Real time, high-speed image stabilization with a retinal tracking scanning laser ophthalmoscope (TSLO) enables new approaches to established diagnostics. Large frequency range (DC to 19 kHz), wide-field (40-deg) stabilized Doppler flowmetry imaging was demonstrated in initial human subject tests. The fundus imaging method is a quasi-confocal line-scanning laser ophthalmoscope (LSLO). The retinal tracking system uses a confocal reflectometer with a closed loop optical servo system to lock onto features in the ocular fundus and automatically re-lock after blinks. By performing a slow scan with the laser line imager, frequency-resolved retinal perfusion and vascular flow images were obtained free of eye motion artifacts. Normal adult subjects and patients were tested with and without mydriasis to characterize flow imaging performance.
© 2004 Optical Society of America
The retina is among the most highly vascularized and metabolically active tissues in the body. Like the central nervous system of which it is a part, it is also susceptible to ischemic injury. Degenerative diseases of the eye often have either hemodynamic consequences or causes, though many mechanisms remain unknown. Improved blood flow imaging diagnostics for retinal circulation and perfusion will aid the detection and management of eye disease, and research on retinal function and metabolism.
The retinal circulation apparent in images generated by scanning laser ophthalmoscopy (SLO) originates from the central retinal artery that passes through the optic nerve head before branching into superior, inferior, nasal, and temporal arteries, into many smaller vessels, and ultimately, into capillary networks. The underlying choroidal vessels and choriocapillaris beneath the retinal pigment epithelium (RPE) account for approximately 90% of the blood flow nourishing the retina . While rapid flow in the retinal vascular tree is readily visualized, the perfusion of the retina through the micro-vasculature on both sides of the RPE is critically important. For eye diseases such as diabetic retinopathy, macular degeneration, and glaucoma, these structures will exhibit early flow defects or the growth of new vessels triggered by metabolic distress and other factors .
Both the retinal and choriodal vessel diameters range from ~5 μm (in the retinal capillary bed and choriocapillaris) to ~0.4 mm (major vessels) . Flow rates range from local quasi-isotropic perfusion rates of tens of μm/s in the capillaries to pulsatile values of several cm/s in the arteries . This range of dimensions and flow parameters presents an extremely demanding diagnostic problem in terms of spatial resolution, field of view, and dynamic range. Dye angiography is a powerful tool for global visualization of retinal vessel topology, occlusions and, uniquely, leakage. Fluorescein and indocyanine green (ICG) dyes have different properties that emphasize different aspects of vascular physiology. However at present, early transit phase dye angiography of both types can provide at best only a fleeting glimpse of dynamic flow characteristics. Few methods accomplish dynamic blood flow imaging non-invasively (i.e., without dyes); fewer still quantitatively; and none with the wide field, high resolution, and dynamic range to characterize retinal hemodynamics globally. Reproducibility is essential for longitudinal studies, and also for sensitive detection of functional correlations with local neuronal activity or pharmacological effects. In short, despite decades of research and the introduction of several advanced systems for measurement of blood flow, retinal blood flow Doppler imaging diagnostics have not yet achieved the clinical prominence that retinal biology would seem to justify.
1.1 Doppler and speckle imaging
When imaging biological tissues with lasers, the phenomenon of speckle is a necessary consequence of coherent illumination. The superposition of scattered photons for an extended source produces a net wavefront at the receiving aperture that varies in amplitude and phase. For imaging instruments whose signal-to-noise ratio is well above the shot noise limit, speckle is the dominant source of the large fluctuations in apparent reflectivity and the granularity in captured images (depending on the degree of confocality). A stationary ensemble of scatterers will produce a stationary speckle pattern. Such variations are not intrinsic reflectivity variations and so from a static imaging viewpoint may be regarded as noise, with concomitant reduction of image contrast and spatial resolution. When particles in motion are of interest, however, the scattered light also has imposed Doppler frequency shifts dependent upon the scattered wave-vectors and velocities of the particles. The frequency content of the imaged light can be measured and velocities inferred. In this sense, in living biological tissue speckle can be regarded as a contrast agent enabling the visualization of dynamic processes [5–9].
