Three-dimensional imaging of the human retina by high-speed optical coherence tomography

Conventional optical coherence tomography is based on A-scans, i.e., the fast scan direction is the z-direction. While this technique has been successfully demonstrated for two-dimensional cross sectional imaging of various tissues, it is rather slow if three-dimensional information is to be obtained. We report on a new technique that combines the transverse scanning approach of a confocal scanning laser ophthalmoscope with the depth sectioning capability of OCT. A stable high-frequency carrier is generated by use of an acousto optic modulator, and high frame rate is obtained by using a resonant scanning mirror for the priority scan (xdirection). Our prototype instrument records 64 transverse images consisting of 256x128 pixels in 1.2 seconds, thus providing the fastest retinal 3D OCT scanning system reported so far. We demonstrate the capabilities of our system by measuring and imaging the fovea and the optic nerve head region of healthy human volunteers in vivo. ©2003 Optical Society of America OCIS codes: (110.4500) Optical coherence tomography; (110.6880) Three-dimensional image acquisition; (170.4470) Ophthalmology. References and links 1. D. Huang, E. A. Swanson, C. P. Lin, J. S. Schuman, W. G. Stinson, W. Chang, M. R. Hee, T. Flotte, K. Gregory, C. A. Puliafito, J. G. Fujimoto, "Optical coherence tomography," Science 254, 1178-1181 (1991). 2. B. E. Bouma and G. J. Tearney, Handbook of optical coherence tomography (Marcel Dekker, New York, 2002). 3. A. F. Fercher and C. K. Hitzenberger, "Optical Coherence Tomography," Progr. Opt. 44, 215-302 (2002). 4. C. A. Puliafito, M. R. Hee, C. P. Lin, E. Reichel, J. S. Schuman, J. S. Duker, J. A. Izatt, E. A. Swanson, J. G. Fujimoto, "Imaging of macular diseases with optical coherence tomography," Ophthalmology 102, 217229 (1995). 5. W. Drexler, O. Findl, R. Menapace, A. Kruger, A. Wedrich, G. Rainer, A. Baumgartner, C. K. Hitzenberger, A. F. Fercher, "Dual beam optical coherence tomography: Signal identification for ophthalmologic diagnosis," J. Biomed. Opt. 3, 55-65 (1998). 6. R. Lattanzio, R. Brancato, R. Pierro, F. Bandello, B. Iaccher, T. Fiore, G. Maestranzi, "Macular thickness measured by optical coherence tomography (OCT) in diabetic patients," Eur. J. Ophthalmol. 12, 482-487 (2002). 7. A. M. Rollins, M. D. Kulkarni, S. Yazdanfar, R. Ung-arunyawee, J. A. Izatt, "In vivo video rate optical coherence tomography," Opt. Express 3, 219-229 (1998), http://www.opticsexpress.org/abstract.cfm?URI=OPEX-3-6-219. 8. A. G. Podoleanu, G. M. Dobre, D. A. Jackson, "En-face coherence imaging using galvanometer scanner modulation," Opt. Lett. 23, 147-149 (1998). 9. A. G. Podoleanu, M. Seeger, G. M. Dobre, D. J. Webb, D. A. Jackson, F. Fitzke, "Transversal and longitudinal images from the retina of the living eye using low coherence reflectometry," J. Biomed. Opt. 3, 12-20 (1998). 10. H.-W. Wang, A. M. Rollins, J. A. Izatt, "High speed, full field optical coherence microscopy," in Coherence domain optical methods in biomedical science and clinical applications III, V. V. Tuchin and J. A. Izatt, eds., Proc. SPIE 3598, 204-212 (1999). (C) 2003 OSA 20 October 2003 / Vol. 11, No. 21 / OPTICS EXPRESS 2753 #3031 $15.00 US Received September 15, 2003; Revised October 14, 2003 11. R. H. Webb, G. W. Hughes, F. C. Delori, "Confocal scanning laser ophthalmoscope," Appl. Opt. 26, 14921499 (1987). 12. E. A. Swanson, D. Huang, M. R. Hee, J. G. Fujimoto, C. P. Lin, C. A. Puliafito, "High-speed optical coherence domain reflectometry," Opt Lett 17, 151-153 (1992). 13. American national standard for safe use of lasers. ANSI Z 136.1 (Laser Institute of America, Orlando, 2000). 14. F. I. Feldchtein, G. V. Gelikonov, V. M. Gelikonov, R. R. Iksanov, R. V. Kuranov, A. M. Sergeev, N. D. Gladkova, M. N. Ourutina, J. A. Warren Jr., D. H. Reitze, "In vivo OCT imaging of hard and soft tissue of the oral cavity," Opt. Express 3, 239-250 (1998), http://www.opticsexpress.org/abstract.cfm?URI=OPEX-36-239. 15. S. A. Boppart, B. E. Bouma, C. Pitris, G. J. Tearney, J. F. Southern, M. E. Brezinski, J. G. Fujimoto, "Intraoperative assessment of microsurgery with three-dimensional optical coherence tomography," Radiology 208, 81-86 (1998). 16. A.G. Podoleanu, J. A. Rogers, D. A. Jackson, S. Dunne, "Three dimensional OCT images from retina and skin," Opt. Express 7, 292-298 (2000), http://www.opticsexpress.org/abstract.cfm?URI=OPEX-7-9-292. 17. B. M. Hoeling, A. D. Fernandez, R. C. Haskell, E. Huang, W. R. Myers, D. C. Petersen, S. E. Ungersma, R. Wang, M. E. Williams, "An optical coherence microscope for 3-dimensional imaging in developmental biology," Opt. Express 6, 136-146 (2000), http://www.opticsexpress.org/abstract.cfm?URI=OPEX-6-7-136. 18. P. Massin, G. Duguid, A. Erginay, B. Haouchine, A. Gaudric, "Optical coherence tomography for evaluating diabetic macular edema before and after vitrectomy," Am. J. Ophthalmol. 135, 169-177 (2003). 19. C. Strøm, B. Sander, N. Larsen, M. Larsen, H. Lund-Andersen, "Diabetic macular edema assessed with optical coherence tomography and stereo fundus photography," Invest. Ophthalmol. Vis. Sci. 43, 241-245 (2002).


