Optical coherence tomography (OCT) has emerged as a powerful imaging instrument and technology in biomedicine. OCT imaging is predominantly performed using wavelengths in the near infrared; however, visible light (VIS) has been recently employed in OCT systems with encouraging results for high-resolution retinal imaging. Using a broadband supercontinuum VIS source, we present a sensorless adaptive optics (SAO) multimodal imaging system driven by VIS-OCT for volumetric retinal structural imaging, followed by the acquisition of fluorescence emission. The coherence-gated, depth-resolved VIS-OCT images used for image-guided SAO aberration correction enable high-resolution structural and fluorescence imaging.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Advancements in optical imaging technology have revolutionized vision science and clinical ophthalmology research. Among various optical imaging instruments, optical coherence tomography (OCT) has become one of the most successful technologies implemented in medicine and clinical research, mostly due to the possibility of non-invasive and non-contact three-dimensional volumetric imaging by detecting back-scattered light. Although most OCT imaging has been performed using near-infrared (NIR) light sources, recent studies have shown the capability of visible light OCT (VIS-OCT)  to perform structural and functional retinal imaging in both humans [2–4] and small animals [5–8]. Aside from strong absorption in the blood vessels, the recent results are extremely encouraging for high-quality retinal imaging .
Although VIS-OCT enables high axial resolution retinal imaging, the transverse resolution of retinal imaging is still limited by the finite size of the eye pupil and imperfections of the optical properties of the eye. Aberrations introduced by the tear film, cornea, and intraocular lens reduce the highest transverse resolution at a given numerical aperture (NA). To approach diffraction-limited in vivo retinal imaging, adaptive optics (AO), a technique that was originally introduced for astronomy to sharpen the images acquired of stars, has been applied to correct the aberrations .
While high-resolution two-dimensional (2-D) tomographic images of the retina allow for detailed visualization of the retinal structure and analysis of pathology, fluorescence imaging has the unique ability to visualize the biological function of the retina through labeled reporter cells [10,11]. The fluorescence images acquired with conventional scanning laser ophthalmoscopy (SLO) are 2-D en face images, and require complementary information to determine the retinal layer in which the fluorescent molecules are located . With the help of AO, the AO-assisted fluorescence SLO (AO-fSLO) can perform optical depth sectioning, as well as diffraction-limited imaging, but still does not provide direct information as to where in the retina the focus is placed. A multimodal AO instrument combining fSLO and NIR-OCT, capable of providing 3-D location of features that are visible in both fluorescence and OCT channels, has been demonstrated [13,14]. However, due to separate light sources used for each imaging modality, the 3-D localization of the fluorophores was not possible for features that did not have an OCT signature.
In this Letter, for the first time, to the best of our knowledge, we present a depth-resolved multimodal visible light sensorless adaptive optics (VIS-SAO) retinal imaging system. By using a supercontinuum light source, VIS-OCT can be combined with fSLO to provide dual-mode acquisition of structural and fluorescence images that are naturally co-aligned without any post-processing registration. Additionally, by leveraging the high depth resolution attainable with VIS-OCT, the specific retinal layer was precisely selected with the assistance of real-time retinal layer tracking software , which enabled coherence-gated depth-resolved SAO optimization. The SAO algorithm used in this Letter is similar to our previous studies [16–19]. The image quality metric was calculated based on the en face VIS-OCT image. With the merit function defined as the image sharpness, we optimized the wavefront of the probe beam in a hill-climbing fashion based on Zernike mode control using deformable optical elements.