Speckle interferometry and its related imaging technologies exploit the temporal characteristics of fluctuations that contain information about the motion of particles within an optically probed or imaged volume. The DC or zero frequency component of the temporal spectrum at an image pixel contains the time-averaged power during the observation, and therefore asymptotically, the intrinsic incoherent reflectivity or the “speckle-free” image. The AC component contains quantitative measures related to particle number density and velocity distributions within the probed volume element (voxel).
Multiple scattering within image volumes gives rise to some complex and often counterintuitive characteristics for Doppler signals. Red blood cells are strongly forward scattering in the near infrared and most of the scattered light per interaction falls within a sharply peaked cone with ~6 deg half angle . The result is usually a distribution of Doppler frequencies even for a single well-defined velocity vector. Direct backscatter from flowing blood is generally quite a weak component of the signal, so the usual conception of the Doppler signal may be misleading. Mainly, light forward scattered by blood is subsequently backscattered by denser tissues. This dual-scatter enables velocities perpendicular to the incident beam to contribute to the Doppler signal, but can result in Doppler signals whose spatial origin is ambiguous. At the largest scales of arteries and veins, the velocities are large with a single well-defined direction. At the capillary or perfusion scale, velocities are small, perhaps with multiple flow directions within a single voxel. Multiple light scattering can cause these fine scales to lose contrast in flow images based upon Doppler frequency shifts. However, the total Doppler signal power per unit of imaged moving tissue volume is approximately preserved .
The first applications of Doppler methods to retinal blood flow diagnostics were developed by Riva et al. [11,12]. In early work, a single laser beam was focused on the retina, and the flow in retinal vessels and capillary perfusion was found to be measurable and quantifiable with laser Doppler flowmetry. Later, two imaging approaches emerged using CCD fundus images (laser speckle imaging or flowgraphy) [13,14] and flying-spot confocal SLO devices such as scanning laser Doppler flowmetry (e.g. Heidelberg Retinal Flowmeter, HRF) [15,16]. Concurrently, color Doppler optical coherence tomography (OCT) or optical Doppler tomography (ODT) was found to provide local anatomical detail with velocity information when blood moves parallel to the probe beam. Such measurements can be difficult to interpret for complex vessel topology [17–21]. Most recently, the remarkable capabilities of high-speed, spectral domain ODT (SDODT) for blood flow measurement have been described . The improvement in retinal flow visualization has been considerable. However, all of these approaches trade resolution, field-of-view, dynamic range, velocity component sensitivity and Doppler frequency range against scan speed and system noise. The motion of the eye, the cardiac and respiratory rhythms and other effects render the lowest frequencies virtually uninterpretable. Low data rates and high-frequency aliasing render the highest flows inaccessible and/or inaccurate.
1.2 Stabilized Doppler Flowmetry (SDF)
Most scanning imaging technologies suffer from practical limitations in the living eye at eye-safe light levels: scan areas or volumes and scan times are restricted by eye motions or other registration issues that corrupt data and are generally not correctable by post-processing. In other words, at present almost all scanning imaging operations must fit within a relatively brief window in time - typically the order of a second, approximately the mean time between small saccades. This severely impacts the size of the measured fields and the trade-offs that must be made in data quality. The resulting difficulties of flow quantification, velocity range, sensitivity, dynamic range, and field of view have not yet been overcome, especially for correlation of precise information about local anatomical features with the wide-field angiographic data familiar to clinicians.
In order to address some of these issues, we have previously combined two technologies, the line-scanning laser ophthalmoscope (LSLO)  and the high-speed retinal tracking system  yielding a tracking scanning laser ophthalmoscope (TSLO) [24,25]. The LSLO images the retina by scanning a line of laser light which is confocal with a linear sensor array. The LSLO is therefore neither a true SLO nor a fundus camera, while possessing some of the advantages of both. These include efficient rejection of stray light, and high speed read-out with no scanning on one axis (along the line). The LSLO provides significant advantages for Doppler imaging. For each of these methods, it is interesting to note the contributions to the image signals of light scattered from (z) planes above and below the focal plane. For the SLO, the out-of-focus contributions fall of as 1/z2 which provides its intrinsic optical sectioning capability. For the fundus camera with flood illumination, there is essentially no rejection of out-of-focus contributions (no z-depndence), which accounts for its poor contrast. Finally, the LSLO exhibits an approximate 1/z dependence in its sensitivity to out-of-focus light. We refer to this intermediate behavior as “quasi-confocality”.