Introduction
Optical coherence tomography (OCT) is now known since more than a decade as a noncontact high resolution technique for obtaining cross sectional images of transparent and translucent structures [1][2][3].The main application field of OCT is imaging of the human retina where several reports have shown the diagnostic power of the technology [4][5][6].Despite the success of OCT in retinal diagnostic applications, the technology still has, in its present state, a major shortcoming: it is rather slow if three-dimensional data are to be obtained.
Conventional OCT is based on A-scans, i.e., the fast scan direction is the z (depth) direction.Typically 100 -400 A-scans/s are recorded with commercially available instruments.To record a 3D data set with transversal resolution of at least 100x100 pixels in x and y direction would require 25 -100 seconds -too slow for imaging of patients.Highspeed OCT with up to 4000 A-scans/s has been demonstrated in human skin and animal tissues [7].The wavelength used in that report (1300 nm) is not applicable for imaging the human retina because it is absorbed by the water contained in aqueous and vitreous humors of the eye.
A different approach of retinal OCT based on transverse scanning has been reported by A. Podoleanu et al. [8,9].In this case the fast scan direction (x) is parallel to the retinal surface.One complete transverse (x-y) image is recorded before the depth position of the coherence gate is changed and the next transverse image is acquired.This scanning scheme (C-mode imaging) has the advantage that ophthalmologists are more familiar with the interpretation of transversal images since they are of similar orientation as those of frequently used fundus cameras and confocal scanning laser ophthalmoscopes (CSLOs).The specific technology of transversal retinal OCT reported so far uses the path length modulation introduced by the x-y galvo scanning mirrors to generate a carrier frequency.This method has the drawback of a varying carrier frequency which requires a broader electronic filtering bandwidth which in turn reduces the sensitivity.This might be one of the reasons why this technology was limited to a speed of ~ 2 C-scans/s.
We report on a new transversal OCT technique for retinal imaging.A highly stable carrier frequency is generated by use of an acousto-optical modulator (AOM) [10].In combination with a high-speed resonant transversal scanner, we have developed a prototype instrument that records a full three-dimensional data set of the human retina in ~ 1 second.This is, to the best of our knowledge, the highest speed of 3D retinal OCT imaging yet reported.Once a 3D data set is recorded, 2D sections through the 3D data set can be derived by software with arbitrary section geometry.We present 3D data sets recorded in vivo in the fovea and nerve head region of healthy human volunteers.The 3D information is displayed in form of movies of adjacent 2D sections along x-y, x-z, and y-z planes.Furthermore, we demonstrate retinal thickness mapping derived from the 3D data information.

