2.1.

mLSO System

The mLSO retinal imaging and treatment system was designed around PSI’s core tracking scanning laser ophthalmoscope (TSLO) technology,^7,8 which includes retinal tracking (RT) and line scanning ophthalmoscope (LSO) hardware. A schematic and SolidWorks opto-mechanical model of the optical head is shown in Fig. 2. The active retinal tracking works by directing a low power beam at a target in the retina (usually the optic nerve head) to detect and compensate for eye movements.^7–9 The tracking beam is dithered at a high frequency (16 kHz) with resonant scanners and detected with a confocal reflectometer. The resultant signal is phase processed to generate $x$- and $y$-error signals, which are fed into the closed-loop galvanometers to stabilize the imaging and PDT beams on fixed retinal coordinates through all eye motion. The retinal tracker hardware includes a phase-locked resonant scanner pair and the optics that constitute the confocal reflectometer. The LSO is a quasi-confocal, line-scanning, wide-field retinal imaging platform.¹³ The LSO hardware includes a cylindrical lens (CL), a single scanning galvanometer, and a linear array detector (LAD, e2v Inc.). The imaging optics consist of custom objectives (O1: scan objective, O2: detector objective) and a front ophthalmoscopic lens (OL). The instrument can be configured with several different ophthalmoscopic lenses (e.g., Volk 40D, 66D, 78D) to achieve a variety of retinal field sizes.

Fig. 2

Schematic (a) and opto-mechanical model (b) of the mLSO optical head.

To the core TSLO hardware, we have made four additions to enable multifunctional capabilities. First, a compact visible (490-nm) laser and coupling optics (filters, fiber-couplers) were added to the instrumentation for fluorescein excitation. Second, an illumination port was added to the optical head to couple the fluorescein excitation laser beam into the optical beam path. Third, a computer-controlled, motorized filter stage was inserted in front of the camera to switch rapidly between reflectance, ICGA, and FA modes for interleaved operation and display. Fourth, a treatment beam port was added for a commercial off-the-shelf PDT laser.

The mLSO has been configured with sources, filters, and dichroic beamsplitters specifically to collect near-infrared confocal reflectance images and ICG and fluorescein angiograms. A low-pass ($\sim 600\mathrm{nm}$ cutoff) dichroic beamsplitter (D1) directs the 490-nm laser source used for FA into the optical path. A high-pass ($\sim 900\mathrm{nm}$ cutoff) dichroic beamsplitter (D2) combines the tracking and imaging beams. A custom band-pass filter (Omega Optical Inc.) at 690 nm (D3) directs the PDT laser into the optical path. The Zemax optical design showed good performance of the optics at both 490 and 795 nm.

Two illumination beams are combined into the main imaging path: the near infrared (795 nm) light for confocal reflectance imaging and ICG angiography, and the 490-nm light for fluorescein angiography. A superluminescent diode (SLD) with 760-nm center wavelength (Exalos Inc.) was also used in place of the 795-nm LD source to generate some LSO images. SLDs produce images with less speckle noise, but no SLDs are currently available at the ICG excitation peak. An excitation filter (751 to 798 nm) prevented illumination light from reflecting back into the detector in the ICG emission band. The 490-nm laser source (Cobolt Calypso laser) is an extremely compact laser with an output power of $\sim 25\mathrm{mW}$. A collimator affixed to a tip-tilt mirror mount is used to couple the light into a single-mode fiber connected to the optical head. Fixed neutral density filters prevent light exposure to the patient above the ANSI safety limits. The system uses a 915-nm superluminescent diode (SLD) for retinal tracking.

The PDT laser added to the system is manufactured by Carl Zeiss Meditec Inc. (Visulas 690 s and Visulink PDT/U) and has an output wavelength of 690 nm. The laser is designed to be mounted on several different manufacturers’ slit lamps. The PDT laser consists of a control box, an optical port, and a foot switch. The controller has a touch screen to allow the user to change parameters such as light dose and exposure time.

A filter stage was designed to switch rapidly between emission filters (EF) for reflectance, ICGA, and FA using a Quickshaft linear actuator (Faulhaber). The actuator has a top speed of 1.9 m/s and uses Hall sensors to achieve a resolution of 6 µm, a repeatability of 40 µm, and a precision of 120 µm over the positioning range. Three filter slots house emission filters for ICGA and FA and Schott glass for confocal reflectance LSO imaging. The FA emission filter passes visible light from 507 to 562 nm, and the ICGA emission filter passes NIR light from 819 to 863 nm. The filters were made the same optical thickness to prevent path length differences that would affect image focus when switching filters. The linear actuator is driven by commands from the PC (serial interface) with a small motion controller board integrated into the instrumentation box. In addition to switching emission filters, an FA illumination source shutter assembly is also mounted to the actuator so that when the filter stage is in any position other than FA, it will block the 490-nm laser beam from entering the eye.