In the present investigation, the TSLO is configured for wide-field, large dynamic range stabilized Doppler flowmetry (SDF). Because of the high fidelity retinal tracking function, most eye motion artifacts are eliminated. In this way the temporal spectrum at each pixel is obtained. High contrast, low-noise, frequency-resolved retinal microvasculature and perfusion maps are thus visualized in a new dye-free angiography mode. With this novel imaging approach and advanced user interface features, the accessible Doppler frequency range for blood flow speeds has been significantly expanded both upwards and downwards, while eliminating confounding imaging artifacts and some limitations of earlier methods.
At first sight, the addition of real time, high-speed retinal image stabilization appears merely to be a convenience. While retinal tracking certainly simplifies some current clinical diagnostic protocols, and virtually eliminates registration issues by stabilizing relative to fixed retinal landmarks, its real contribution is potentially more fundamental. Tracking enables the duration of any given scan or scan sequence to be effectively indefinite. The familiar continuous spatial scan dimensions are the A-scan (z axis), B-scan (x or y-axis), and C-scan (y or x-axis). Precision stabilization robustly adds a new scan dimension to clinically practical diagnostics: the T-scan (the time axis). Any spatial and temporal scan sequence can be specified with combinations of these: e.g., a BC scan is an xy-plane; a BT scan is a single line re-scanned in time.
2. Materials and methods
The TSLO served as the platform for the SDF blood flow experiments and has been described in detail elsewhere [24,25]. A 25-μW, 880-nm LED tracking beam is dithered on a retinal feature - usually the lamina cribrosa in the optic nerve head - and detected with a confocal reflectometer. The processed reflectance signals were used to create error signals proportional to eye movements and fed back to an x-y galvanometer-driven mirror pair. The TSLO is meant to operate non-mydriatically but does not include dynamic focus capability. The required pupil size is 3mm or less, depending upon the magnification chosen. For some subjects, the long duration scans used in SDF are greatly facilitated by mydrasis which minimizes accommodation. A 300-μW, 780-nm laser illumination beam was fanned out on the retina and detected confocally with a linear image sensor. The transverse resolution is approximately 40 μm. The estimated depth of focus is approximately 1mm for the quasi-confocal LSLO imaging, but this parameter is somewhat arbitrary and is not the same as the usual confocal parameter for SLOs. The videos presented in this paper were acquired with a linear detector with a line rate of 7.5k lines/sec (Physical Sciences Inc.), which precluded measurement of the peak velocities in large vessels but allowed for higher resolution images of retinal perfusion. An alternate image sensor with a line rate up to 39k lines/sec (Fairchild Imaging Inc.) was also used to measure peak arterial velocity. In the optic nerve head, where the major vessels turn, the peak Doppler frequencies may remain above the bandwidth limit of this instrument. In the conventional LSLO imaging mode, the line is swept over the retina at a rate that can exceed 60 frames/sec.
An algorithm has been implemented to automatically re-lock the system after subject blinks. The subject is instructed to look at a fixation target. If lock is lost, the tracking system will wander or slew rapidly depending on bias settings because the closed loop-gain of the servo is high. When the tracking system detects an out-of-bounds target reflectance or tracking angle, tracking is suspended. After a brief delay to allow for the blink to be completed, the tracking mirrors are re-positioned and a re-lock command is issued. At present, we depend upon the subject looking within a few degrees (± ~2 deg) of the fixation target which is the re-lock capture range. Figure 1 shows a video and position signals during tracking and re-lock through blinks.
