1. Introduction

Emerging imaging modalities have always had a significant impact on ophthalmic practice and clinical research. A-scan and B-scan ultrasonography have routinely been used in ophthalmologic biometry and diagnosis over several decades. These measurements require physical contact with the eye, therefore creating a potential risk for infection and typically provide 150 µm longitudinal resolution by using a 10 MHz transducer or 20–30 µm with limited penetration of a couple of millimeters using a 50 MHz transducer. Scanning laser ophthalmoscopy represented a major development in ophthalmic fundus imaging in the late 1970s providing an en face view of the fundus with high transverse resolution and contrast. However, pupil aperture and ocular aberrations limit the axial resolution in the retina to ~300 µm. In contrast to standard or confocal microscopy, axial and transverse resolution in optical coherence tomography (OCT) are decoupled, with the axial resolution determined by the optical bandwidth of the light source and the transverse resolution determined by the focusing of the measurement beam onto the tissue. These properties make OCT a unique ophthalmic diagnostic modality in which high axial resolution can be accomplished despite a long depth of focus (field) – a situation encountered with in vivo retinal imaging [1–3]. In addition, the eye is essentially transparent, providing easy optical access to the anterior segment as well as the retina. For these reasons, ophthalmic and especially retinal imaging has been not only the first, but also the most successful clinical application for OCT. Furthermore OCT provides information of retinal structure that cannot be obtained by any other non-invasive diagnostic technique enabling enhanced understanding of retinal pathogenesis and response to therapy [4, 5].

1.1. Data acquisition speed in retinal optical coherence tomography

Until recently the majority of commercial (about 10,000 sold from 2000 to 2007) as well as laboratory prototype OCT systems were based on time domain OCT. This technique has been demonstrated to provide scanning speeds of up to 8 kHz, i.e., 8,000 A-scans per second [6]; however, the lower system sensitivity at the faster scanning speeds in combination with limited allowed optical power has effectively limited their imaging speed in clinical applications to below 500 Hz. En face OCT, a special version of time domain OCT [7, 8] achieves very high imaging speeds by acquiring an en face retinal image at a specified depth. A powerful alternative to time domain OCT for significantly enhanced data acquisition speed are Fourier domain detection techniques where the entire depth resolved tissue reflectance (A-scan) is obtained simultaneously, thereby removing the need for depth scanning [9–12].

The first demonstration of OCT using swept source based frequency domain OCT was performed a decade ago [13, 14]. However, it was not until six years later that the sensitivity and speed advantage were recognized when OCT at 16,000 A-scans/second with ~14 µm axial resolution in the 1300 nm wavelength region was reported [15]. Recently, a new laser technique called Fourier domain modelocking (FDML) was developed that enables a ten-fold increase in imaging speed [16]. FDML achieves ultrafast tuning speeds by using a long laser cavity, consisting of an optical fiber delay line, and synchronously tuning the laser at the optical round trip time. Using FDML lasers, OCT imaging with 370,000 A-scans/second were demonstrated. The laser had a tuning range of >100 nm at 1300 nm and achieved axial image resolutions of ~10 µm [17]. Ultrahigh speed CMOS camera based spectrometer based FD OCT systems recently demonstrated in vivo retinal imaging with up to 312,500 A-scans/s at 800 nm [18].

1.2. Resolution in retinal optical coherence tomography

Since its invention, the original concept of OCT has been to enable non-invasive optical biopsy, i.e., the in situ imaging of tissue microstructure with a resolution approaching that of histology, but without the need for tissue excision and tissue post-processing. A critical step towards this goal was the introduction of ultrahigh resolution OCT (UHR OCT). Improving axial OCT resolution by one order of magnitude from the 10–15 µm to the 1 µm region enabled a noticeably superior visualization of tissue microstructure, e.g., all major intraretinal layers [19]. In retinal OCT imaging, the transverse resolution on the order of ~20 µm at 800 nm wavelength using a 1–2 mm incident beam diameter is typically achieved. The highest transverse resolution for OCT imaging is determined by the smallest achievable spot size on the retina. Studies have shown that the largest pupil size, which still yields diffraction limited focusing is ~3 mm, enabling theoretical retinal spot sizes of 10–15 µm [20]. In practice, however, for large incident beam diameters, ocular aberrations limit this minimum focused spot site on the retina, even for monochromatic illumination. Adaptive optics (AO) is a promising approach to correct ocular aberrations in order to decrease the spot size on the retina and improve transverse resolution in OCT [21]. AO has been successfully interfaced to ophthalmic imaging techniques, such as conventional, flood illuminated [22] or scanning fundus imaging systems [23] used to provide higher contrast and transverse resolution, achieving in vivo, en face visualization of photoreceptors as well as ganglion and RPE cells in animal models [24]. Integrating AO with ultrahigh resolution 3D-OCT promises to enable retinal visualization with isotropic resolution of a few micrometers [25–29].