Figure 1 shows the schematic of the system used for multimodal mouse retinal imaging. A supercontinuum laser (Fianium WhiteLase Micro and VARIA tunable filter, NKT Photonics Inc.) that provides tunable broadband illumination was used as the light source. The excitation beam was divided into sample and reference arms by a 50/50 fiber coupler. The sample arm consisted of a segmented deformable mirror (PTT-111, IrisAO Inc.) for aberration correction and a variable focus lens (Arctic 39N0, Varioptics) to control the focal plane in the sample. A telescope was used to relay the imaging beam to the galvanometer-scanning mirrors (6210H, Cambridge Technology Inc.) for 2-D transverse scanning of the sample. In the reference arm, a dispersion compensation block was placed to match the dispersion between the two arms of the interferometer. On the detector side, we used a custom-built spectrometer with a 50 mm focal length achromatic collimating lens at the input. In the spectrometer, the interference signal generated by light returning from the sample and reference paths was spectrally dispersed with a VIS transmission volume phase holographic grating (1800 lines/mm, Wasatch Photonics Inc.) before being imaged onto a CMOS line scan camera (BASLER Sprint spl4096-140 km) using another achromatic doublet pair with 100 mm focal length. The A-scan (depth-scan) rate was configured at 40 kHz, resulting in an acquisition rate of 1 volume per second with acquisition parameters of sampling points. Real-time image acquisition was performed with our custom-built GPU-accelerated program .
The SAO aberration correction algorithm used in this Letter is similar to our previous reports on SAO for human [16,17,21] and mouse [18,19,22] retinal imaging. The en face image was generated from a manually selected depth region in the OCT volume by maximum intensity projection (MIP). The sum of squared intensity (sharpness) of the VIS-OCT en face image was used as the merit function for the SAO aberration correction. The optimization of the Zernike modes was performed in a hill-climbing fashion, where the coefficient producing the sharpest image was selected as the optimized value. The volume size was decreased to during the optimization, and 11 different coefficients were applied to the deformable mirror for each Zernike mode. Searching up to the 5th radial order (omitting the first three Zernike modes: piston, tip, and tilt; standards for reporting the optical aberrations of eyes ), the resulting optimization time was . For fluorescence imaging, the fluorescence emission from the sample was transmitted through a multi-edge filter beam splitter (89402bs, Chroma Technology) to the photo-multiplier tube (PMT) detector (H10723, Hamamatsu Photonics Inc.). A multi-edge filter (89402m, Chroma Technology Inc.) was used as a clean-up filter to reject any residual excitation light, and a lens and pinhole were used to reject out-of-focus light with a confocal aperture times that of the Airy disk. The digitization of the PMT signal was synchronized to the acquisition of the VIS-OCT A-scan to ensure that both the OCT and fluorescence images were perfectly registered. OCT-guided SAO optimization was first performed using the en face OCT image, followed by switching the imaging software to a fluorescence mode. Fluorescence images with the same wavefront correction were acquired with sampling points, resulting in a frame rate of 10 frames per second.
Initial imaging experiments on a tissue phantom were performed using a center wavelength of 470 nm and full width at half maximum (FWHM) of 30 nm for simultaneous VIS-OCT and fluorescence imaging. This wavelength range is compatible with excitation of fluorescein and enhanced green fluorescent protein (EGFP), which is readily available in many different transgenic mouse models and has bright fluorescent emission. Using the blue light, phantom imaging was performed with a model eye constructed from a 2.97 mm focal length aspheric lens (C660TME-A, Thorlabs Inc.) and lens tissue fibers labeled with fluorescein. The lens configuration of the final optical relay was and . Aberrations were created by placing a gel between two non-uniform plastic surfaces, generating astigmatism and coma. Figures 2(a) and 2(b) show the en face OCT and fluorescence images, respectively, in the presence of the aberrations. Next, the SAO optimization process was performed on the OCT en face image to correct the induced aberrations. Figures 2(c) and 2(d) are the images after the optimization, and the intensity plots across the fluorescence image marked by a dashed red line in Fig. 2(b) and a solid blue line in Fig. 2(d) are shown in Fig. 2(e). In both OCT and the fluorescence images, the features in the lens paper become brighter and better resolved after the SAO optimization; improvements of and were measured in the en face OCT and fluorescence images, respectively, with the same optimization parameters. This result demonstrates that the SAO improved the en face OCT image and, simultaneously, improved the fluorescence image.