The SDF technique introduces a new, fast, and efficient approach to Doppler image capture. By performing slow infrared line scans, tracking the eye and re-locking after blinks, frequency-resolved images of retinal flow are obtained. Figure 2 illustrates the technique in comparison to the HRF. The linear array is read out 512 times at one line on the retina and stored as a single BT image plane (i.e., y-line vs. time) before the next line is scanned. Each BT plane of the image cube therefore comprises an image of 512 spatial pixels by 512 or more temporal pixels. Since the linear sensor integrates the light collected, a nearly continuous temporal profile is obtained at each pixel for the duration of the BT scan. The line rates for the respective linear sensors give inter-line times of 133 μs and 26 μs, and the integration duty factor is approximately 80% of these values. This significantly diminishes the susceptibility to aliasing. In a slow C-scan, 512 such BT planes are captured across the fundus, taking 35 seconds to complete a full BTC image cube at 7.5K lines/s, or 6.7 seconds at 39K lines/sec. Two scans in immediate succession are now routinely performed, so that blanking that occurs during blinks can be overwritten by the valid data of the twin scan. Scans can be precisely repeated after any intervening interval by simply locking onto the same retinal feature.
The subject is asked to fixate on a bright green target with the fellow eye. A dull red raster is visible during normal TSLO imaging. When the SDF scan commences, a bright red vertical line can be seen moving slowly across the visual field (currently left to right, but there is no reason that it could not be always nasal to temporal). The passage of the line over the fovea can induce some momentary nystagmus and vergence/near triad responses which are undesirable. While the tracking system can compensate most of these motions, future systems will incorporate internal (non-stabilized) fixation to minimize these. The scan direction could also be superior/inferior, though this has not yet been tried.
After image cube acquisition, a fast Fourier transform (FFT) was applied to extract the full power spectrum at each image pixel. The frequency spectra were normalized by the DC value to remove intensity reflectance variation across the image. To create frequency videos, the spectra at each pixel were binned with several different windows, including one with a fixed size and one with octave increments (2,4,8…). The most common processing shown in the videos presented herein, used an overlapping moving window with an increment and halfwidth of 8 and 16 frames (117 and 234 Hz for the PSI camera), respectively. After binning, the individual frequency frames were linearly contrast-stretched between minimum and maximum image intensity in 8-bit image planes. The processing takes less than 3 minutes with a 1.8-GHz Pentium IV processor. The motivation for the video representation used here is to reveal the interesting transition of flow rates and vessel patterns from the smallest to the largest size scales, and is intended to evoke the visual effect of early phase angiography. Another aim is to present a novel means of interpreting this new data which more fully exploits the improved dynamic range and signal-to-noise ratio, beyond conventional “volume, flow and velocity” maps associated with the HRF.
In contrast to SDF with the TSLO, the HRF samples each pixel of each line at a rate of 4k lines/sec (Fig. 2b). A significant impact of this lower line rate - determined by the mechanical hardware - is that the Nyquist Theorem limits the maximum frequency detected to 2 kHz. Even more important is the fact that the duty cycle will be an extremely small percentage (1/256) contingent upon the number of pixels in the line. This detection approach based upon sampling leads to aliasing and consequently, velocity errors. Moreover, since tracking is not employed in the HRF, frequencies below 125 Hz are disrupted by motion and not processed or displayed. Eye motion also prevents acquisition of even moderately-sized and sampled retinal fields. The BTC scan dimensions of the HRF are 256×128×64. The TSLO with scan dimensions of 512×512×512 has 16 times the spatial resolution, and 4 times the temporal resolution. Depending on the linear sensor, the TSLO has 1.875 or 9.5 times the accessible peak Doppler frequency, and extends to as low as 14 Hz.
We utilized the TSLO in SDF mode for wide-field blood flow imaging and have performed SDF scans in seven normal subjects and four patients recruited from a retinal practice. Videos created from two normal subjects and two patients are shown below. The first frame of each video is the reconstructed fundus image created from the DC value. The videos are displayed at 10 frames/sec but each frequency bin can be viewed by stepping through frames in Windows Media Player or Apple Quicktime.
Figure 3 shows the blood flow video for a healthy 24-year old subject. The vertical line noise at the lowest frequency bin are caused by small amplitude (<1 pixel) transient tracking stability artifacts. In all images, dark vertical bands near the fovea (only in undilated subjects) are due to nystagmus invoked by occasional fixation on the scanning line as it passes across the fovea, causing some vignetting by small pupils. Such effects are adequately compensated by DC normalization. Young healthy subjects often show bright specular reflection, presumably from the nerve fiber, which are DC only and affect Doppler images through the normalization.