The mLSO also includes an integrated LED-array fixation target (fellow-eye) and the PDT laser delivery component. Fellow-eye fixation may introduce eye position errors, but the fixation target is only intended to coarsely position the subject’s gaze direction to lock and re-lock tracking on particular targets.

The final design for the optical head includes the dual illumination port, which fits entirely within the external enclosure of the standard TSLO optical head, the Maxwellian PDT port, which includes relay optics (lens and turning mirror), galvanometer-driven mirrors to position the PDT beam anywhere on the retina, and the Zeiss PDT laser mounted to the side of the optical head, and the angiography filter sub-assembly, composed of the linear actuator and the filter slide mounted to the same bracket used to mount the PDT laser.

All system electronics and instrumentation are contained in a single instrumentation box. The instrumentation box and system computer are housed in a compact 19-inch rack. The electronics are built around two custom-printed circuit boards and a motherboard and field programmable gated array (FPGA)-based tracker controller board; they also include several commercial OEM driver electronics boards. The primary function of the tracker controller board is to run the feedback loop independently between the detector and galvanometers in real time.

The motherboard has hardware to interface with the OEM driver boards for the galvanometers (Cambridge Technology), resonant scanners (EOPC Inc.), and laser diodes and SLD sources (Exalos and QPhotonics). It also includes integrated LD/SLD thermo-electric coolers and drivers, integrated custom silicon detector electronics (for the tracking reflectometer), power regulation electronics, and header slots to mount and communicate with the custom retinal tracker control board. The tracker control electronics are composed of two boards: a Xilinx Virtex-4 FPGA mini-module board and an analog front-end (AFE) that includes a six-channel 250-kHz analog-to-digital converter (ADC), a four-channel, 100-kHz digital-to-analog converter (DAC), digital input/output (DI/O) lines, and serial communication with the host PC. The detector signal is sampled at 208 kHz, and the eye position data are transferred to a PC at a decimated rate of 1 kHz. Photographs of the mLSO system and optical head are shown in Fig. 3. Key specifications for the mLSO components are highlighted in Table 1.

Fig. 3

Photograph of mLSO system. (a) Complete system. (b) Optical head.

Table 1

Key specifications for mLSO components.

2.2.

Subjects

The mLSO system was tested in two human subject studies. The first, performed at PSI, was on a small number of subjects without retinal diseases ($n=3$) to validate, characterize, and optimize LSO imaging and retinal tracking. The second, performed at Boston University Medical Center (BUMC), was a pilot clinical investigation on 10 subjects with retinal disease and one without retinal disease. The BUMC study was designed to demonstrate line scanning angiography. The diseases examined included diabetic retinopathy (DR, with macula edema), central serous chorioretinopathy (CSC), sickle cell retinopathy, and a possible case of macular degeneration. IRB protocols were submitted and approved by New England IRB for the initial PSI study and the BUMC IRB for the pilot clinical study. All subjects gave informed consent prior to imaging.

2.3.

Imaging Protocol

For the PSI study, LSO images were acquired, and the system optics were aligned. No angiography was performed, but artificial fluorescent targets were used to verify operation of the FA and ICGA modes prior to the BUMC study. For validation and optimization of retinal tracking, the subjects were asked to gaze at the illuminated light on the LED array (spacing $\sim 2.5\mathrm{deg}$.) during a patterned illumination sequence while tracking was active. The performance of the automatic blink/re-lock algorithm was also verified.

The BUMC pilot clinical study was designed to perform complementary imaging on patients undergoing standard fundus camera angiography to verify mLSO performance. The mLSO system was set up in the clinical space immediately adjacent to a TopCon Fundus Photography System (model #TRC50X). Initial clinical testing at BUMC was limited to verification and optimization of the angiography imaging modes alone. For the subject without retinal disease and the subject with CSC, we obtained FA and ICGA sequences during the initial dye-infusion stage. In all other patients, the subjects were first imaged on the TopCon system during dye injection and then moved over to the mLSO after a few minutes. This procedure somewhat limited the visualization of early stage dye infusion, when the dye concentration is largest; however, it did not hinder resolution of late stage leakage.