1.3. Penetration in retinal optical coherence tomography

So far, the majority of clinical ophthalmic OCT studies have been performed in the 800 nm wavelength region. Excellent contrast especially when sufficient axial resolution is accomplished, enables visualization of all major intraretinal layers, but only limited penetration beyond the retina, resulting in limited visualization of the choriocapillaris and choroid. This limitation is mainly due to significant scattering and absorption at the retinal pigment epithelium but is also dependent on the individual fundus pigmentation of the investigated person’s eye. In clinical applications of retinal OCT imaging, cataracts also represents a significant challenge when imaging the retina due to significantly increased scattering resulting in a weak signal and therefore instrument sensitivity losses. In the 600–1200 nm region, scattering decreases monotonically with increasing wavelength, in particular the scattering behavior of light in biological tissues shows significant decrease with longer wavelengths. For this reason, OCT imaging at 1050–1060 nm can achieve deeper tissue penetration into structures beneath the retinal pigment epithelium (RPE), as well as better delineation of choroidal structure.

In vitro and in vivo time domain based OCT imaging in the 1050–1060 nm wavelength region has successfully demonstrated enhanced choroidal visualization [30, 31]. The clinical feasibility of three-dimensional spectrometer based frequency domain OCT at 1060 nm in a cross-section of patients with retinal disease and with cataracts was investigated [32, 33]. Recently, 3D-OCT retinal imaging was demonstrated with FDML lasers at 1050 nm. Axial scan rates of 236,000 A-scans/s with a tuning range of 63 nm, yielding ~11 µm axial resolution in tissue, were achieved [34]. Probing the effects of retinal stimulation by visible shorter wavelength is an important application of imaging at longer wavelengths since they do not stimulate the retina. In this context, it has been demonstrated that OCT can be used to detect depth resolved physiological correlates of neuronal activity [35, 36] within the retina.

2. Methods

2.1 Three-dimensional OCT-systems at two wavelengths with different resolution and speed

Four different spectrometer based frequency domain OCT systems have been used in the present study. All of them use a fiber optic interferometer setup and special optics optimized to the respective wavelength region and employed optical bandwidth of the light source. The key technological parameters of all systems are summarized and compared in Table 1. The performance of the three-dimensional retinal OCT system at 800 nm (3D-OCT) complied with those that are at the moment commercially available and served as the baseline to benchmark the three laboratory prototype OCT instruments. The ultrahigh speed threedimensional OCT system (UHS 3D-OCT) and the 1060 nm three-dimensional OCT system (1060 nm 3D-OCT) both utilized a fundus camera patient module (OCT-2, Carl Zeiss Meditec, Dublin, CA) with a collimator, two high speed galvanometric mirrors in close pair configuration and variable focusing optics in front of the subject’s eye optimized for the respective wavelength region. The adaptive optics OCT (AO OCT) system employed an all free space measurement arm that incorporated two optically conjugated scanning mirrors as well as a wavefront sensor and a high stroke deformable mirror. The wavefront sensor was a commercially available (HASO 32 EYE) Hartmann-Shack aberrometer that consisted of an array of square microlenses of 110 µm width. The deformable mirror (Mirao52, Imagine Eyes, France) used a novel technology with a unique performance in terms of the amplitude and linearity of the deformation and a set of 52 independent magnets placed beneath the mirror that exert a magnetic force deforming the flexible mirrored membrane with up to ±50µm stroke, allowing for correcting highly aberrated normal or pathologic eyes [29, 37].

Control of the scanning mirrors in all three laboratory prototype OCT systems was performed by a reconfigurable input/output card (NI PCI-7830R, National Instruments, USA) based on a FPGA (Field Programmable Gate Array) that was connected to a Full-CameraLink frame grabber (NI PCIe-1429, National Instruments, USA) capable of collecting up to 680 MB/s from the CMOS line camera. Software interaction was realized via a Labview (National Instruments, USA) interface that controls the FPGA subroutines and processes inbound data after reaching the frame buffer. It was composed of detector response to frequency mapping, dispersion shifting and transformation to the time domain for simultaneous display at sub-sampled arrays during the scan. In the current device display speeds during acquisition of up to 6 frames/s, at 512 lines/frame could be achieved independent of the acquisition speed.

