1. Introduction

Optical coherence tomography (OCT) is an emerging technology for micrometer-scale, cross-sectional imaging of biological tissue and materials.[1] A major application of optical coherence tomography (OCT) is ophthalmic imaging of the human retina in vivo.[2,3] The Spectral-Domain OCT (SDOCT) improvement of the traditional time domain OCT (TDOCT) technique, known also as Fourier domain OCT (FDOCT), makes this technology suitable for real-time cross-sectional retinal imaging at video rate. [4–8]. At high speed, the need for vertical realignment of “A-scan” depth profiles is effectively eliminated across single B-scans, revealing a truer representation of retinal topography and the optic nerve head. Although B-scan image distortion by involuntary eye movement is reduced, transverse eye motion remains an issue for 3-D imaging and individual scan registration.[9,10] On the expense of higher complexity, stabilized 3D OCT imaging [11,12] can provide an en face fundus views for locating any given B-scan relative to retinal landmarks. Alternatively, simultaneous or interleaved live fundus imaging can also provide good retinal coordinates for a given B-scan, subject to the limitations of inter-frame eye motion. Work performed by Podoleanu’s group [13] shows that scanning transversal SLO/OCT imaging techniques can also accomplish en face and B-scan co-registration directly, although with more complex instrumentation.

The fusion of wide-field, line scanning laser ophthalmoscope (LSLO) retinal imaging with SDOCT imaging can enhance the clinician’s ability to quickly assess pathologies in linked, complementary views with a simple, compact instrument. As will be seen, this concept evolved naturally from the like requirements of the LSLO and SDOCT imaging instruments (e.g., both require a linear array detector, both use galvanometers for transverse scanning, both operate at tens of kHz line rates, with similar optical power and bandwidth requirements, etc.). In order to make the ocular interface of future SDOCT systems more efficient, cost-effective, compact, and eventually field portable, neither complex motion stabilization systems nor optomechanical integration of secondary fundus cameras is desirable. Yet without precise knowledge of the OCT scan coordinates within the live fundus image to guide scan acquisition and interpretation, the diagnostic utility of this powerful imaging modality is limited.

In this paper we present a novel integration method for real-time LSLO-SDOCT image co-registration using a relatively simple optical scheme that shares the same imaging optics and line scan camera for both imaging modalities. OCT and LSLO interleaved images are obtained at 15 frames/second. The short time interval between LSLO and SDOCT scans minimize registration errors.

2. Methods and materials

A simplified schematic of our proposed approach is shown in Figure 1. The same scanning lenses, imaging lenses and linear array detector are used for the acquisition of both LSLO and OCT images.

The LSLO portion of the system consists of a line-confocal reflectometer, which contains a cylindrical lens (CL) that fans out the imaging light beam (920 nm SLD) to a line at the fundus, and imaging lenses, including a scan element (O1), an ophthalmoscopic lens (OL), an imaging objective (O2), an aperture separating/beam-combining element (BC) that isolates the entrance and exit apertures, a galvanometer scanner (Gy), and a linear array detector (LA). The line scanner simultaneously scans a line across the ocular fundus and de-scans the backscattered return onto a linear array camera. The LSLO is fully confocal in only one dimension since it uses line illumination rather than a flying spot. However, our previous work demonstrates that high quality, high contrast near-infrared (NIR) fundus images can be obtained with this simple but efficient approach.[9]

The SDOCT portion of the system shares most of the LSLO parts: the SDOCT spectrometer shares the same imaging lens system and array detector with the LSLO system; and the sample arm of the SDOCT system shares the same scanning lenses. The separation of the LSLO and OCT beam input paths (via BC) is accomplished using two identical dichroic mirrors, one mounted on scanner (DGx) and one in a fixed position (D). The LSLO uses only one scanner (Gy) to generate a raster, while the SDOCT can use one or both scanners to generate arbitrarily oriented cross-sectional or raster images. During the LSLO portion of scan DGx is fixed. The SDOCT signal beam path de-scans to the SLD fiber and the LSLO signal beam de-scans to the linear array.