The mLSO was configured to acquire FA and ICGA images at 1 frame per second (fps). The power at the cornea was $\sim 0.6\mathrm{mW}$ and $\sim 1\mathrm{mW}$ for the individual FA (490 nm) and ICGA (795 nm) sources, which were not used simultaneously. Because the 490-nm source is extremely bright to the patient, we configured the mLSO imager to modulate this source by collecting three frames at 490 nm, followed by seven LSO frames for every 10 frames acquired. The NIR LSO beam power was reduced to $<50-100\mu W$ for the longer integration time that accompanied the lower frame rate (the LSO is typically run at 15 to 60 fps). The software was configured to switch automatically between appropriate bandpass filters, trigger the ICGA shutter, and change to image brightness scaling to accommodate the different FA and LSO reflectance modes during acquisition of this sequence. For ICGA, the videos were acquired continuously.

Because the mLSO and the TopCon always imaged an eye at different time points with respect to the dye infusion, a direct comparison between the two devices was not possible. Nonetheless, we were able to move some subjects fast enough between instruments to get a general idea of the appearance of early and late phase angiograms on both instruments. Also, while we do make a comparison in the results that follow, the principal aim of the investigation was to demonstrate the feasibility of line scanning angiography and not to compare it to current clinical techniques.

3.1.

LSO Imaging

Preliminary system characterization was performed on artificial targets and resolution charts. The wide-field imager had good flatness across the field. Over the $\sim 48$-deg field (13.9 mm), the system can resolve objects with a resolution of $\sim 15\mathrm{\mu m}$ (image pixel $\text{size}=13.5\mathrm{\mu m}$ referenced to the retina). The PDT beam power measurements also verified expected losses through the system of $\sim 25\%$. The adjustment range for the PDT beam diameter is 0.8 to 4.9 mm. The spot size measurements resulted in an average error across this range of $\sim 3.5\%$, indicating proper beam alignment and relay lens positioning.

The initial human subject testing also showed good preliminary results in terms of imaging performance and retinal tracking. A comparison of single and composite averaged images with and without tracking for one subject are shown in Fig. 4. The LSO images are nearly equivalent to flying-spot confocal images in terms of contrast, reduced scatter, and depth sectioning. Because this subject was a good fixator, even the non-tracking composite image [Fig. 4(c)] shows coarse retinal features. However, in terms of the resolution of fine features (see inset of Fig. 4), such as capillaries around the foveal avascular zone (FAZ), the tracking composite image [Fig. 4(b)] has nearly the same appearance as the single frame [Fig. 4(a)]. In contrast, features are doubled and blurred in the non-tracking average.

Fig. 4

mLSO imaging and tracking performance for one PSI subject (male, normal vision, age 25, excellent fixator). (a) Single LSO frame. (b) 50 frames ($>3\sec$.) with tracking. (c) 50 frames without tracking. Inset for each show $2\times$ magnification of region between fovea and ONH. Foveal pit, nerve fibers, and small capillaries around FAZ are clear in single and tracked frame but blurred in non-tracked frame. Bright reflections near the macula and ONH are from nerve fiber layer. Scale $\mathrm{bar}=3\mathrm{deg}$.

3.2.

Retinal Tracking

Retinal tracking in two subjects during eye movement tasks is shown in Fig. 5. The composite (i.e., averaged) image and corresponding eye movement positions are shown. The accompanying videos (Videos 1 and 2) show real-time dynamic tracking during the tasks. Retinal tracking was robust on the small number of subjects tested. Tracking performance was similar to other recent reports.^10–12

Fig. 5

Retinal tracking performance for eye movement tasks in two PSI subjects. Shown are composite averages (a) and (c) and eye movement charts (b) and (d). Subjects were asked to gaze at patterned illuminated fixation targets with spacing of $\sim 2.5\mathrm{deg}$ during tracking. (a) Composite image generated from 142 video frames ($\sim 10\sec$) acquired during eye movement sequence for subject 1. Overlay shows display as viewed in the acquisition software with tracking beam position in the ONH, fixation target position, and 2-deg scale bar. (b) Eye position data for box/diamond task. (Video 1, MOV, Quicktime, 6.0 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.2.026008.1]. (c) Composite images generated from 166 video frames ($\sim 11s$) acquired during eye movement sequence for subject 2. (d) Eye position data for cross task. (Video 2, MOV, Quicktime, 7.1 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.2.026008.2]. Scale bar in (c) $=3\mathrm{deg}$. Subject 1: 38-year-old male with –7D of myopia, imaged with contacts. Subject 2: 49-year-old emmetropic and presbyopic male.