3D-OCT used a high speed linear CCD array (ATMEL M2, CameraLink 2048 px, 20,000 A-scans/s). UHS 3D-OCT and AO OCT used a CMOS Basler sprint spL4096-140k camera (Basler AG Germany) running at 80 MHz pixel clock utilizing up to 8 taps at 8, 10 or 12 bit and therefore filling 430 MB/s of the possible 680 MB/s full signal bandwidth of the CameraLink (CL) interface at 1536 px. The spectrometer design for UHS 3D-OCT and AO OCT that was designed and built in collaboration with BaySpec (BaySpec Inc. CA, USA) implemented refractive optics and a holographic transmission grating. While the UHS and the AO system were using Ti:sapphire lasers (Femtolasers Integral (Δλ=140 nm, λ_c=800 nm emitting ~60 mW and modified versions of the Femtolasers Femtosource pro with Δλ=120–160 nm, depending on optional bandpass filters) the commercial 3D-OCT system utilized a broadband superluminescent diode with a bandwidth of Δλ=50 nm, centered at λ_c=840 nm from Superlum 37-HP. At the 1060 nm water-window two alternative amplified spontaneous emission light sources were utilized (Δλ=50 nm, λ_c=1038 nm, Multiwave, Portugal, Δλ=72 nm, λ_c=1042 nm, NP-Photonics Inc. Arizona, USA). The key component of the 1060 nm 3D-OCT system is a recently developed high speed InGaAs camera (SU-LDH 1024, manufactured by SUI-Goodrich, NJ) with a linear 1024 pixel array. The array is optimized for high speed spectrometry and can deliver a sustained line rate of approximately 47 kHz sampling at 14 bit per pixel. The OCT spectrometer, specifically designed for the camera, utilizes all reflective, off the shelf components in an optically simple and mechanically stable Czerny-Turner geometry with two spherical mirrors and gold coated 1,200 lines/mm grating [32, 38].

Table 1. Key technological parameters of OCT systems used in this study.

View Table | View all tables in this article

2.2 In vivo three-dimensional OCT imaging

For retinal imaging of normal subjects and pathologic retinas written consent was received acknowledging that the method, purpose and safety issues associated with the imaging was explained and understood, according to the local ethic protocol. The optical power at the cornea was set below 850 µW in all experiments in the 800 nm wavelength region and below 3 mW in the 1060 nm wavelength region to stay below ANSI and ICNIRP safety limits for 1s exposure at a 7mm pupil (1.1mW@800 nm, 3.5mW@1050nm) in case of a scanning mirror problem. Alignment of the measurement beam was performed by locating the iris, then scanning in both orthogonal directions to center the beam with the nodal point at the iris, focusing and adjustment of the reference position in the reference arm. Processing of the images involved reconstruction of the spatial information in the time domain by denoising, resampling, dispersion correction and transformation to the spatial domain. An adaptive wavelet shrinkage filter[39] was used to remove excess noise and speckle and finally B-mode-scan shifts of the slow scanning axis due to motion were corrected via a modified pyramidal registration algorithm[40] that also compensated continuous motion distortions during the zig-zag (bidirectional) B-mode scan via de-warping with a vector spline.

3. Results

3.1 Three-dimensional retinal imaging in normal subjects

Figure 1 demonstrates 3D-OCT at 800 nm and 1060 nm 3D-OCT in a normal (cf. Fig. 1(A–G)) and light pigmented subject (cf. Fig. 1(H–L)). State-of-the-art commercially available 3D OCT at 800 nm with 20,000 A-scan/s is compared to 1060 nm 3D-OCT with 47,000 A-scans/s. In addition to volumetric imaging of the retina, 1060 nm 3D-OCT enables significantly better visualization of the choroidal morphology, choroidal-scleral interface (cf. Fig. 1(F, G, I, K, L)) and even the sclera - especially in light pigmented subjects (cf. Fig. 1(G, L)) as compared to 3D-OCT at 800 nm. The higher data acquisition speed of 47,000 A-scans/s enables nearly motion artifact free wide-field 3D OCT imaging. Volumes have been corrected for motion artifacts by registration along the slow scan axis. Due to this expansion of the dimension and reduction by cropping of excess voxels the volume datasets have a slightly different size than the original scan.

