We propose a spectral domain optical coherence tomography (SD-OCT) system that uses a single line-scan detection scheme for balanced detection. Two phase-opposed spectra, generated by two optical fiber couplers, were detected by using a spectrometer with fast optical switching. A 2.69 km optical fiber was introduced to provide a proper time delay to prevent phase errors caused by the difference in measurement time between the two opposing spectra and unstable output voltages for controlling the galvano-scanner. Hence, a phase difference of π was obtained between the spectra over the sample depth without a phase error, which improved sensitivity by approximately 6 dB compared to that of conventional SD-OCT. We directly showed and compared the OCT images before and after applying the proposed balanced detection method in a phantom and in vivo sample.
© 2015 Optical Society of America
Optical coherence tomography (OCT) is a noninvasive and depth-resolved in vivo imaging technique that can image tissue structure with a high axial resolution [1,2]. Unlike conventional time domain OCT (TD-OCT), Fourier domain OCT (FD-OCT) obtains depth-resolved information of the tissue through Fourier transform of the detected interference signal. FD-OCT can be generally classified as spectral domain OCT (SD-OCT) and swept source OCT (SS-OCT). FD-OCT images intrinsically contain zero-delay lines and autocorrelation (AC) artifacts. This noise causes image distortion. Balanced detection (BD) has improved the signal-to-noise ratio (SNR) and efficiently removed DC noise and AC artifacts in the OCT configuration using a dual balanced detector such as in the cases of TD-OCT and SS-OCT [3–5]. On the other hand, most existing SD-OCT techniques use the unbalanced detection (UD) method to configure a single spectrometer in the detection arm. The background subtraction is an essential signal procedure to minimize DC noise; however, lower sensitivity and AC noise remain its drawbacks.
Recently, several balanced detection SD-OCT (BD-SD-OCT) designs have been proposed. One such proposed SD-OCT system used two different spectrometers. However, this system suffered from alignment issues while acquiring two opposing fringe signals [6,7]. Although the two spectrometers were tightly aligned, the output fringes often contained minor phase errors, which decreased the SNR significantly, especially with increasing optical path difference (OPD) between the sample and reference arms. In addition, for optimal use of the two-spectrometer system, two cameras should have the same quantum efficiency and uncorrected noise during signal acquisition. Furthermore, using two spectrometers makes the system expensive and bulky. Another BD-SD-OCT system used a single-area CCD camera . In the previous study, the spectrometer comprised two input fibers and an area CCD camera to simultaneously measure the two opposing phased fringe spectra. However, this BD method utilized the whole detection area of the CCD camera from which only two spectral lines were selected to balance the interference signals. The unused data-flow made the data acquisition slower and limited the overall A-scan rate to lower than a few frames per second.
Optical switches have served various purposes in OCT, such as AC removal, single spectrometer polarization-sensitive OCT (PS-OCT), and simultaneous imaging of an anterior segment and retina [9–12]. Also, buffering methods using a delay line (DL) have been widely applied in Fourier domain mode locked laser (FDML) sources, system speed improvement, imaging depth enhancements, etc [13–17].
In this work, we propose a novel BD method for SD-OCT with a single line-scan camera. The proposed method applies a 1 × 2 optical switch and a time delay to achieve balanced interference signal acquisition. Using our method, we successfully removed the phase mismatch between two anti-phase interference signals and enhanced the dynamic range of the OCT system.
