Soliton microcomb based spectral domain optical coherence tomography

TL;DR

This paper investigates the use of soliton microcombs generated in photonic chipscale resonators as a novel illumination source for spectral domain optical coherence tomography, demonstrating advantages in bandwidth, noise performance, and potential for advanced imaging techniques.

Contribution

It introduces the application of soliton microcombs in SD-OCT, showing their superior bandwidth, lower noise floor, and potential for improved imaging and circular ranging capabilities.

Findings

Soliton microcombs can exceed the bandwidth of commercial SLDs.

Soliton states exhibit a noise floor about 3 dB lower than SLDs at the same power.

Experimental SD-OCT imaging demonstrates the viability of soliton microcombs for biomedical imaging.

Abstract

Spectral domain optical coherence tomography (SD-OCT) is a widely used and minimally invaive technique for bio-medical imaging [1]. SD-OCT typically relies on the use of superluminescent diodes (SLD), which provide a low-noise and broadband optical spectrum. Recent advances in photonic chipscale frequency combs [2, 3] based on soliton formation in photonic integrated microresonators provide an chipscale alternative illumination scheme for SD-OCT. Yet to date, the use of such soliton microcombs in OCT has not yet been analyzed. Here we explore the use of soliton microcombs in spectral domain OCT and show that, by using photonic chipscale Si3N4 resonators in conjunction with 1300 nm pump lasers, spectral bandwidths exceeding those of commercial SLDs are possible. We demonstrate that the soliton states in microresonators exhibit a noise floor that is ca. 3 dB lower than for the SLD at identical power, but can exhibit significantly lower noise performance for powers at the milliWatt level. We perform SD-OCT imaging on an ex vivo fixed mouse brain tissue using the soliton microcomb, alongside an SLD for comparison, and demonstrate the principle viability of soliton based SD-OCT. Importantly, we demonstrate that classical amplitude noise of all soliton comb teeth are correlated, i.e. common mode, in contrast to SLD or incoherent microcomb states [4], which should, in theory, improve the image quality. Moreover, we demonstrate the potential for circular ranging, i.e. optical sub-sampling [5, 6], due to the high coherence and temporal periodicity of the soliton state. Taken together, our work indicates the promising properties of soliton microcombs for SD-OCT.

Equations

$$\pm z_{\max }=\pm \frac{1}{4\delta k_{\mathrm{CCD}}}$$

$$z_{\mathrm{ambiguity}}=\frac{c}{2n_{\mathrm{tissue}}}\frac{1}{f_{\mathrm{rep}}}\approx \frac{c}{2n_{\mathrm{tissue}}}\frac{2\pi }{{\mathrm{D}}_1}$$

Full text

Soliton microcomb based spectral domain optical coherence tomography

Paul J. Marchand

paul-james.marchand@polymtl.ca

Swiss Federal Institute of Technology Lausanne (EPFL), Laboratoire d’optique biomédicale (LOB), Lausanne, CH-1015, Switzerland

Department of Electrical Engineering, École Polytechnique de Montréal, Canada

  

J. Connor Skehan

Institute of Physics, Swiss Federal Institute of Technology Lausanne, CH-1015, Switzerland

  

Johann Riemensberger

Institute of Physics, Swiss Federal Institute of Technology Lausanne, CH-1015, Switzerland

  

Jia-Jung Ho

Institute of Physics, Swiss Federal Institute of Technology Lausanne, CH-1015, Switzerland

  

Martin H. P. Pfeiffer

Institute of Physics, Swiss Federal Institute of Technology Lausanne, CH-1015, Switzerland

  

Junqiu Liu

Institute of Physics, Swiss Federal Institute of Technology Lausanne, CH-1015, Switzerland

  

Christoph Hauger

Carl Zeiss Meditec AG, Rudolf-Eber-Straße 11, 73447 Oberkochen, Germany

  

Theo Lasser

Swiss Federal Institute of Technology Lausanne (EPFL), Laboratoire d’optique biomédicale (LOB), Lausanne, CH-1015, Switzerland

  

Tobias J. Kippenberg

tobias.kippenberg@epfl.ch

Institute of Physics, Swiss Federal Institute of Technology Lausanne, CH-1015, Switzerland

Spectral domain optical coherence tomography (SD-OCT) is a widely used and minimally invasive technique for bio-medical imaging Drexler and Fujimoto (2015). SD-OCT typically relies on the use of superluminescent diodes (SLD), which provide a low-noise and broadband optical spectrum. Recent advances in photonic chipscale frequency combs Kippenberg et al. (2018); Kues et al. (2019) based on soliton formation in photonic integrated microresonators provide an chipscale alternative illumination scheme for SD-OCT. Yet to date, the use of such soliton microcombs in OCT has not yet been analyzed. Here we explore the use of soliton microcombs in spectral domain OCT and show that, by using photonic chipscale \ceSi3N4 resonators in conjunction with 1300 nm pump lasers, spectral bandwidths exceeding those of commercial SLDs are possible. We demonstrate that the soliton states in microresonators exhibit a noise floor that is ca. 3 dB lower than for the SLD at identical power, but can exhibit significantly lower noise performance for powers at the milli-Watt level. We perform SD-OCT imaging on an ex vivo fixed mouse brain tissue using the soliton microcomb, alongside an SLD for comparison, and demonstrate the principle viability of soliton based SD-OCT. Importantly, we demonstrate that classical amplitude noise of all soliton comb teeth are correlated, i.e. common mode, in contrast to SLD or incoherent microcomb states Ji et al. (2019), which should, in theory, improve the image quality. Moreover, we demonstrate the potential for circular ranging, i.e. optical sub-sampling Siddiqui and Vakoc (2012); Siddiqui et al. (2018), due to the high coherence and temporal periodicity of the soliton state. Taken together, our work indicates the promising properties of soliton microcombs for SD-OCT.

