERS GmbH (Vienna, Austria), Carl Zeiss Meditec, Inc. (Dublin, California), and Maxon Computer GmbH (Friedrichsdorf, Germany).