The foveal avascular region is clearly visible at low frequencies. In the lower frequency bins (f < 400 Hz), or the “perfusion range”, flow in the retinal capillary beds and the choriocapillaris creates a bright field that is shadowed by the overlying retinal vessels, which happen to be dark in these frequency bins. This implies that the choroidal contribution to the flow maps can be attenuated by the RPE in darkly pigmented eyes, and perhaps by the focus and depth of field of the quasi-confocal imaging system. In middle frequency bins (f = 400 – 1000 Hz), very small retinal vessels emerge, as well as pronounced peripapillary flow. At the highest frequencies (f > 1000 Hz), the small vessels gradually fade and only the signal from the largest vessels in the retina and choroid can be detected. Note also the change in the reflectance of the lamina cribrosa as the frequency increases.
A normal subject with lighter pigmentation shows much more choroidal flow detail in Fig. 4. This subject was highly myopic, accounting for the larger variation in focus across the field in the reconstructed fundus image. The foveal avascular region is again apparent at lower frequencies. At higher frequencies, vertical banding caused by pulsatile flow (at about 15 lines per beat) is quite strong in this subject in the choroid visible below the retinal circulation. It is interesting to observe that very little of this pronounced choroidal structure is visible in the reconstructed SLO image. Intense pulsatile flow occasionally gives arteries a beaded appearance. The perfusion of the choriocapillaris is undoubtedly contributing to the perfusion maps, but it does not overwhelm the retinal signal.
A patient shows some sign of inflammation associated with central serous chorioretinopathy indicated by blue arrows in the reconstructed fundus image in Fig. 5. However, perfusion appears to be normal in the low frequency frames of the video. Choroidal flow was difficult to observe but visualization of medium-sized vessels at middle frequencies was more pronounced in this darkly pigmented eye.
In the same patient, a higher-frequency (19 kHz), higher-magnification scan was acquired to determine the peak of laminar Poiseuille flow in the larger retinal arteries (Fig. 6). At frequencies greater than about 15KHz, the Doppler images appeared as featureless white noise at constant mean amplitude. This was taken to be the system noise floor and subtracted from the Doppler spectra. After noise correction and normalization, the average flow margin position (FWHM) for a small section of artery was measured at each frequency bin and the resultant data fit to a parabolic profile. At frequencies greater than 6 kHz, the flow signal was confined to a region smaller than the approximate resolution of the imaging system (~40 μm on the retina) and no further decrease in diameter was seen. The fit corresponds well with the frequency bin of the maximum resolvable signal in that vessel (8.3±0.6 kHz) and with previously reported measurements .
The estimated peak velocity, up, can be computed from the center frequency,fp as,
where λ is the laser wavelength, n is the refractive index of blood, and θB is the estimated forward scattering half-angle of blood (~6 deg). For the peak shown in Fig. 6, up = 4.6±0.4 cm/s (at the particular point in the cardiac cycle where the vessel was scanned).
Figure 7 shows the blood flow video from a 68-year old patient with macular degeneration. Perfusion anomalies associated with apparent pathology are evident in the low frequency images, but also a large area of reduced perfusion temporally. At present these effects are difficult to categorize and interpret. The binned images show significant detail, even though the DC fundus image is not optimally focused through the relatively optically poor ocular media in this subject. The detail of the retinal vessels indicates the possibility for diagnosis of retinal-choroidal anastomoses and retinal vascular anomalous complexes , associated with severe vision loss. Middle frequency peripapillary flow signal that was evident in all (younger) normal subjects was not visible in this patient.
Figure 8 illustrates an alternate representation of Doppler blood flow information with false color maps. To generate these maps, the transformed videos were averaged in three arbitrarily chosen ranges roughly representing capillary perfusion (f < 400 Hz), middle flow in arterioles and venules (f = 400 – 1000 Hz), and high flow in larger arteries and veins (f > 1000 Hz). The three bins were then DC-normalized, scaled and mapped into RGB colors, and displayed. Two different scaling techniques were employed. The first, shown in the upper row of Fig. 8, normalized and scaled each color plane individually (with the same linear contrast stretch in the videos above) so that the overall intensity range of each flow bin was matched in the composite image. This creates a black pixel where no flow at any frequency was present. Where high or low flow dominated, the pixel is more blue or red. This normalization and scaling scheme emphasizes vascualar anatomy, and can be used to compare structures within a Doppler image, but not from subject-to-subject because the Doppler band-to-band normalization is relative The second normalization technique is shown in the lower row of Fig. 8. In this representation, the maximum pixel value in each color plane is scaled to the average Doppler power per frequency bin. In this representation, the overall map color will be determined by which flow region dominated according to the true Doppler spectrum shape. Thus, the second normalization scheme can be used to make comparisons between subjects.