The instrument can run in three modes: LSLO mode only, OCT mode only, and frame-interleaved LSLO/SDOCT mode. No moving parts are required to change imaging modes: a simple software switch controls the hardware configuration for each imaging mode “on the fly”. When switched, the desired source is turned on (and the other off) and the camera gain is changed if necessary, as are the transverse scan parameters of the data acquisition card. Thus, the LSLO and SDOCT systems are integrated in a unique manner with a common detection path that conserves sub-system capabilities and minimizes size, cost, and complexity.

 

Fig. 1. Schematic of the hybrid LSLO/SD-OCT approach.

Download Full Size | PDF

A schematic of the command lines, imaging raster, and timing sequence are shown in Figure 2.

 

Fig. 2. Hybrid timing. (a) Instrumentation, (b) Example of raster scan, and (c) frame timing diagram.

Download Full Size | PDF

As shown in Fig. 2(a) a cameralink framegrabber (IMAQ) is used for data collection while a data acquisition (DAQ) board is used for instrument timing. The analog outputs are used to control both scanners, while the digital lines are used to control the camera gain through the RTSI line and also to turn “on” and “off” any of the light sources. The timing diagram (Fig. 2(c)) shows the combined LSLO/SDOCT mode when SDOCT data are acquired. The OCT light source is turned “on” while the LSLO source is turned “off”, the y galvanometer moves to the selected y coordinate, and the x galvanometer scans the OCT beam over a smaller distance.

Based on the above mentioned approach we have built a preliminary benchtop version of the LSLO/SDOCT system. A picture of this system is shown in Figure 3. A broadband super-luminescent diode (SLD-37MP, Superlum-Russia) with 830 nm central wavelength and approximately 50 nm bandwidth is used as OCT light source. A 920 nm SLD (QSDM-920-2, Q-Photonics) with about 35 nm FWHM and 2 mW output power is used for LSLO imaging. Custom designed objectives that include meniscus lenses to control field flatness and chromatic aberration are used for the scan lens system, imaging lens system, and OCT collimators. Short focal length scan and ophthalmoscopic lenses are used to reduce system dimensions and susceptibility to reflections.

 

Fig. 3. (a). Photograph of the LSLO/SDOCT system. (b). Photograph detail showing most of the major parts of the LSLO/SDOCT system.

Download Full Size | PDF

The core of the SDOCT system is the spectrometer. The optical spectrum is dispersed by a holographic diffraction grating (Wasatch Photonics, 1200 lines per mm) and imaged by a custom designed lens system onto a Silicon CCD array line scan camera (Atmel AVIIVA M2 CL 1014). The CCD has 1024 detector pixels with a 14 µm pitch and can operate at a maximum 60 MHz data rate. The output of the camera is connected to a camera link board (NI PCI-1429). The sampled data are transferred continuously to computer memory. The usual λ to ω (or k) interpolation is performed for our spectrometer. A discrete Fourier transform (DFT) is performed on each set of 1024 data points acquired by the CCD to produce an axial depth profile of the sample (A-line).

The graphical user interface for the system is shown in Fig. 4. The LSLO and SDOCT images are displayed separately or simultaneously (depending on the imaging mode), and the SDOCT scan can be positioned anywhere in the LSLO raster (see the yellow line). Other controls for SDOCT processing, display, and saving or streaming to disk are in a tab box at the bottom. The raw spectrum and processed profile are shown below the images and the integrated fixation target is also displayed.

3. Results and discussion

The current design of our instrument allows imaging with a speed of 30 frames/sec in LSLO mode, 15 frames/sec in OCT mode, and 15 frames/sec in the combined LSLO/SDOCT mode. Each LSLO frame has 1024×1024 pixel resolution, while the OCT frame has 1024×512 pixel resolution. However, even though OCT data are acquired at15 frames/sec, current processing speed of SDOCT lines prevents the displaying of OCT images at high rates. We have configured our system to acquire 15 frames/sec in the OCT mode and display every other fifth frame, which is equivalent to 3 frames/sec speed. In the LSLO/SDOCT mode the system is configured to display one OCT frame at every 14 LSLO frames. Another speed limitation is given by the transition from OCT mode to LSLO mode, which requires some additional time to change the gain of the camera. This gain is increased only during the LSLO scans to maintain a low illumination power. Since the gain is controlled via the serial lines on the cameralink interface, the synchronicity of gain switching is only software configurable and is thus prone to jitter. This limits the potential speed at which the frames can be interleaved given the present hardware limitations, which are not fundamental. However, a custom processing board is under development to provide more sophisticated timing and real-time processing of SDOCT data. With the real-time processing board, the system will eventually be able to aquire data at rates of 30 frames/sec or more, and interleave and display the LSLO and SDOCT frames consecutively at 15 frames/sec each.