Methods
Contrary to conventional OCT, our system is based on transversal scanning, combining the principles of CSLO [11] and OCT. Figure 1 shows a basic sketch of the operating principle of such a combined instrument.A low coherent light source emits a beam of short coherence length, centered at optical frequency ν 0 .This beam illuminates an interferometer where it is split into two components: a reference beam and a sample beam.The reference beam traverses an AOM, is reflected at the movable reference mirror, recombined with the sample beam at the beam splitter, and both beams finally travel towards the detector.The function of the AOM is to shift the light frequency of the reference beam from ν 0 by an amount ∆ν to a new frequency ν 1 .The reference mirror, which is coupled to a stepper motor, acts as a variable path delay unit (PDU) that is used to vary the length of the reference path between successive C-scans, thus changing the position of the coherence gate within the retinal tissue.The sample beam is scanned transversally over the retina by an x-y scanning unit.A 3D data set of the retina is recorded in the following way: initially, the PDU is set to place the coherence gate slightly anterior to the anterior retinal surface (the inner limiting membrane, ILM).The x-y scanner scans the complete x-y plane corresponding to the depth position of the coherence gate (the layer thus scanned is the so-called coherence layer; its thickness is equal to the coherence length).If the sample beam hits a backreflecting or backscattering site within the coherence layer, the backscattered sample beam will interfere with the reference beam.Since sample and reference beams differ in frequency by an amount ∆ν, the resulting interference signal will oscillate at ∆ν (the beat frequency).∆ν constitutes a very stable carrier frequency that can easily be extracted by filtering with a band pass filter centered at ∆ν.The magnitude of the signal thus extracted is a measure of reflectivity of the sample as a function of the transversal position at the depth defined by the PDU.The signal magnitude is encoded on a gray scale.After one such C-scan, the PDU setting is changed to move the coherence layer deeper into the retina, and another C-scan is recorded.Step by step, the coherence layer is shifted through the retina until its final position is beyond the posterior retinal surface (the retinal pigment epithelium, RPE).In this way, a 3D data set of a selected region of the retina is recorded.Figure 2 illustrates the transversal scanning scheme, as compared to the conventional, A-scan based scheme.Our prototype instrument is based on a commercial CSLO platform (TopSS, Laser Diagnostic Technologies, Inc.) which was modified in several ways.The laser was replaced by a high power super luminescent diode (SLD; λ 0 = 840 nm, ∆λ = 21.3 nm, P 0 = 20 mW) as the imaging light source.An interferometer (whose operation is functionally equivalent to that of fig. 1) was integrated into the beam path.The instrument operates at a carrier frequency ∆ν = 40 MHz.The light is detected by a silicon detector plus amplifier unit with an electronic signal bandwidth of 2 MHz.Because of ample light intensity of the high power SLD, the AOM insertion loss (diffraction efficiency specified by manufacturer ~ 85%) can be neglected.The x-y scanning unit consists of a fast resonant scanner (4 kHz) for x-scanning and a galvo scanner for y-scanning.