First demonstrated in 1991 by Huang Huang et al. (1991), optical coherence tomography (OCT) has become an important technique for non invasive imaging of biological tissues Fercher et al. (2003). Today, OCT is a standard diagnostic tool in ophthalmology and has been extended to intravascular imaging Kassani et al. (2017) and brain imaging Vakoc et al. (2009); Bolmont et al. (2012); Srinivasan et al. (2012). Over the past decade, frequency domain methods (FD-OCT), i.e. spectral-domain OCT (SD-OCT) and swept-source OCT (SS-OCT), have superseded time domain OCT through their higher sensitivity Leitgeb et al. (2003); de Boer et al. (2003); Choma et al. (2003); Choi and Wang (2015); Drexler (2004). Since then, light sources and detectors for FD-OCT (both SD and SS-OCT) have been improved, providing low noise, larger bandwidths and faster acquisition rates. Recently, sources comprised of a set of discrete frequencies have been proposed for FD-OCT, as they offer an increased depth-sensitivity Tsai et al. (2009); Bajraszewski et al. (2008), reduced power exposure while maintaining a high axial resolution Jung et al. (2008) and an extended imaging range through optical-domain subsampling Siddiqui et al. (2018). The periodicity in the tomogram offered by this novel acquisition scheme enables significantly extending the OCT imaging range in a data efficient manner and shows great promise for imaging of non-planar samples, such as in intra-operative scenarios Siddiqui and Vakoc (2012).

One promising implementation of such discrete sources for SD-OCT are soliton microcombs. First discovered in 2007, these microcombs are generated by the nonlinear conversion processes inside micro-resonators Del’Haye et al. (2007); Kippenberg et al. (2011). Through adjustment of laser power and detuning, a dissipative Kerr soliton (DKS) state can be excited, providing coherence lengths and bandwidths comparable to continuous-wave and femtosecond lasers, respectively Herr et al. (2014). The spectrum of a DKS state consists of fully coherent laser lines with linewidths equal to the CW pump laser linewidth (typically $\sim 100\,\mathrm{kHz}$), resulting in kilometer scale coherence lengths. Its overall spectral bandwidth can be tailored via dispersion engineering Okawachi et al. (2014) and can reach up to octave-spanning coverage Pfeiffer et al. (2017). In addition to their spectral properties, recent advances in fabrication technology have significantly reduced the power requirements for DKS generation, thus allowing for direct integration with semiconductor pump lasers Liu et al. (2018a); Stern et al. (2018). Altogether, through their exceptional optical properties and wafer-scale fabrication, DKS microcombs are promising candidates as sources for imaging and in particular OCT. Here, we demonstrate for the first time the use of a soliton microcomb for SD-OCT.

Dissipative Kerr solitons as a source for SD-OCT. We designed novel microcombs sources for OCT imaging operating in the second optical window (NIR-II), at 1300 nm, for its relatively low water absorption and reduced tissue scattering properties. We fabricated three \ceSi3N4 resonators (one shown in the inlet of Fig. 1 e)) following the established photonic Damascene process Pfeiffer et al. (2016) with free spectral range (FSR) of $\sim 100,\,\ 200\,\ \mathrm{GHz}$ and $1\,\ \mathrm{THz}$, respectively (Fig. 1 e). Through their large waveguide cross sections, the resonators achieve anomalous group velocity dispersion (GVD) in the NIR-II imaging window (see Methods and S.I. for details). A microcomb, as shown in Fig. 1 a), is generated by the nonlinear frequency conversion processes inside a micro-resonator Kippenberg et al. (2011). The mutual interplay between (non-)degenerate four-wave mixing processes and self- and cross-phase modulations provides an optical gain to the resonator modes adjacent to the pumped mode. The Kerr comb generation is achieved by sweeping the pump laser frequency from the effective blue-detuned to a defined point at the effective red-detuned side of the selected cavity resonance. For DKS comb generation, the laser sweeping typically stops at a multi-soliton state and proceeds to a single soliton state through a backward frequency tuning technique Guo et al. (2016).