3.3.

Line Scanning Angiography

In the results that follow, the images from the mLSO are presented, along with images from the TopCon system. We tried to choose frames that were nearly equivalent in terms of dye infusion times, but as mentioned above, this was often difficult because the subjects were imaged at different times on each device. The optical resolution of the TopCon is better than the present mLSO e2v line-camera implementation ($1024\times 1024$ pixels versus $1900\times 1472$ across a similar field size), which should be considered in comparing the images presented. The mLSO can be designed to accommodate higher pixel density.

The subject without retinal disease was imaged first on the mLSO and then on the TopCon with both FA and ICGA. The results are shown in Fig. 6. The upper row shows the TopCon images, and the lower row consists of the mLSO images. The TopCon images include a red-free image (a), a late phase FA (b), and a late phase ICGA (c). The mLSO images include the NIR reflectance image (d), an early phase FA (e), and an early phase ICGA (f). Videos 3 and 4 show the accompanying FA video (at 15 fps). Video 3 shows the raw FA image acquisition with LSO and FA images interleaved. Video 4 shows the FA images extracted from Video 3. The mLSO FA image shows good resolution of all the retinal vasculature. As a consequence of the mLSO confocality, the choroidal vessels in the FA image are not as visible as in the TopCon. However, the mLSO ICGA image shows good contrast of the choroidal vessels.

Fig. 6

Comparison of single fundus and fluorescence images from a human subject without retinal disease. Top row (a) to (c): Images from TopCon fundus photography system. Bottom row (d) to (f): Images from mLSO system. (a) Red-free image. (b) Late-phase FA image (04:06 after injection of dye). (c) Late-phase ICGA image. (d) LSO confocal reflectance image. (e) Early-phase FA image (00:46). (f) Early-phase ICGA image (00:48). Scale $\mathrm{bar}=3\mathrm{deg}$. Field of view is $\sim 50\mathrm{deg}$ for both imagers. (Video 3, MOV, Quicktime, 15.7 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.2.026008.3] shows raw FA video for entire dye infusion cycle corresponding to (e). (Video 4 MOV, Quicktime, 14.2 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.2.026008.4] shows FA frames extracted from Video 3.

The subject with central serous chorioretinopathy (CSC) was imaged first on the mLSO and then on the TopCon with both FA and ICGA. The results are shown in Fig. 7 with a similar arrangement of panels as Fig. 6. Video 5 shows the accompanying ICGA video (at 15 fps). In this subject, the area of the disease is clear in the FA images, with hyper-fluorescence in lesions immediately adjacent to the disc on the temporal side. The ICGA images show good contrast and hypo-fluorescence in the choroidal vasculature in the diseased region.

Fig. 7

Comparison of single fundus and fluorescence images from a human subject with stable central serous chorioretinopathy (CSC). Top row (a) to (c): Images from TopCon fundus photography system. Bottom row (d) to (f): Images from mLSO system. (a) Red-free image. (b) Late-phase FA image. (c) Late-phase ICGA image. (d) LSO confocal reflectance image. (e) Early-phase FA image (01:02). (f) Early-phase ICGA image (00:39). Scale $\mathrm{bar}=3\mathrm{deg}$. Field of view is $\sim 50\mathrm{deg}$ for both imagers. (Video 5, MOV, Quicktime, 18.2 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.2.026008.5] shows entire dye infusion cycle corresponding to (f).

For the CSC subject, we examined the disease features in the FA and ICGA images in more detail. Figure 8 shows the images zoomed into the diseased region. The TopCon images are shown on the left, and the mLSO images are shown on the right. The top row presents the FA images, and the bottom row consists of the ICGA images. The features resolved in both are indicated by white arrows while the features resolved in the TopCon but not the mLSO are indicated by black arrows. The mLSO was able to resolve all the fine features of the disease areas of hyper- and hypo-fluorescence and many of the small capillaries. It could not resolve some of the finest capillaries, as expected due to the resolution of the camera. The mLSO ICGA images show higher contrast imaging of the choroidal vessels from rejection of scattered light and resolution of nearly all of the same disease features.

Fig. 8

Detailed comparison of FA and ICGA images from TopCon and mLSO imager in the region of CSC lesions. (a) TopCon FA image. (b) mLSO FA image. (c) TopCon ICGA image. (d) mLSO ICGA image. White arrowheads indicate features (capillaries, hypo-pigmentation, hyper-pigmentation) resolved in both imagers. Black arrowheads indicate features (primarily the smallest capillaries) resolved in the TopCon imager but not the mLSO. Scale $\mathrm{bar}=3\mathrm{deg}$.