Longitudinal
The instrument presently has the following imaging parameters: The scan field size is variable: in x-y direction from 5° x 5° to 15° x 15°, consisting of 256(x) x 128(y) pixels; the focal spot diameter is estimated from Gaussian beam optics to ~ 15 µm (FWHM diameter of field amplitude distribution, in analogy to the definition of depth resolution [3]).In z-direction the scan depth can be varied between 350 and 1050 µm (in air), consisting of 64 slices (i.e., 64 pixels in z-direction).The depth resolution is ~ 16 µm (~ 12 µm in tissue) for the largest scan depth of 1050 µm (limited by the number of pixels), for smaller scan depths it is limited by the (round trip) coherence length [12] to ~ 14.5 µm (~ 10 -11 µm in tissue).The recording time is 1.2 seconds for the full set of 64 C-scans, or ~ 53 C-scans/s.This corresponds to up to 1.75 Mega voxels/s (spatially resolved voxels).The beam power illuminating the eye is 5 -10 mW (depending on field size), which is safe according to ANSI standards for repeated exposures within the imaging time of 1.2 s [13].The system sensitivity, defined as the ratio of the squared interference fringe amplitude obtained from a mirror (reflectivity 1) and the noise variance, is ~ 83-85 dB.
3D OCT data sets were recorded in the eyes of healthy human volunteers after informed consent was obtained.OCT recordings were taken from the fovea and the optic nerve head region.No mydriatic pupil dilation was used.Once a 3D data set is recorded, the next step is to project the 64 C-scans on top of each other to obtain a 2D image similar to a conventional CSLO image.After that step, cross sectional images of arbitrary orientation can be derived by software from the 3D data set at any location where features of interest are observed.Furthermore, movies of successive adjacent cross sections can be generated, showing the full 3D data set.

Results
Figure 3 is a movie showing the recording sequence of a 3D data set in a human fovea.An area of 10° x 10° is covered, 64 C-scans were recorded with a depth increment of ~ 16 µm between successive scans.During the first frames of the movie, the coherence layer was positioned in the non-backscattering vitreous, anterior to the retinal surface, hence the first frames are dark.As the coherence layer is moved deeper into the eye, the first bright structure appearing is the ILM.The ILM is, due to the curved geometry of the eye ball, first visible at the periphery of the scan field.As the coherence layer moves deeper, the ILM and deeper tissue structures cover more and more of the visual field.The foveal depression is indicated by the central dark disc whose diameter decreases with increasing depth.After the coherence layer has moved beyond the foveal pit, a second bright structure, even brighter than the ILM, is observed.This is attributed to the RPE.Furthermore, during the shift of the coherence layer through the retinal tissue, vessels can be observed at peripheral parts of the scan field.These vessels light up brightly when they are located in the coherence layer.As the coherence layer moves through the RPE, the dark shadows of the vessels on the RPE can be observed.After recording of the 64 C-scans, they are mounted to a 3D data set.Figure 4     A retinal area of great diagnostic interest is the region of the optic nerve head.Retinal thickness, retinal nerve fiber layer thickness, as well as the dimensions and shape of the nerve head excavation are important for glaucoma diagnostics.3D data sets of the optic nerve head region were recorded in healthy volunteers in the same way as discussed in the context of Fig. 3. Figure 6