2.1 Experimental setup
Figure 1 shows the schematic of the proposed system based on a Mach-Zehnder interferometer. A superluminescent diode (SLD, DenseLight Semiconductors Incorporated, Singapore) provides a broadband optical source with a center wavelength of 1310 nm and FWHM spectral width of 150 nm that is split into a reference arm and a sample arm by a 1 × 2 90/10 fiber coupler (FC1). Reflected light from each arm recombines in a 2 × 2 50/50 fiber coupler (FC2) to form interference signals. After the signals pass through a second coupler (FC2), the phase difference between the divided signals is naturally π. One of the FC2 output is directly guided to a fast 1 × 2 optical switch (maximum switching speed of 1 MHz, Boston Applied Technologies, USA) and the other travels along an optical delay line. The optical switch delivers two anti-phase fringes from the 2 × 2 coupler to the single spectrometer in sequence. BD utilizes two consecutive A-scan signals from adjacent, but different, sample positions, which inevitably creates errors without the DL. The DL prevents this problem. The SD-OCT system consists of a linear-in-wavenumber spectrometer (Bayspec, USA) to improve the sensitivity roll-off. An InGaAs line-scan camera (Goodrich Corporation, USA) in the spectrometer has 2048 pixels, 12-bit resolution, and a maximum line rate of 76 kHz. Axial resolution and transverse resolution were measured to be 7.9 μm and 21 μm in air, respectively. A CUDA (computer unified device architecture) and multithread programming allowed for real-time display of OCT images.
2.2 Spectral fringe signal flow with optical switching and time delay
Figure 2(a) represents the synchronized signals timing diagram between the camera, the optical switch, and the 2-D galvano-scanner controlled by the DAQ board. BD OCT requires two anti-phase fringe signals from the same sample spot. However, the DAQ analog output signal instability easily affects the interference signals. Even though the same voltage signal drives the galvano-scanner during the acquisition of each of the two anti-phase signals, a small, unexpected OPD change can be introduced, as shown in Fig. 2(b). The OPD change causes a random phase shift between the two interference signals generated at time t1 and t2, losing the π phase difference. Balancing with the two mismatched interference signals deteriorates the sensitivity and the visibility of the OCT images.
Figure 3 shows signal flow diagrams following the optical switch redrawn from the red box of Fig. 1 with and without corrective time delay. S and S* refer to mutually opposing spectra collected simultaneously. Figure 3(a) shows the signal flow chart without DL. When channel 1 of the optical switch is on at time t1, S1, one of the two spectra formed at time t1, is transmitted to the spectrometer, but S1* is blocked in channel 2, which is switched off. When channel 2 of the optical switch is on at time t2, S2*, which is one of the two spectra collected at time t2, is transmitted to the spectrometer; however, S2 is blocked in channel 1, which is switched off. Figure 3(b) shows the signal flow-chart when the DL is introduced. The spectrometer measures the first fringe signal, S1, at time t1. The signal S1* generated from the same spot as S1, but with an anti-phase of π due to an intrinsic property of the fiber coupler, travels through the DL. By the time the signal S1* reaches the end of the DL, the optical switch has switched channel 2 on. Then, the signal S1* is detected. This sequence is repeated for every A-scan. To summarize, the camera sequentially detects different interference signals S1 and S2*, each generated at different times, without the DL. In contrast, the fringe pair, S1 and S1*, is generated at the same scanning position and time, with the DL. The two measured fringe signals have an exact phase difference of π. Thus, the BD calculations introduced in  can be easily performed without the expected potential errors shown in Fig. 3(a).
In our design, an extra DL that consists of a single mode fiber (SMF) several kilometers in length is incorporated prior to the optical switch to match the anti-phase signals. The modulation frequency of the optical switch is half of the camera line rate and the SMF length is given by:13].
Throughout this study, OCT images were obtained at a camera line rate of 76 kHz, optical switching rate of 38 kHz, and camera exposure time of 9.21 μs. The optical fiber delay line was 2.69 km length and the effective refractive group index was 1.43 at 1.3 μm wavelength.
3.1 Roll-off with depth for three different detection methods
To evaluate our design’s performance, we compared each acquired image with an image acquired by the conventional SD-OCT with UD scheme. All images obtained by UD was acquired after background subtraction to eliminate the reference (DC) power.
Spectrally BD optimally compensated for the difference in overall signal levels caused by a non-uniform coupling ratio of the 3 dB fiber coupler and insertion loss of the DL . It is important to match the spectral envelope of the interference signals to reduce DC term. Spectrally BD also eliminates undesirable fixed pattern noise due to light source fluctuations over time.