A subject with diabetic retinopathy (DR) was imaged first on the TopCon and then on the mLSO. Although the subject was imaged at different time points, we were able to assemble comparisons at similar times for the early, mid-, and late phase of the dye infusion. Figure 9 shows the FA images that were collected. The multiple lesions around the arcades on the nasal side of the disc were caused by previous laser treatments. The mLSO images show good resolution of dye, including late-phase leakage in lesions between the fovea and the disc.

Fig. 9

Comparison of single fundus and FA images at several time points after dye infusion in a human subject with diabetic retinopathy and macular edema. Top row (a) to (d): Images from TopCon fundus photography system. Bottom row (e) to (h): Images from mLSO system. (a) Red-free image. (b) Early-phase FA image (01:30). (c) Mid-phase FA image (09:23). (d) Late-phase FA image (17:32). (e) LSO confocal reflectance image. (f) Early-phase FA image (03:56). (g) Mid-phase FA image (05:56). (h) Late-phase FA image (14:04). Scale $\mathrm{bar}=3\mathrm{deg}$.

A detail comparison of the early phase images in a region with multiple laser lesions just superior to the fovea is shown in Fig. 10. The detail is labeled similar to Fig. 8. The mLSO imager resolved the same areas of hypo- and hyper-fluorescence in the laser lesions and nearly all the same capillaries as the TopCon system. It can be misleading to attempt quantitative comparisons of angiography images between the two very different imagers, especially in this case, where the images were collected at different times after dye infusion. However, some quantitative information can be gained by comparison of line profiles across a small capillary as shown in Fig. 10(c). While both imagers were able to resolve this vessel (width of 2 pixels for both), the background noise floor for the TopCon imager was smoother, indicating a higher signal-to-noise ratio (SNR). The standard deviation of 20 pixels immediately surrounding the peak was 3.3 for the TopCon, compared to 9.8 for the mLSO. The vessel contrast (difference between peak and floor) was slightly higher in the mLSO (68.6 versus 45.8). However, this analysis was done on processed images. A more informative comparison performed on the raw output was not possible because the raw output from the TopCon was not accessible.

Fig. 10

Detailed comparison of FA and ICGA images from TopCon and mLSO imager in the superior macula in a subject with DR. (a) TopCon FA image. (b) mLSO FA image. (c) Line profiles across capillary on the left side of the images. White arrowheads indicate features resolved in both imagers. Black arrowheads indicate features resolved in the TopCon imager but not the mLSO. Scale $\mathrm{bar}=3\mathrm{deg}$.

The subject with sickle cell retinopathy was imaged first on the TopCon and then on the mLSO (Fig. 11). This subject had a normal macula but large peripheral lesions. We were able to compare images from the mLSO and TopCon at similar points. Extremely large hyper-fluorescence from dye leakage was seen in the late-phase angiograms.

Fig. 11

Comparison of single fundus and FA images in a subject with sickle cell retinopathy. Top row (a) to (d): Images from TopCon fundus photography system. Bottom row (e) to (h): Images from mLSO system. (a) Red-free image. (b) FA image of normal macula (00:54). (c) FA image of peripheral lesions (02:29). (d) Late-phase FA image of peripheral lesions (06:42). (e) LSO confocal reflectance image of peripheral lesions. (f) FA image of normal macula (04:07). (g) FA image of peripheral lesions (04:57). (h) Late-phase FA image of peripheral lesions (15:38). Scale $\mathrm{bar}=3\mathrm{deg}$.

One of the powerful features of the mLSO is the ability to merge the reflectance, FA, and ICGA images to obtain a more clear picture of ocular health. In a future design, this could be presented to a clinician in real-time with multiple sensors. Figure 12 shows pseudo-color maps for two subjects (without disease and with CSC) generated by merging the gray-scale images from FA and ICGA modes into the green (G) and blue (B) planes of a color image. The merged FA and ICGA image delineates the retinal and choroidal vasculature nicely and enhances disease visualization in the CSC subject. We also added the reflectance image in the red plane (not shown), which distinguished between the disc margin, the scleral crescent, and the lesions in the CSC eye, but did not contribute further information in the eye without disease.

Fig. 12

Pseudo-color images created by merging FA and ICGA images from mLSO for two subjects with: (a) No retinal disease and (b) CSC. Color planes: R: none, G: ICGA, B: FA.