In vivo retinal multimodal VIS-SAO imaging experiments were performed using light with a center wavelength () of 560 nm, a FWHM of 30 nm, and a probe power of at the mouse cornea. Initial in vivo mouse retinal imaging was performed with a beam diameter of 0.7 mm at the mouse eye, corresponding to an NA of 0.18 ( & ). Figures 3(a) and 3(b) show single VIS-OCT B-scan images with this NA. These VIS-OCT results were obtained with a 470 nm wideband fiber coupler (TW470R5A2, Thorlabs Inc.). The low SNR was further exacerbated in the field-of-view (FOV) image [Fig. 3(b)] compared to the FOV image [Fig. 3(a)], which was acquired with the same number of samples. Although the mouse retinal structure was more clearly visualized after averaging 100 and 200 B-scan images for small [Fig. 3(d)] and large [Fig. 3(e)] FOVs, respectively, the low SNR of the VIS-OCT still caused several limitations for the SAO optimization process, affecting the retinal tracking and layer selection. In order to improve the SNR of VIS-OCT, the 470 nm wideband fiber coupler was replaced with a 560 wideband 50/50 coupler (TW560R5A2, Thorlabs Inc.) for enhancing the back-coupling efficiency at the center wavelength of 560 nm, while still preserving single-mode delivery of the excitation light at . With the improvement in OCT image quality, subsequent mouse retinal imaging was performed with the probe beam increased to 0.8 mm at the mouse eye, corresponding an NA of 0.21 ( & ). With the high NA setting, a single VIS-OCT volume was acquired with the focus at the retinal nerve fiber layer (RNFL). A representative VIS-OCT B-scan image is shown in Fig. 3(c), and the en face MIP image extracted from the region within the dashed red rectangular window in Fig. 3(c) is shown in Fig. 3(f). The experimental results shown in Figs. 3(c) and 3(f) demonstrate successful VIS-OCT retinal imaging at a center wavelength of 560 nm that is suitable for VIS-OCT-guided SAO aberration correction. The limitation with using this wavelength, however, is that there are not many mouse models expressing fluorophores (such as red fluorescent protein) within the retina. The solution to this limitation was to separate imaging modes by controlling the tunable filter, permitting VIS-OCT optimization at 560 nm and fluorescence excitation at 470 nm. As a result, following the VIS-OCT imaging and SAO optimization at the 560 nm wavelength range, the output from the supercontinuum was tuned to a center wavelength of 470 nm for fluorescence excitation.
Figure 4 shows representative fluorescein retinal angiography acquired from a wild-type mouse. The SAO aberration correction was performed on the RNFL in VIS-OCT; then the focus was shifted to the outer plexiform layer (OPL) to image the capillaries with the fluorescence channel. In this case, the RNFL was used for SAO aberration correction, because it provided high-quality en face VIS-OCT images to guide the image-based optimization. After the aberration correction, an increase in the image brightness () was observed, as well as more clearly resolved features within the inner retinal layers of the VIS-OCT B-scan, as shown in Figs. 4(a) and 4(c), which corresponded to an improvement in the lateral resolution of the fluorescein angiography of the capillaries, as shown in Figs. 4(b) and 4(d).
The imaging system performance was also demonstrated on a mouse with ganglion cells labeled with EGFP. As in the previous example, the optimization was performed on the RNFL layer using the VIS-OCT image data on a small FOV of . Comparing the B-scan images in Figs. 5(a) and 5(d), we see that there is a contrast improvement () of the VIS-OCT at the RNFL layer after the SAO aberration correction; the Zernike coefficients for the aberration correction are presented in Fig. 5(h). Following the SAO optimization, the FOV was increased, and the wavelength was tuned to 470 nm to excite the EGFP labeled cells. The improvement in lateral resolution was demonstrated by the fluorescence images in Figs. 5(b), 5(c), 5(e), and 5(f), and the line spread function of the system in Fig. 5(g).