Figure 2 depicts UHS 3D-OCT with about 3 µm axial resolution with 20,000 (cf. Fig. 2(A,D)), 80,000 (cf. Fig. 2(B, E)) and 160,000 (cf. Fig. 2(C, F)) A-scan/s in a normal subject covering about 8 degrees with 512×512 pixels. By minimizing dynamic range losses (cf. Fig. 2(D–F)), improving data acquisition speed by a factor of 6 significantly removes motion artifacts of the isotropically sampled volumes (cf. red arrows in Fig. 2(A–C)). In addition UHS 3D-OCT at 800 nm enables highly sampled (Gvoxel; 1024×1024×1024 voxel) volumetric retinal imaging resulting in high definition OCT en face fundus images (cf. Fig. 2(H) vs. fundus photo of the same subject depicted in Fig. 2(G)) as well as high definition volumetric rendering (cf. Fig. 2(I)).

Combining ultrahigh axial resolution, UHS 3D-OCT at 800 nm with pancorrected adaptive optics based on a single state-of-the-art deformable mirror with a unique high stroke capability (Mirao52, Imagine Eyes, France) allows volumetric cellular resolution retinal imaging with isotropic resolution of 2–3 µm. Figures 2 J and N reveal AO-OCT at two parafoveal locations (2° and 4°, respectively) of a normal human retina. Due to high transverse resolution, the resulting depth of focus was centered at the outer segments of the photoreceptors. AO-OCT retinal imaging not only provides qualitative (cf. Fig. 2(K, O, L, M, P, Q)) but also quantitative cellular resolution information of intraretinal morphology (cf. Table 2).

3.2 Three-dimensional retinal imaging in retinal pathologies

Figure 3 shows 3D-OCT and AO-OCT at 800 nm of the right eye of a 60 year old, white, female patient with Type 2 Macular Telangiectasia (Mactel or Idiopathic Juxtafoveal Perifoveal Telangiectasia) with 6/15 visual acuity. State-of-the-art commercially available 3D-OCT at 800 nm was used to pre-screen a large retinal volume centered in the fovea (cf. Fig. 3(E–H)) to identify regions of interest that were then investigated with cellular resolution AO-OCT @ 800 nm (cf. Fig. 3(I–T)). In this case the central foveal region was significantly impaired (cf. Fig. 3(I–N)), whereas AO-OCT revealed reasonable cellular morphology (cf. also Table 2) at 6° parafoveal (cf. Fig. 3(O–T)). Figure 3(K) depicts a slightly peripheral region that seems to cover three different stages with normal (green arrow), affected (yellow arrows) and impaired (red arrows) appearance. While the green portion seems to be comparable to normal, increased signal in the inner nuclear layer is detected (cf. Fig. 3(K) green arrow). In the yellow region being free of any cyst, a signal increase in the outer nuclear layer and a slight signal loss in the inner plexiform layer.

Furthermore, the increase in the signal reflected from the external limiting membrane (cf. Fig. 3(K) second yellow arrow from the top) might indicate changes in the tight junctions. The region that is associated with the photoreceptor outer segment (cf. Fig. 3(K) 3rd+4th yellow arrow from top) has a significant signal decrease which might indicate photoreceptor outer segment atrophy, or retraction of the retinal pigment epithelium. In the red region the cyst (cf. Fig. 3(K) 1^st red arrow from top) separates the inner plexiform layer from the underlying layers. The photoreceptor inner segment seems to be shorter than normal, the junction between the inner and outer photoreceptor segment is almost invisible. Cellular resolution AO OCT in the same eye at 6° nasal reveals reasonable appearance of the nerve fibers (cf. Fig. 3(Q)), capillaries at the level of the inner nuclear layer (cf. Fig. 3(R)) as well as photoreceptor density (cf. Table 2) at the level of the inner/outer segment junction (cf. Fig. 3(S)) and the tips of the outer photoreceptor segments (cf. Fig. 3(T)).