 

Fig. 4. Graphical user interface (GUI) of the LSLO/SDOCT system

Download Full Size | PDF

In order to achieve the optimal resolution OCT images consistent with the light source bandwidth, it was necessary to carefully compensate the dispersion between the two arms of the interferometer. The standard technique balances the dispersion of the sample by adding a dispersive material in the reference arm. However, this correction technique might provide sub-optimal compensation of dispersion and requires adjustment from one sample to another. Our numeric dispersion algorithm enables automatic dispersion correction of depth reflectivity profiles at different positions with the eye. A detailed description of this algorithm is presented elsewhere. [12].

Preliminary cross-sectional retinal OCT images acquired from the hybrid system are presented in Fig. 5 The cross-sectional image was obtained over 5 mm transverse scan range and includes 1024 A-lines. Each processed A-line has 512 pixels. The dynamic range within the image was approximately 44 dB. Most of the anatomic layers can be recognized in these images. It is to be noted that the external limiting membrane (ELM) can be clearly seen over the whole width of the image. This layer is in general not visible if dispersion correction is not performed properly.

 

Fig. 5. Preliminary SDOCT retinal cross-sectional scans acquired from the system. (a) Single image, (b) Composite image averaged from 20 single frames. NFL: nerve fiber layers, GCL: ganglion cell layer, IPL: inner plexiform layer, INL: inner nuclear layer, OPL: outer plexiform layer, ONL: outer nuclear layer, ELM: external limiting membrane, CC: connecting cilia between photoreceptor inner and outer segments, RPE: retinal pigment epithelium, C: choriocapillaris and choroid.

Download Full Size | PDF

We have carefully designed our OCT system to provide good sensitivity, over 95dB, which has proven capable of letting us see the most important anatomical features within the retina. Inherent losses on all optical elements, especially on the dichroic mirrors (about 10%), have some impact on the system’s sensitivity. The LSLO mode sensitivity is affected as well by the wavelength dependent losses on the dichroic mirrors. Fortunately, LSLO imaging tolerates this loss since LSLO power is well bellow the ANSI limit and therefore a slight increase in the power can compensate these losses. However, we regard the simplicity of the optical layout and its compactness as net benefits which more than compensate other limitations. Future work will also focus on minimizing losses with new sources and improved custom optical coatings.

4. Conclusions

In conclusion, we have demonstrated that real-time interleaved SDOCT imaging and wide-field LSLO fundus imaging with a compact, single sensor design is possible. Both imaging modalities are necessary to provide to the clinician the optimal diagnostic information. The more conventional en face LSLO view of the entire retina provides the bigger picture on global ocular health and orientation, while the cross-sectional OCT view provides a high resolution detail of retinal layers and cellular and sub-cellular structures in a region of interest. By overlaying fiducial lines or boxes over the live LSLO image display, the position and orientation of the SDOCT scan(s) to be captured in the subsequent frame can be precisely scaled and oriented to show retinal features at the desired location and resolution.

In future work, improved spectrometer and high throughput optical design will be used to improve the imaging resolution as well as the dynamic range and the imaging depth. Higher frame-rate and more flexible image acquisition and processing systems are currently in development. More intuitive LSLO registration and novel SDOCT imaging modalities will be integrated simply by creating new scan configurations with corresponding overlays. For example, local 3D image capture and display of very small features or lesions is possible. Small square “micro-rasters” can be captured in real time with many small amplitude scans instead of single long linear scan, and displayed together as one composite B-scan for high-resolution SDOCT representation of compact volumes. Such an approach would be ideal for elucidating columnar lesions due to laser damage or other localized pathology.