Figure 8 was created with data from the second normal subject and both patients presented previously. In the lightly pigmented normal subject, the higher choroidal flow (blue) is apparent in the nasal portion of Fig. 8a. In the patient with higher pigmentation, medium flow retinal vessels (green) are dominant (Fig. 8b). In both patients, the high pulsatile flow is also visible (Fig. 8b and c). Changes visualized with the second normalization technique (Fig. 8d–f) are dramatic in subjects with global blood flow defects (e.g., overall orange appearance of macular degeneration patient) compared to those without defect. However, care must be exercised in interpretation of this result. It is not yet known how poor image LSLO image quality in some patients affects the apparent distribution of Doppler power.
The partitioning of the flowmetry data into normalized frequency bins, which divides the images by vessel size and/or flow rate, is only useful with the low noise levels achieved with the SDF scans. The sharpness of the reconstructed DC fundus images is evidence of tracking fidelity during the long scans. The perfusion images (f < 400 Hz) have similar appearance and correspond to Doppler velocities less than approximately 2 mm/s. These images combine retinal and choroidal contributions.The remaining bins begin to reveal the various vessel size and flow speed echelons in the vascular tree, up to in-plane velocity component values of ~20 mm/s. The forward scattering angle through the blood in the larger vessels may be several times higher. In a large superficial retinal artery, we used a linear array sensor with a maximum line rate of 39k lines/sec to measure laminar Poiseuille flow and a peak velocity of ~5 cm/s. Depending on pigmentation and pathology, the larger vessels and highly pulsatile flow of the choroid can be clearly seen in some eyes, but less well in others.
The current approach provides whole retina information by taking advantage of the relatively large depth of field of the SDF (TSLO) optical system (~1 mm) and the fact that the line scanning system is only quasi-confocal and does not possess the full intrinsic sectioning capability of purely confocal flying spot scanning systems. While this mixes information from the choroid and retina, especially at low flow rates, it allows a single data set to be generated to characterize global ocular health. Alternate modes are possible to provide finer depth sectioning of the vasculature including stereo imaging with a dual detector and split pupil or low-coherence imaging techniques. These capabilities are being evaluated with the TSLO, but have not yet been applied to SDF scans.
Future work will include additional clinical testing, development of higher dynamic range systems, improved depth resolution, and cardiac synchronization for wide-field pulsatile flow characterization. The large dynamic range available with SDF will be used with higher magnification imaging to validate improved quantitative blood flow analysis methods, and to characterize local perfusion defects or neovascular growth.
This work was supported by NIH Grants EY11577, EY14106, and EY07624, and Physical Sciences Inc. The authors thank Scott Meyer for helpful discussions.
References and links
1. J. J. Weiter, R. A. Schachar, and J. T. Ernest, “Control of intraocular blood flow, I. Intraocular pressure,” Invest Ophthalmol. 12, 327–331 (1973). [PubMed]
2. A. Alm, “Ocular circulation,” in Alder’s Physiology of the Eye. Clinical Applications, W. M. Hart, ed. (Mosby-Year Book, St. Louis,1992).
3. M. E. Hartnett, J. J. Weiter, G. Staurenghi, and A. E. Elsner, “Deep retinal vascular anomalous complexes in advanced age-related macular degeneration,” Ophthalmol. , 103, 2042–2053 (1996).
4. A. E. Elsner, D. Bartsch, J. J. Weiter, and M. E. Hartnett, “New devices in retinal imaging and functional evaluation,” in Practical Atlas of Retinal Disease and Therapy, W. Freeman, ed. (Lippincott-Raven, New York,1998).