Figure 4 depicts 3D-OCT, AO-OCT at 800 nm and 1060 nm 3D-OCT of the left eye of a 67 year old, white, male patient with retinitis pigmentosa. Retinitis pigmentosa (RP) is an inherited disease that causes visual loss primarily through degeneration of cone and rod photoreceptors. Initial symptoms of RP include night vision loss and mid-peripheral visual field loss, progressing with age into increased peripheral fields (known as tunnel vision), and possibly extending into the macular field with further progression. Loss of photoreceptors is accompanied by alterations in retinal pigment epithelial (RPE) cell structure as RPE cells detach from Bruch’s membrane and migrate into the inner retina, accumulating around blood vessels. In addition to a thin choroid visualized by 1060 nm 3D-OCT (cf. Fig. 4(F, G)) AO-OCT reveals atrophy in the photoreceptor with ‘drusen like’ appearances especially in the foveal center (cf. Fig. 4(H, I)). Quantitative analysis of AO-OCT at about 4° nasal (cf. Fig. 4(J–M)) reveals significantly reduced cone density (cf. Table 2).

Table 2. Measured cone density for different subjects.

View Table | View all tables in this article

 

Fig. 1. 3D-OCT at 800 nm and 1060 nm of a normal retina (A-G) and a light pigmented retina (H-L): (A) fundus photo; (B) 3D-OCT at 800 nm over 20°×20°, 2× axially stretched to emphasize layer structure over wider fields, 512×128 depth scans (View 1); (C) 1060 nm 3D-OCT over 20°×20°, 2× axially stretched, 512×128 depth scans (View 2); (D) en face fundus image of the choroid using 3D-OCT, extracted from (B); (E) high definition (4096 depth scans) 3D-OCT scan over 35°; (F) en face wide field (35°×35°) fundus image of the choroid using 1060 nm 3D-OCT; (G) high definition (2048 pixel) 1060 nm 3D-OCT scan over 35°; (H) en face wide (35°×35°) field fundus image of the retina using 1060 nm 3D-OCT, extracted from (I); (I) wide field 1060 nm 3D-OCT, 2× axially stretched, 512×512 pixel (View 3), also represented in the traditional OCT color map (View 4) and color coded to emphasize the choriocapillaris and the deeper choroidal vasculature (View 5); (J) volumetric rendering of wide field 1060 nm 3D-OCT, extracted from (I); (K) en face wide (35°×35°) field fundus image of the choroid using 1060 nm 3D-OCT, extracted from (I); (L) high definition (2048 pixel) 1060 nm 3D-OCT scan over 35°. Additional color maps and transfer functions are available for use with the datasets in this paper (Color Maps).

Download Full Size | PDF

 

Fig. 2. UHS 3D-OCT (A-I) and AO OCT (J-Q) at 800 nm of a normal retina: en face fundus image over 8°×8° (512×512 depth scans) of the retina using UHS 3D-OCT at 20,000 (A), 80,000 (B) and 160,000 (C) A-scans/s; red arrows indicate significant reduction of motion artifacts with increasing scanning speed; (D-F) representative central cross-sectional tomograms extracted from the three respective volumes; (G) fundus photograph for comparison with (H); Gvoxel UHS 3D-OCT (1024×1024×1024 voxel) enables high definition OCT en face fundus image (H) and volumetric rendering (I). Cellular resolution retinal imaging using AO OCT: isotropic volumetric AO OCT in the photoreceptor region at 2° (J, View 6) and 4° (N, View 7) parafoveal; (K, O) volumetric rendering at 2° and 4°, respectively; (L, P) en face image at the level of the inner/outer photoreceptor junction at 2° (L) and 4° (P), respectively; (M, Q) en face image at the level outer part of the tips of the outer photoreceptors at 2° (M) and 4° (Q), respectively (cf. also Table 2 for cone densities).

Download Full Size | PDF

 

Fig. 3. 3D-OCT (E-H) and AO OCT (I-T) at 800 nm of a patient with Type 2 Macular Telangiectasia: (A) fundus photo; (B) autofluorescence fundus image; (C) fluorescein angiography (early phase); (D) fluorescein angiography (late phase); (E-G) representative cross-section from 3D-OCT, taken from (H); (H) 3D-OCT at 800 nm over 20°×20° (512×128 depth scans, 2× axially stretched View 8); Cellular resolution retinal imaging using AO OCT: (I) isotropic, volumetric AO OCT at 0° (View 9), retinal location indicated by white dashed line in (G); (J) volumetric rendering at 0°; cross-sections (K, L) and en face images at the level of the outer nuclear layer (M) and retinal pigment epithelium (N) at 0°; arrows in (K) indicate areas of little (green), medium (yellow) and significant (red) impairment; (O) isotropic volumetric AO OCT at 6° parafoveal location (View 10), retinal location indicated by yellow dashed line in (G); en face images at the level of the nerve fiber bundles at 6° (Q); capillaries in the inner nuclear layer at 6° (R); inner/outer photoreceptor junction at 6° (S) and at the level of the tips of the outer photoreceptors at 6°(T) (cf. also Table 2 for cone densities).