Figure 4 shows the interference fringes obtained by the BD-SD-OCT with and without the DL. A slightly tilted mirror was placed at the sample arm and laterally scanned. The phase shift without DL is not clear, as shown in Fig. 4(a), whereas the two fringe spectra with DL exhibit almost an exact π phase shift in Fig. 4(b). Figure 4(c) shows the balanced signals from Figs. 4(a) and 4(b) before and after correcting phase error using DL. The resulting difference fringe amplitudes are displayed in red and black in Fig. 4(c), respectively. As expected, the fringe amplitude is larger when DL is used. The sensitivity of the BD-SD-OCT system can be quantified after inverse Fast Fourier Transform (FFT) of the balanced signal, presented in Fig. 4(d). To measure the sensitivity of the system, 11.2 dB and 5.3 dB neutral density filters were placed in the sample and the reference arm (round trip 33 dB attenuation), respectively, and a 5 dB coupling loss occurred by tilting the sample mirror. The measured sensitivities of the BD without and with the DL were 56.85 dB and 61.64 dB at 200 μm, respectively. Therefore, the sensitivity improved by approximately 4.79 dB with the DL. The difference depends on the tilt angle of the mirror at the sample arm and the A-scan interval. When compensating for systematic attenuation, the peak sensitivity reaches 99.15 dB with DL.
Figure 5 shows the signal sensitivity fall-off as a function of depth. We measured the sensitivity of our SD-OCT system for three different cases: UD, BD without DL, and BD with DL. The effective imaging range for a 10 dB sensitivity drop was 3 mm for all cases. Throughout the depth range, the sensitivity difference between the BD-SD-OCT with DL and the UD OCT is 6 dB. The 6 dB sensitivity enhancement over the whole imaging range agrees with theoretical prediction. This implies that the π phase difference between the two spectra is maintained accurately, regardless of the OCT imaging depth.
3.2 A multilayer tape imaging
2D OCT images of a rolled tape are shown in Fig. 6. As before, an image for UD, BD without DL, and BD with DL are presented in Figs. 6(a)-6(c). The scan range of these 2D OCT images was set to 10 mm. As described earlier, the DC noise, shown in Fig. 6(a), was reduced by background subtraction. In addition, AC noise was minimized by increasing the reference power. BD naturally reduces DC noise. A noticeable change after applying BD method is the enhanced visible depth and overall sensitivity. As shown in Fig. 6(c), when BD was employed, the deeper region of the tape was more clearly visualized. It is obvious that imaging depth and brightness are improved in the BD. Comparing Fig. 6(b) and Fig. 6(c), we can see that the signal level of the tape image obtained by BD with the DL is higher than that of an image obtained by BD without the DL. To represent this in more detail, we extracted three A-scans for each of the red dotted lines in Figs. 6(b) and 6(c), and averaged them. The depth-resolved signals through FFT of the averaged A-line data are shown in Fig. 6(d). Figure 6(e) is the depth-resolved signal with a low pass filter. The boundary between air and the sample surface was located 0.8 mm from DC line. We found the mean intensity values of air and the sample regions before and after applying the DL. The mean value at the air region almost coincides with the noise level of the system. The mean noise levels in BD with and without the DL are −20.1 dB and −17.05 dB, respectively, marked by dotted lines in Fig. 6(e). It shows that noise in BD with the DL is lower than that in BD without the DL. On the other hand, the mean values of the sample signals detected by the BD method with and without the DL are −7.7 dB and −10.2 dB, respectively. The sample signal detected by BD with the DL is approximately 2.75 dB higher than that detected by BD without the DL.
3.3 In vivo imaging of rat-tail and human eye
Figure 7 shows in vivo rat-tail OCT images obtained by the three detection methods as in the previous experiment. Each image consists of 350 (axial) × 500 (lateral) pixels and has a scan range of 10 mm. Comparing the three images, we see that the results are quite similar to those obtained in the previous multilayer tape experiment. We can easily observe that the image contrast of Fig. 7(c) is higher than that of other images, Figs. 7(a) and 7(b). Besides, the visible depth range is also extended owing to an increase in sensitivity.