Compared to the conventional OCT with NIR light, VIS-OCT inherently has a higher lateral resolution at a given NA, and can achieve a high axial resolution with a narrower spectral bandwidth. Besides the advantages of VIS-OCT, there are also some limitations. For example, since melanin in the RPE is highly absorbing in the VIS range , the penetration depth into the choroid is limited, and visible excitation light has potential to cause photobleaching and photodamage throughout the retina. The supercontinuum source used for the VIS-OCT system also has stronger inherent noise than the superluminescent diodes commonly used for NIR-OCT systems. As a result, the combination of a lower illumination power and higher light source noise in VIS-OCT lead to relatively lower SNR than in NIR-OCT . In this Letter, by employing the SAO technique to correct aberrations and optimize the incident wavefront, we demonstrated an improvement in the SNR, as well as the transverse resolution. In the visible spectrum, wavelength-dependent variations in the refractive index are larger compared to the NIR range, leading to increased chromatic aberration. Longitudinal chromatic aberration has potentially the most deleterious impact on image quality ; hence, we used a relatively narrow bandwidth to achieve adequate depth sectioning while minimizing the effects of chromatic aberrations.
VIS-OCT with blue light has been demonstrated by other research groups, for example Refs. [5,26]. Our experience was that switching to 560 nm for VIS-OCT provided mouse retinal images with higher SNRs compared to 470 nm. However, there are not many mouse models that express red fluorescent protein in the retina. In this Letter, the SAO aberration correction was performed at a longer wavelength than the fluorescence emission wavelength, instead of optimizing the aberration correction at the excitation wavelength range. The cross-sectional imaging ability of OCT () was utilized to correct aberrations with direct access to a specific retinal layer. After acquiring high-resolution depth-resolved OCT volume data, the center wavelength was tuned to 470 nm for fluorescence imaging. A significant benefit of this approach was that the fluorophore did not photobleach, while performing the SAO optimization using VIS-OCT at the wavelength longer than the fluorescence excitation wavelength, and with a low incident power on the mouse eye. Since the SAO optimization algorithm is purely based on an image quality metric and not on direct wavefront measurement with a wavefront sensor device, there is flexibility in terms of wavelength tuning for imaging and for aberration correction. As shown the results, multiple wavelengths can be used for VIS-OCT and fluorescence imaging with different fluorophores. As an example, transgenic mice expressing different fluorophores in the retina can be imaged with different spectral bands to visualize different cells and features within the same FOV.
In conclusion, we have, for the first time, to the best of our knowledge, demonstrated a dual-mode VIS sensorless AO retinal imaging system with a single supercontinuum light source, which can provide both high-resolution structural and fluorescent imaging of mouse retina. The proof-of-concept results encourage further investigation of this approach as a valuable tool for vision research applications.
Michael Smith Foundation for Health Research (MSFHR); Alzheimer Society Research Program (ASRP); National Research Foundation of Korea (NRF) (NRF-2018K1A4A3A02060572).