Discussion and conclusion
Although the first report on OCT was published in 1991, rather few papers on 3D OCT have yet been published.Among the applications reported so far were dental OCT [14], imaging of vessels and nerves [15], retina and skin [16], and developmental biology [17].The reason for the rather small amount of published work on 3D OCT, at least in case of retinal imaging, is probably the recording time of at least several seconds, up to several minutes, which makes in vivo 3D imaging rather difficult.
We have developed a new OCT system that overcomes this limitation.Our system is based on transversal, or C-scan imaging.Previously reported transversal OCT systems use the path length modulations caused by the scanning mirrors [8,9,16] or a vibrating reference mirror [17] for carrier frequency generation.Our system uses an AOM for this purpose.This method of carrier frequency generation was already used for optical coherence microscopy, where 4 -8 C-scans/s were reported [10].The drawback of this method is that an additional electro optic device is needed.The advantage is that a very stable carrier frequency well above the imaging bandwidth is generated by the AOM which allows for high speed imaging.If OCT systems employing different scan patterns are to be compared in terms of speed, a useful performance figure is the number of spatially resolved voxels per second.Spatially resolved voxel means total scan length (in each spatial dimension) divided by the corresponding sampling function width (but of course limited by the number of sampled data points in each scan direction).In other words: only those voxels are spatially resolved whose sampling functions do not overlap.In transverse direction, this condition means that the width of the focused beam is equal or smaller than the pixel width, a condition fulfilled by our instrument in the case of the 15° x 15° scan field (cf.Fig. 5).In depth direction, this condition means that the sampling interval is not smaller than the coherence length (i.e., oversampling in depth direction, as frequently done in A-scan based OCT, does not generate more spatially resolved voxels than sampling intervals that are matched to the coherence length; it should be mentioned, however, that spatial oversampling can be used to improve the image quality to some extent).Commercially available retinal OCT instruments (Carl Zeiss Meditec) record typically 100-400 A-scans of ~ 2 mm length per second.With a coherence length of ~ 10 µm, this corresponds to 20 -80 kVoxels/s (spatially resolved).Our instrument records up to 1.75 MVoxels/s (spatially resolved).This is an improvement of 1-2 orders of magnitude in recording speed.
Since our system records a full 3D data set in only ~ 1 sec, it can easily be used for imaging in patients, and the full advantages of 3D imaging becomes available: A full 3D data set is recorded in a single session.After the recording of the data set, 2D sections of arbitrary section geometry can be derived at any location of interest.Since the full information is available, no structural details are missed which can easily happen if only a single 2D section is available.Features of interest can be viewed in detail with several adjacent sections and with different section orientations without the necessity of re-positioning and re-measuring the patient.3D image acquisition should also be very favorable in follow-up studies since it is often difficult to re-position a patient to record exactly the same retinal location with conventional 2D OCT.Another advantage of our 3D technique is in retinal thickness mapping.Conventional OCT typically records 6 radial scans of the macula [18].Near the foveal center, the sampling density is quite good with this scanning pattern, however, the measurement results depend on the fixation capabilities of the patient, and towards the periphery of the scanned area, gross averaging is necessary, causing a poor azimutal resolution [19].This is cumbersome if the retina is scanned for macular edema, e.g., in patients with diabetic retinopathy, since smaller edema can easily be missed.Our technique has a constant high transversal resolution throughout the entire scan field that should avoid this problem.Further resolution improvement by dynamic focus tracking can easily be implemented in our C-scan based technique.Finally, our technique should be very useful for volumetric measurements, e.g., for measuring the volume of macular edema, which is important for diagnosis and treatment follow-up of patients with diabetic retinopathy.
One of the problems of our present instrument is the limitation to 64 C-scans (caused by the presently used frame grabber).This causes a trade-off between depth resolution and achievable total image depth.E.g., the slight undersampling in depth that is used in case of the largest depth step size of 16 µm, as compared to the depth of the sampling function of 14.5 µm (the coherence length), degrades the achievable depth resolution by the same amount.If we adapt the depth step size to the coherence length, we are limited to a depth range of 64 x 14.5 µm ≅ 930 µm or ~ 670 µm in tissue.While this depth range is enough for imaging normal (thickness ~ 300 µm) and moderately thickened retinas, there might be problems with retinas containing thick edema or with deeply excavated optic nerve heads.Therefore, we plan to double the depth scanning range of the instrument to 128 C-scans which should be sufficient for imaging even extreme cases of retinal pathologies.
In conclusion, we have developed a new system for retinal OCT that combines the principles of CSLO and OCT.The instrument allows to record 3D data sets of the human retina in ~ 1 sec.We have demonstrated the capability of the instrument to measure and image the fovea and the optic nerve head region of healthy volunteers.The acquired data can be used to derive 2D sections of arbitrary orientation and location through the retina and to construct retinal thickness maps of constant high transverse resolution throughout the scan field.