Intensity profiles of the three red dotted line marked on Figs. 7(a)-7(c) are shown in Figs. 8(a)-8(c). Each A-line was extracted for the three detection cases and was constructed by taking average of 7 adjacent A-scans. The A-line was normalized after low pass filtering. Observing the change in intensity obtained by BD without the DL, we see that the intensity profile of BD with the DL exhibits an intensity profile similar to the UD case in Fig. 8(a). Figure 8(b) shows that the intensity of BD without the DL lies somewhere in the middle of UD and BD with the DL in depth of the sample. In contrast to Fig. 8(a), the intensity of BD without the DL is analogous to the intensity of BD with the DL. We calculated the normalized intensity mean values in the range between 0.3 mm and 1.4 mm (inside region of the tissue) for the three detection methods as shown in Fig. 8(d). Normalized intensity mean values of BD with the DL are constantly about 1.6 times higher than that of UD for the three A-scan locations shown in Fig. 7. On the other hand, the intensity mean values obtained by BD without the DL are changed randomly, depending on the later position of the tissue. It should be noted that if the phase difference between adjacent A-lines is ambiguous due to the high-scattering tissue, then the proposed system does not detect phase difference accurately, without the DL. If the proposed BD-SD-OCT system is implemented without the DL, the sensitivity will vary according to the A-scan position.
We performed an in vivo experiment on a human eye using our BD-SD-OCT system combined with the off-pivot scanning method for a full range complex imaging [19,20]. The resolution of the spectrometer was 0.073 nm and a maximum ranging depth of the UD OCT was limited by 5.9 mm. The depth range for full range imaging was extended to 11.8 mm. We imaged an anterior segment of a human volunteer’s eye, as shown in Fig. 9. The incident light power on the eye was measured to be 10 mW, which satisfies ANSI safety standards . The off-pivot scanning method requires a large number of A-lines, at least more than 1000, for efficient removal of mirror artifacts [22,23]. The 2D eye OCT images consisted of 2048 (axial) × 2500 (lateral) pixels. Images taken with UD, BD without DL, and BD with DL SD-OCT are presented in Figs. 9(a)-9(c), respectively. Three normalized intensity profiles of each yellow dotted line in Figs. 9(a)-9(c) (10 adjacent A-lines were averaged) was shown in Figs. 9(d)-(f). Red dotted lines in Figs. 9(d)-9(f) were obtained after low pass filtering. As shown in A-line profile of UD in Fig. 9(d), the signal level of the cornea was very close to the noise level, so it is hard to differentiate between sample signals and noise. However, Figs. 9(e) and 9(f) showed that the cornea was easily distinguished through BD. In Figs. 9(d)-9(f), the signal levels of red lines at the lens surface were 0.4 (UD), 0.58 (BD without DL), and 0.7 (BD with DL), respectively. Signals from the human eye was the highest in BD with DL. The difference of signal level between BD with and without DL could have been more pronounced if a smaller interval had been used between two adjacent A-scans. If the interval is smaller, the two consecutive A-line data used for BD will be almost identical, thus, the phase error becomes minimal. In this experiment, the signal level at the lens surface of BD with DL was only 1.2 times than that of BD without DL due to the use of 2500 A-scans. DL greatly reduced speckle noise, as seen in comparisons of Figs. 9(b) and 9(c). In Figs. 9(d)-9(f), Mean noise levels from 512 axial pixels to 1010 axial pixels in the anterior chamber region were 0.1 (UD), 0.18 (BD without DL), and 0.09 (BD with DL), respectively. Noise level of BD without DL was about twofold than that of BD with DL and it was also higher than that of UD after background subtraction.
4. Discussions and conclusion
BD in OCT requires two signals with exactly opposite phases from a single A-scan location to effectively remove undesirable noise, improving SNR. Since SD-OCT, unlike SS-OCT, uses a spectrometer as a detector, BD cannot be easily implemented. As was demonstrated, SD-OCT with BD can be achieved using a single line-scan camera and an optical switch, but such quick consecutive scanning creates slight variations in the A-scan positions of the two fringe signals with opposite phases. To solve this problem, a simple fiber optic DL is used.