1. X. Shu, L. Beckmann, and H. F. Zhang, J. Biomed. Opt. 22, 121707 (2017). [CrossRef]
2. S. Chen, X. Shu, P. L. Nesper, W. Liu, A. A. Fawzi, and H. F. Zhang, Biomed. Opt. Express 8, 1415 (2017). [CrossRef]
3. S. P. Chong, M. Bernucci, H. Radhakrishnan, and V. J. Srinivasan, Biomed. Opt. Express 8, 323 (2017). [CrossRef]
4. P. L. Nesper, B. T. Soetikno, H. F. Zhang, and A. A. Fawzi, Vision Res. 139, 191 (2017). [CrossRef]
5. Z. Nafar, M. Jiang, R. Wen, and S. Jiao, Biomed. Opt. Express 7, 3220 (2016). [CrossRef]
6. S. Chen, J. Yi, and H. F. Zhang, Biomed. Opt. Express 6, 2840 (2015). [CrossRef]
7. R. P. Mcnabb, T. Blanco, D. R. Saban, J. A. Izatt, and A. N. Kuo, Invest. Ophthalmol. Vis. Sci. 55, 1779 (2014).
8. S. P. Chong, C. W. Merkle, C. Leahy, H. Radhakrishnan, and V. J. Srinivasan, Biomed. Opt. Express 6, 1422 (2015). [CrossRef]
9. S. P. Chong, T. Zhang, A. Kho, M. T. Bernucci, A. Dubra, and V. J. Srinivasan, Biomed. Opt. Express 9, 1477 (2018). [CrossRef]
10. D. R. Williams, Vision Res. 51, 1379 (2011). [CrossRef]
11. P. F. Sharp, A. Manivannan, H. Xu, and J. V. Forrester, Phys. Med. Biol. 49, 1085 (2004). [CrossRef]
12. L. Zhang, W. Song, D. Shao, S. Zhang, M. Desai, S. Ness, S. Roy, and J. Yi, Biomed. Opt. Express 9, 25 (2018). [CrossRef]
13. D. X. Hammer, F. R. Daniel, M. Mujat, A. Patel, E. Plumb, N. Iftimia, T. Y. P. Chui, J. D. Akula, and A. B. Fulton, J. Opt. Soc. Am. A 29, 2598 (2012). [CrossRef]
14. P. Zhang, A. Zam, Y. Jian, X. Wang, Y. Li, K. S. Lam, M. E. Burns, M. V. Sarunic, E. N. Pugh, and R. J. Zawadzki, J. Biomed. Opt. 20, 126005 (2015). [CrossRef]
15. M. Cua, S. Lee, D. Miao, M. J. Ju, P. J. Mackenzie, Y. Jian, and M. V. Sarunic, J. Biomed. Opt. 21, 026007 (2016). [CrossRef]
16. K. S. K. Wong, Y. Jian, M. Cua, S. Bonora, R. J. Zawadzki, and M. V. Sarunic, Biomed. Opt. Express 6, 580 (2015). [CrossRef]
17. Y. Jian, S. Lee, M. J. Ju, M. Heisler, W. Ding, R. J. Zawadzki, S. Bonora, and M. V. Sarunic, Sci. Rep. 6, 27620 (2016). [CrossRef]
18. M. Cua, D. J. Wahl, Y. Zhao, S. Lee, S. Bonora, R. J. Zawadzki, Y. Jian, and M. V. Sarunic, Sci. Rep. 6, 32223 (2016). [CrossRef]
19. D. J. Wahl, Y. Jian, S. Bonora, R. J. Zawadzki, and M. V. Sarunic, Biomed. Opt. Express 7, 1 (2016). [CrossRef]
20. Y. Jian, K. Wong, and M. V. Sarunic, J. Biomed. Opt. 18, 26002 (2013). [CrossRef]
21. M. J. Ju, M. Heisler, D. Wahl, Y. Jian, and M. V. Sarunic, J. Biomed. Opt. 22, 121703 (2017). [CrossRef]
22. Y. Jian, J. Xu, M. A. Gradowski, S. Bonora, R. J. Zawadzki, and M. V. Sarunic, Biomed. Opt. Express 5, 547 (2014). [CrossRef]
23. L. N. Thibos, R. A. Applegate, J. T. Schwiegerling, R. Webb, V. Standards, and T. Members, J. Refract. Surg. 18, 652 (2002).
24. J. Yi, S. Chen, X. Shu, A. A. Fawzi, and H. F. Zhang, Ophthalmology 121, 1435 (2014). [CrossRef]
25. W. J. Brown, S. Kim, and A. Wax, J. Opt. Soc. Am. A 31, 2703 (2014). [CrossRef]
26. C. Dai, X. Liu, and S. Jiao, J. Biomed. Opt. 17, 080502 (2012). [CrossRef]