5. D. Gabor, “Laser speckle and its elimination,” IBM J. Res. Dev. 14, 509–514 (1970). [CrossRef]
6. A. F. Fercher and J. D. Briers, “Flow visualization by means of single-exposure speckle photography,” Opt. Commun. 37, 326–329 (1981). [CrossRef]
7. A. E. Ennos, “Speckle interferometry,” in Laser Speckle and Related Topics, J. C. Dainty, ed. (Springer-Verlag, Berlin,1984).
8. J. W. Goodman, “Statistical properties of laser speckle patterns,” in Laser Speckle and Related Topics, J. C. Dainty, ed. (Springer-Verlag, Berlin,1984).
9. J. D. Briers, “Speckle fluctuations and biomedical optics: implications and applications,” Opt. Eng. 32, 277–283 (1993). [CrossRef]
11. C. Riva, B. Ross, and G. Benedek, “Laser Doppler measurements of blood flow in capillary tubes and retinal arteries,” Invest. Ophthalmol. 11, 936–944 (1972). [PubMed]
13. N. Konishi and H. Fujii, “Real-time visualization of retinal microcirculation by laser flowgraphy,” Opt. Eng. 34, 753–757 (1995). [CrossRef]
15. G. Michelson, B. Schmauss, M.J. Langhans, J. Harazny, and M.J.M. Groh, “Principle, validity, and reliability of scanning laser Doppler flowmetry,” J. Glaucoma 5, 99–105 (1996). [CrossRef] [PubMed]
16. G. Michelson, J. Welzenbach, I. Pal, and J. Harazny, “Functional imaging of the retinal microvasculature by scanning laser Doppler flowmetry,” Int. Ophthalmol. 23, 327–335 (2001). [CrossRef]
18. J. A. Izatt, M. D. Kulkarni, S. Yazdanfar, J. K. Barton, and A. J. Welch, “In vivo bidirectional color Doppler flow imaging of picoliter blood volumes using optical coherence tomography,” Opt. Lett. 22, 1439–1441 (1997). [CrossRef]
19. T. G. van Leeuwen, M. D. Kulkarni, S. Yazdanfar, A. M. Rollins, and J. A. Izatt, “High-flow-velocity and shear-rate imaging by use of color Doppler optical coherence tomography,” Opt. Lett. 24, 1584–1586 (1999). [CrossRef]
20. S. Yazdanfar, A. M. Rollins, and J. A. Izatt, “Imaging and velocimetry of the human retinal circulation with color Doppler optical coherence tomography,” Opt. Lett. 25, 1448–1450 (2000). [CrossRef]
21. B. R White, M. C. Pierce, N. Nassif, B. Cense, B. H. Park, G. J. Tearney, B. E. Bouma, T. C. Chen, and J. F. de Boer, “ In vivo dynamic human retinal blood flow imaging using ultra-high speed spectral domain optical coherence tomography,” Opt. Express 11, 3490 (2003), http://www.opticsexpress.org/abstract.cfm?URI=OPEX-11-25-3490. [CrossRef] [PubMed]
22. D. X. Hammer, R.D. Ferguson, T. Ustun, G. Dadusc, and R.H. Webb, “Hand-held digital line-scanning laser ophthalmoscope (LSLO),” in Ophthalmic Technologies XIV, F. Manns, P. G. Söderberg, and A. Ho, eds., Proc. SPIE5314, 161–169 (2004).
23. R. Daniel Ferguson, “Servo tracking system utilizing phase-sensitive detection of reflectance variation,” U.S. Patents #5,767,941 and #5,943,115.
24. D. X. Hammer, R. D. Ferguson, J. C. Magill, M. A. White, A. E. Elsner, and R. H. Webb, “Image stabilization for scanning laser ophthalmoscopy,” Opt. Express 10, 1542–1549 (2002), http://www.opticsexpress.org/abstract.cfm?URI=OPEX-10-26-1542. [CrossRef] [PubMed]
25. D. X. Hammer, R. D. Ferguson, J. C. Magill, M. A. White, A. E. Elsner, and R. H. Webb, “Compact scanning laser ophthalmoscope with high speed retinal tracker,” Appl. Opt. 42, 4621–4632 (2003). [CrossRef] [PubMed]