An advantage of using the DL becomes pronounced when dealing with imaging samples with rough surfaces (see Figs. 7 and 9). If the DL is omitted, the two consecutive A-scan signals, which may not originate from the exact same spot, are combined to generate a balanced signal. Sharper surface morphology changes create larger phase ambiguities in the balanced signal. This effect becomes more significant when a smaller number of A-lines are used to generate the same sized 2D image. Increasing the number of A-lines from 500 to 2000 in BD-SD-OCT negates the effect of the DL (data not shown). This suggests that, especially with a smaller number of A-lines, the image quality could be much improved using our system. Comparing the proposed BD method with UD applied off-pivot full range technique, imaging depth of UD with off-pivot technique is doubled, but the sensitivity of the system remains unchanged. In addition, as a large number of A-scans will be required to eliminate mirror artifacts effectively, the speed of the system decreases accordingly. In contrast, the imaging depth of BD with DL is enlarged as the sensitivity of the system is improved, without increasing the number of A-scans. One shortcoming of our design is the image acquisition speed, which was reduced by half, since we measured two times at the same sample position using an optical switch for BD. Careful observation of Fig. 9 reveals that a simple background subtraction produces lesser speckle noise than the BD-SD-OCT without the DL. However, considering the fact that background noise can vary with time, the BD-SD-OCT still produces more stable images. It is clear that the BD-SD-OCT with the DL produces minimal speckle noise.
In this study, a very long SMF bundle acted as the DL. The wavelength range of the SD-OCT system spanned from 1235 nm to 1385 nm. The theoretical dispersion coefficient (D(λ)) in the fiber is approximately −7.3891 ps/(km·nm) at 1237 nm and 5.9916 ps/(km·nm) at 1384 nm [24, 25]. The estimated group delay is about 36 ps, after passing through the 2.69 km DL. The camera exposure time (9.21 μsec) is much larger than this group delay, therefore, we can ignore the dispersion effect caused by the long DL. Additionally, polarization mode dispersion (PMD) of the optical switch and the delay line are less than 0.1 ps and 0.3 ps, respectively. There values are much shorter than the camera speed, so it is negligible as the dispersion.
To summarize, we have demonstrated a new SD-OCT design for BD that has clear advantages over previously reported systems. An ultrafast optical switch and a time delay line were added in order to accurately detect interference fringes with opposite phases using a single camera. In comparison with conventional SD-OCT, the sensitivity was improved by more than 6 dB, which improved image quality by eliminating AC and DC noise.
This study was supported by a grant of the Korean Health Technology R&D Project, Ministry of Health & Welfare, Republic of Korea (HI13C1501), the Original Technology Research Program for Brain Science through the National Research Foundation of Korea (NRF) funded by the Ministry of Science ICT and Future Planning (NRF-2015M3C7A10290304), the new growth power equipment competitiveness reinforcement program (10047580), the Industrial Strategic Technology Development Program (10047943) funded by the Ministry of Trade, industry & Energy, and the “APRI Research Program” through a grant provided by the Gwangju Institute of Science and Technology in 2015. All correspondence to Beop-Min Kim, E-mail: firstname.lastname@example.org; Tae Joong Eom, E-mail: email@example.com.