Download Full Size | PDF

 

Fig. 4. 3D-OCT, 1060 nm 3D-OCT and AO OCT at 800 nm of a patient with retinitis pigmentosa: (A) fundus photo; (B) 3D-OCT at 800 nm over 20°×20° (512×128 depth scans, 2× axially stretched, View 11); (C) 1060 nm 3D-OCT over 20×20° (512×512 depth scans, 2× axially stretched, View 12); (D) en face fundus image of the choroid using 3D-OCT, extracted from (B); (E) high definition (4096 depth scans) 3D-OCT scan over 35°; (F) en face wide field (~35°×35°) fundus image of the choroid using 1060 nm 3D-OCT; (G) high definition (2048 pixel) 1060 nm 3D-OCT scan over 35°; cellular resolution retinal imaging using AO OCT: (H) cross-section and en face image (I) of s volume at the level of the retinal pigment epithelium; retinal location indicated by white dashed line in (G); (J) volumetric AO OCT at 4° isotropic parafoveal (View 13); retinal location indicated by yellow dashed line in (G); (K) volumetric rendering at 4°; (L) en face images at the level of the inner/outer photoreceptor junction at 6°; (M) at the level of the tips of the outer photoreceptors at 6°(M) (cf. also Table 2 for cone densities).

Download Full Size | PDF

4. Discussion

4.1 Illumination limits

The signal to noise ratio is fundamentally limited by the number of photons collected at the detector. Power limitations for in vivo infrared imaging are mainly due to induced thermal processes in biological materials. Especially ocular and in particular retinal tissue is strongly affected by over-exposure. According to ANSI [41] or ICNIRP-standards [42] the maximum power for continuous 1s exposure at a 7mm pupil is 1.1mW@800 nm and 3.5mW@1050nm. Although power limits might be adjusted for the lower local exposure of scanning beams, the risk of scanner failures does not allow to safely increase the power above static, long term exposure times. Higher sampling rates (SR) result in fewer collected photons from the sample during the reduced exposure time Δt=1/SR, which results in corresponding losses [43]

These losses can be significantly reduced if the power in the reference arm is compensated to maintain a constant number of photons per scan. The interference signal is then limited by the power on the detector and the associated sampling rate dependent loss exhibits a reduced slope as compared to the linear loss when the reference power is kept constant (Fig. 5 left).

The power from the sample arm can be varied using unequal splitting ratio in favor of the light back-reflected from the sample. However, for broadband systems at 800 nm where optical circulators are not readily available or extremely expensive, this necessitates excess power available in the light source, which can be achieved with common Titanium:sapphire lasers. The axial resolution in OCT specifies the number of photons per depth sample. In contrast to the time domain denser sampling in the frequency domain only increases the depth range and thereby reduces depth ambiguity, but does not affect the sensitivity. Within every volume element (voxel) that is defined by the minimum resolvable volume Δx²·Δz, with isotropic transversal resolution Δx and axial resolution Δz, one finds the number of photons as,

with P _max being the maximum power, v_avg the mean optical frequency, sample reflectivity R and transversal sampling rate SR. I.e. the maximum available number of photons for a single depth scan at 50 kHz and a reflectivity R=-90 dB, excluding sample specific losses is ~5,300.

4.2 Motion artifacts

In a clinical ophthalmic device, the size and density of the scanning pattern are limited by the measurement duration, as well as by movement artifacts that disrupt the quality of information collected. In retinal imaging the minimum blinking rate of an average human subject of 0.2–0.3 Hz sets the maximum time for a single continuous measurement. The stability of the tearfilm, required for good optical quality, is closely associated to this time. Although longer measurement times of up to 15 s are achievable commonly clinical application above 3–5 s is impaired and >10 s are problematic. Motion occurring within the measurement period results in spatial distortions or breaks (tearing) in the volume. Standard fluctuations in the retinal position are due to slow continuous drifts or due to high speed microsaccades that occur every 0.5–2 seconds and have typical amplitudes between 0.17° and 4° [44] (1° external viewing angle ≈288 µm image size). In contrast to well fixating young subjects a considerable range of patients suffer from stabilization problems, not only those with apparent functional disorders like nystagmus, but also subjects with visual impairment, where locating the fixation target can become a problem leading to stronger drifts. The average eye-motion during fixation is reported to reach excursions of 0.25–2.3° depending on the time scale of the re-centering binocular saccades [45], with accelerations of up to 14,500°/s² and speeds of 30 to ~360°/s [46] with a typical maximum of ~100°/s [44] and drifts with an order of magnitude lower excursions and speeds. This transversal motion at the retina is superimposed by axial and rotational motion at lower speeds, causing distortions in the scanned volumetric image [47].