Fig. 3 .
Fig. 3. (1.16 MB) Frame no. 10 of movie showing the recording sequence of a 3D OCT data set in the fovea of a healthy volunteer.Image size: 10° x 10°.Depth step size between individual movie frames: 16 µ m. 45 transversal frames out of a total of 64 are shown (the first frames corresponding to positions entirely in the vitreous are omitted).
shows a movie of software derived B-scans (x-z) through this data set.The upper part of the figure shows a projection of the 64 C-scans on top of each other, yielding a CSLO-like intensity image.The dark area near the center corresponds to the foveal depression.The lower part of the figure shows software derived B-scans (image size ~ 3 mm horizontal x 1 mm vertical).An indicator line on the right side of the CSLO-like projection image shows the position corresponding to the B-scan of the movie frame momentarily in display in the lower part of the image.Recognizable features in the B-scans are the ILM (uppermost bright line), layers within the retina, and the RPE (bright line at the posterior boundary of the retina).As the indicator line approaches the central area, the foveal depression can clearly be observed.ILM and RPE are visible throughout the entire scan field.This allows the derivation of retinal thickness maps.

Fig. 4 .
Fig. 4. (717 kB) Frame no.33 of movie showing software derived B-scans through the fovea of a healthy volunteer.Same data set as in fig. 2. The upper part of the figure shows an SLO-like projection image (10°(x) x 10°(y)) of the foveal area, the lower part shows the B-scan corresponding to the actual movie frame (10°(x) x 1.05 mm(z)).The red indicator line at the right hand side of the projection image indicates the y-position corresponding to the actual movie frame.

Figure 5
Figure 5 shows a retinal thickness map derived from a 3D data set recorded with a 15° x 15° scan field in the macular area.The distance ILM-RPE is false color coded and shown as a function of position.The foveal depression near the center is clearly recognized.With increasing radial distance from the fovea, retinal thickness increases up to a distance of ~ 3 -4° (~ 1 mm), further at the periphery there is a slight decrease of retinal thickness.

Fig. 5 .
Fig. 5. Retinal thickness map (distance ILM -RPE) of the macular area of a healthy human volunteer derived from 3D OCT data set.Image size: 15° x 15°.Color bar: retinal thickness in µm.
shows a movie of B-scans derived by software from such a 3D data set.The scan field size was 10° x 10° (x-y) x 1.05 mm (z).The large squared part of the figure (left side) shows again the CSLO-like projection image, the B-scan movie frames (orientation y-z, left: inner segments, right: outer segments of the eye) are shown at the right hand side of the figure.An indicator at the bottom of the figure marks the position of the respective B-scan.The retinal nerve fiber layer (first bright layer from left in B-scans), the RPE/choriocapillaris complex (second bright layer from left in B-scans), the excavation of the optic nerve head, and blood vessels can clearly be observed.The change of these structures with position allows a 3D reconstruction.

Fig. 6 .
Fig. 6. (772 kB) Frame no.35 of movie showing software derived B-scans through the optic nerve head of a healthy volunteer.The left part of the figure shows an SLOlike projection image (10°(x) x 10°(y)) of the nerve head area, the right part shows the B-scan corresponding to the actual movie frame (10°(y) x 1.05 mm(z)).The red indicator line at the bottom of the projection image indicates the x-position corresponding to the actual movie frame.

Figure 7
Figure 7 shows another view of the 3D data set.The projection image and B-scans derived along x-z and y-z planes at selected cursor positions (indicated by white lines) are shown.This figure clearly indicates the possibilities of obtaining B-scan images at arbitrary locations and orientations from a 3D data set.

Fig. 7 .
Fig. 7. 3D OCT data set of the optic nerve head of a healthy volunteer (same data set as in fig.5).Bottom right: SLO-like projection image.Top: y-z (10° x 1.05 mm), and bottom left: x-z (10° x 1.05 mm), cross sectional B-scan images derived at indicated cursor positions from 3D data set by software.