References and links
1. D. Huang, E. A. Swanson, C. P. Lin, J. S. Schuman, W. G. Stinson, W. Chang, M. R. Hee, T. Flotte, K. Gregory, C. A. Puliafito, and A. Et, “Optical coherence tomography,” Science 254(5035), 1178–1181 (1991). [CrossRef] [PubMed]
5. K. Takada, “Noise in optical low-coherence reflectometry,” IEEE J. Quantum Electron. 34(7), 1098–1108 (1998). [CrossRef]
8. W.-C. Kuo, Y.-S. Lai, C.-M. Lai, and Y.-S. Huang, “Balanced detection spectral domain optical coherence tomography with a multiline single camera for signal-to-noise ratio enhancement,” Appl. Opt. 51(24), 5936–5940 (2012). [CrossRef] [PubMed]
9. H. Wang, Y. Pan, and A. M. Rollins, “Extending the effective imaging range of Fourier-domain optical coherence tomography using a fiber optic switch,” Opt. Lett. 33(22), 2632–2634 (2008). [CrossRef] [PubMed]
10. S.-W. Lee, H.-W. Jeong, and B.-M. Kim, “High-speed spectral domain polarization- sensitive optical coherence tomography using a single camera and an optical switch at 1.3 µm,” J. Biomed. Opt. 15(1), 010501 (2010). [CrossRef] [PubMed]
11. K. S. Park, W. J. Choi, T. J. Eom, and B. H. Lee, “Single-camera polarization-sensitive full-field optical coherence tomography with polarization switch,” J. Biomed. Opt. 18(10), 100504 (2013). [CrossRef] [PubMed]
12. J. Ai and L. V. Wang, “Spectral-domain optical coherence tomography: Removal of autocorrelation using an optical switch,” Appl. Phys. Lett. 88(11), 111115 (2006). [CrossRef]
13. R. Huber, M. Wojtkowski, and J. G. Fujimoto, “Fourier Domain Mode Locking (FDML): A new laser operating regime and applications for optical coherence tomography,” Opt. Express 14(8), 3225–3237 (2006). [CrossRef] [PubMed]
14. D. Nankivil, A.-H. Dhalla, N. Gahm, K. Shia, S. Farsiu, and J. A. Izatt, “Coherence revival multiplexed, buffered swept source optical coherence tomography: 400 kHz imaging with a 100 kHz source,” Opt. Lett. 39(13), 3740–3743 (2014). [CrossRef] [PubMed]
15. A.-H. Dhalla, K. Shia, and J. A. Izatt, “Efficient sweep buffering in swept source optical coherence tomography using a fast optical switch,” Biomed. Opt. Express 3(12), 3054–3066 (2012). [CrossRef] [PubMed]
16. J. Zhang, J. Jing, P. Wang, and Z. Chen, “Polarization-maintaining buffered Fourier domain mode-locked swept source for optical coherence tomography,” Opt. Lett. 36(24), 4788–4790 (2011). [CrossRef] [PubMed]
17. M. Gora, K. Karnowski, M. Szkulmowski, B. J. Kaluzny, R. Huber, A. Kowalczyk, and M. Wojtkowski, “Ultra high-speed swept source OCT imaging of the anterior segment of human eye at 200 kHz with adjustable imaging range,” Opt. Express 17(17), 14880–14894 (2009). [CrossRef] [PubMed]
19. B. Baumann, M. Pircher, E. Götzinger, and C. K. Hitzenberger, “Full range complex spectral domain optical coherence tomography without additional phase shifters,” Opt. Express 15(20), 13375–13387 (2007). [CrossRef] [PubMed]
20. H.-W. Jeong, J.-G. Lim, H.-J. Kim, W. Chung, and B.-M. Kim, “Complex artifact suppression using vestigial sideband filter in Fourier-domain optical coherence tomography,” Opt. Lett. 37(23), 4859–4861 (2012). [CrossRef] [PubMed]
21. American National Standard for Safe Use of Lasers ANSI Z136, 1–2007 (American National Standards Institute, Inc., 2007)
22. R. A. Leitgeb, R. Michaely, T. Lasser, and S. C. Sekhar, “Complex ambiguity-free Fourier domain optical coherence tomography through transverse scanning,” Opt. Lett. 32(23), 3453–3455 (2007). [CrossRef] [PubMed]
24. Halina Abramczyk, “Dispersion phenomena in optical fibers,” http://www.mitr.p.lodz.pl/evu/wyklady/.
25. Corning® SMF-28® Optical Fiber, http://course.ee.ust.hk/elec342/readings/corning%20smf-28.pdf.