 

Fig. 5. Left: Decrease of optical OCT signal strength with speed. Signal loss with uncompensated reference arm (red), signal loss (green) with power increase in the reference arm (purple) is approximately halved and permits to increase the speed with less impact on the sensitivity, measured sensitivity (black, including optical losses). Right: Imaging angle and width of the fast axis suppressing motion artifacts due to fixation drift (gold) critically sampled at 20 µm transversal resolution (10 µm spacing). At speeds above ~50 kl/s equivalent to 5°×5° the overall motion free imaging range is governed by the blinking rate when the drift is corrected; 3s blink range for a square image (green, linear inlet). To correct normal micro-saccades (MS) within 5°×5° a speed above 400 kl/s is necessary (red). Quasi-static wide-field imaging where all motion artifacts (350°/s) can be suppressed is found around 10 Ml/s sampling rate. Microsaccades can also be prevented by acquisition within the 0.5–2 s timeframe between two MS (black). Snapshot type-imaging of the macula (~8°) can be achieved with speeds above 150 kl/s.

Download Full Size | PDF

While it is possible to avoid distortion artifacts by actively tracking eye motion with an additional instrument, numeric correction methods keep the acquisition system simple. Distortion due to image motion may be compensated by linear numeric registration and correction without additional information such as a fundus photograph if two necessary conditions are met.

For critical sampling the linear scanning range r_lin is dependent on the typical size of the scanning OCT laser spot on the retina described by the transversal resolution Δx. Multiplying the distance between sample points by a given number s_f of samples along the fast axis yields the linear range r_lin across the retina and division by the angle to distance conversion factor c=288 µm/° gives the angular scanning range r_deg given by

 

Fig. 6. Drift motion artifact ambiguity free imaging zones (solid squares) and blink free imaging zones (square frames) at critical sampling, assuming a max. drift of 10°/s and a blink rate of ~3 Hz for the investigated systems. Maximum motion due to high speed, high displacement microsccades (inner dotted circle normals, outer pathological) distorts the scan, limiting the width of the fast axis scan. Increase in the sampling speed significantly enlarges the “snapshot” region towards the perifovea. Adaptive optics, due to its much higher transversal resolution stays limited to small sub-regions and sensitive to microsaccades.

Download Full Size | PDF

with OS being a transversal oversampling factor. Simple numeric registration and correction algorithms permit the reconstruction of the original image in 3D if neighbored tomograms are not overlapping with the necessary condition that the slowest scanning speed is higher than the motion of the object v₀ as described above. For a two dimensional raster scan the slow and fast axis scan speeds v_s and v_f are coupled via the sampling rate and the number of sampling points. Assuming an equivalent transversal sampling interval Δx/(2·OS) in both directions the velocities are obtained as:

thereby a duty cycle DC is included that reduces v_s in order to account for delays between successive tomograms. With the necessary condition vs≥v0, (4) and (3) we finally obtain the following expression for the sampling rate:

The implications (cf. fig. 6 right) for the systems used in the manuscript are summarized in Table 3, using a zig-zag scan with a duty cycle of DC≈100%. For a given scanning rate SR, lateral resolution Δx, and oversampling factor OS, there is a size r_lin of the rectangular scanning pattern that can be corrected for motion induced distortions. Figure 6 represents the drift motion artifact-free zones (shaded regions) and the 3 s-‘blink’ regions (lines) graphically for the four systems. Higher sampling speed significantly improves the ‘snapshot’ image regions without motion artifacts, reaching up to the perifovea in the case of the UHS-OCT system. The limits can be relaxed to typical drift motions ~3–5°s, which permits imaging a four times larger area in each case, but reduces the reliability of the device. Especially AO-OCT due to its high resolution is strongly affected by eye motion and limited to a very small field of slightly more than 200² µm² even in the relaxed case. All systems are prone to microsaccades, since the high speeds involved (100–360°/s) would need sampling speeds over 1 Ml/s to be directly corrected (Fig. 5 right). The circumvention of the problem by scanning faster than the minimum microsaccade delay (~0.5 s) only makes the 1060 nm device feasible within foveal rim (~5°) and the UHS-OCT-system for the whole macula. Again, the microsaccade limit can be relaxed to 1–2 s for most of the subjects and multiple successive volumes. In this case the gain is smaller and the UHS-device scan covers the perifovea or the fovea and the optic nerve head in one image, while at 47 kl/s the macula is covered without distortions. The 20 kl/s device, however, only reaches the foveal rim. With these relaxed settings the 160 kl/s AOsystem is still limited to a little bit more than one degree, making it practicable only for preselected sub-regions. Improvements can be made by adjusting the scanning direction, so that the majority of the motions is directed along the fast scan direction (typically horizontal), or by limiting the fast axis length to perform rectangular rather than square scans. As an example the image Fig. 2(C) covering the macula exhibits the same distortions as the wider scan (Fig.2(H)). For the same macular region, however the distortions are smaller at the expense of lower sampling density.

Table 3. Sampling parameters of OCT systems used in this study.

View Table | View all tables in this article

Another way to get around the microsaccades is additional tracking support like matching the OCT image with a fundus image or using an eye tracker capable of extremely high response time ~2 ms and acceleration and in the AO-case very fine resolution, which complicates the OCT devices significantly. This effect is seen in the en face scans acquired with the AO-OCT device in Fig. 3(Q–T), where two horizontal stripes caused by microsaccades tile the image, while in Fig. 4(L–M) the object appears shifted obliquely with duplications of the same structure.

In the images produced by the systems compared in this manuscript (cf. Table 3) an oversampling factor of 1–2 was used to permit isotropic sampling with high image quality across wide scanning angles. By suppressing rather the average drift ~3°/s rather than the peak drift of 10°/s most of the subjects could be measured and motion could be compensated. In some cases, especially subjects with severe pathologies and lack of good fixation microsaccades commonly distorted the scans multiple times, which had to be compensated by non-linear stitching of mosaics (Fig. 4(L,M) original, Fig. 4(J) stitched)

5. Conclusions

State-of-the-art photonics and detection technology enables significantly improved performance of three-dimensional retinal OCT imaging regarding data acquisition speed, axial and transverse resolution as well as penetration into the choroid. In the present study, high speed imaging at 1060 nm with 47,000 A-scans/s, ultrahigh speed at standard resolution with 160,000 A-scans/s at 800 nm as well as cellular resolution retinal imaging at 160,000 A-scans/s at 800 nm has been compared to standard commercially available 3D-OCT at 800 nm with about 20,000 A-scans/s in normal and pathologic retinas. Whereas improved speed enables nearly motion artifact-free wide-field imaging of the choroid using 1060 nm 3D-OCT, UHS OCT at 800 nm enables ‘volumetric snap-shot OCT’ recording volumes with 512×128×768 voxels in 0.2 s or ultrahigh definition (Gvoxel) volumetric imaging. Theoretical analysis of the influence of eye motion on three-dimensional OCT imaging revealed that scanning protocols should be favorably short in the fast axis direction, but long in the slow one. In the present study the peak drift was not fully compensated, resulting in residual motion artifacts despite high data acquisition speeds. Although 3D-OCT at 800 nm is capable of visualizing superficial parts of structure below the retinal pigment epithelium (strongly depending in the investigated eye’s fundus pigmentation), 1060 nm 3D-OCT enables significantly improved visualization of choroidal structure including the choroidal-scleral interface in eyes with clear ocular media. In addition, 1060 nm 3D-OCT is clinically more feasible for retinal imaging in patients with turbid anterior eye segments (e.g., corneal haze or cataract) [32]. Finally the concept of clinical cellular resolution retinal imaging has been introduced. 3D-OCT at 800 nm was used to pre-screen a large retinal volume to target suspicious sites, where AO OCT has then successfully been used for three-dimensional qualitative and quantitative cellular resolution retinal imaging.

According to these preliminary results it seems that improved data acquisition speed, enhanced penetration and isotropic cellular resolution clearly provide synergistic morphologic information of intraretinal morphology. More systematic studies that investigate the clinical feasibility of these complementary OCT modalities presented in the present manuscript on a large number of patients with different retinal pathologies will enable to evaluate if the extra technological effort is worth the additional diagnostic information they enable.