As illustrated in Fig. 2, the nonlinear frequency conversion bandwidth of the 1 THz microcombs can readily reach and exceed the bandwidth of SLDs. This is demonstrated for two distinctly different operational Kerr frequency comb states: the DKS and the chaotic modulation instability (MI) states (shown in Fig. 2 c). The DKS state, shown in green, exhibits a characteristic $\mathrm{sech}^{2}$ spectral envelope and reaches a spectral coverage similar to the reference SLD source. The cross section of the 1 THz DKS waveguide, $780\times 1450\,\mathrm{nm^{2}}$, provides an anomalous GVD ($\mathrm{D_{2}/2\pi}\sim 40\,\mathrm{MHz}$) for soliton pulse formation. The 3 dB bandwidth of the DKS spectrum, highlighted in Fig. 2 c), is $\sim 8.3\,\mathrm{THz}$, corresponding to a 38 fs transform limited pulse.

The chaotic Kerr comb state, shown in blue in Fig. 2 c), provides a spectral coverage well exceeding the SLD’s, due to the lower GVD ($\mathrm{D_{2}/2\pi}\sim 20\,\mathrm{MHz}$) originating from its smaller micro-resonator cross section ($730\times 1425\,\mathrm{nm^{2}}$). The resulting spectral envelope is overall flat but, in contrast to the DKS state, exhibits local power variations caused by avoided mode crossings.

Noise characteristics of soliton microcombs. To assess the noise characteristics of these novel sources and their applicability to OCT imaging, we first measure their relative intensity noise $\mathrm{RIN}=\frac{S_{P}(f)}{\langle P^{2}\rangle}$, with $S_{P}(f)$ denoting the single sided power spectral density of the intensity fluctuations (shown in Fig. 2 d), and demonstrate that while the MI state provides a broader spectral coverage, its chaotic nature results in an increase in RIN of nearly 20 dB, extending to very high offset frequencies in the GHz domain (Fig. 2 d) Herr et al. (2012). These measurements were performed for different FSRs (i.e. $100$, $200$ $\mathrm{GHz}$ and $1$ $\mathrm{THz}$), and resulted in similar RIN profiles between resonators (data not shown here). Accordingly, although chaotic comb states in a $\mathrm{Si_{3}N_{4}}$ microresonator have been demonstrated in OCT imaging Ji et al. (2019), their higher noise should ultimately limit OCT performance as compared to SLDs, especially at elevated imaging speeds. Meanwhile, we also show that the DKS soliton state has comparable intensity noise with the SLD, at frequencies higher than 10 kHz. In the low frequency regime, mechanical modes of the input and output lensed fiber-coupling result in a broad noise peak spanning from 100 to 1000 Hz for both the MI and DKS states, which can be mitigated through optimized packaging or feedback loops.

Even more so, the ultimate performance limit of coherent sources at high offset frequencies, such as the DKS comb, is given by the photon shot noise (RIN = $\frac{2\hbar\omega}{P}=\mathrm{-145~{}dBc/Hz}$ with 20 $\mu$W power on the detector) and improves with optical power. In contrast, in the case of broadband, incoherent light sources, the RIN is limited by spontaneous emission beat noise Yurek et al. (1986); Baney (1998) ($\mathrm{RIN}=1/B_{0}=\mathrm{-136~{}dBc/Hz}$ for a 45 THz rectangular bandwidth SLD source), which ultimately limits the dynamic range gain with high source powers in the reference arm Sorin and Baney (1992).

Next, we explore one unique feature of soliton microcombs; the high-degree of coherence between individual comb lines. This is especially important in the context of OCT, as line-by-line intensity noise of the frequency comb’s retrieved spectra (Fig. 3 a) and c)) corresponds to pixel-by-pixel noise in the retrieved image (Fig. 3 b) and d)). Indeed, as an image in SD-OCT is produced via a Fourier transform of the interferogram, only uncorrelated intensity noise between various pixels degrades the final image Yun et al. (2004). Noise in the amplitude of the spectrum’s enveloppe will be act only on the DC component of the tomogram (Fig. 3 a) and b), whereas uncorrelated intensity fluctuations between the different optical frequencies will lead to a higher noise level at all depths of the tomogram (Fig. 3 c) and d)). To investigate these intra-tone noise properties, we performed the cross-correlation of intensity fluctuations on pairs of comb lines, using the experimental setup described in Fig. 3 e) Siegman and Siegman (1971). From both DKS or MI combs, individual comb lines are filtered and time traces are recorded and cross-correlated (Fig. 3 f) and g) for various sampling speeds. The corresponding cross power spectral densities (PSD) are depicted in Figures 3 g) and i). In practice, we chose two lines, at 1272 $\mathrm{nm}$ and at 1320 $\mathrm{nm}$ (lines 1 and 2 respectively in Fig. 3). In the DKS state, we observe a peak correlation coefficient between the two chosen lines of approximately 0.95, corresponding to a sampling rate of $\mathrm{500\,\ kSa/s}$. The maximum correlation coefficient near zero lag stays well above 0.8 for sampling frequencies up to $\mathrm{5\,\ MSa/s}$, indicating that intensity noise between DKS comb lines is highly correlated even at elevated frequencies. In contrast, for the fully developed MI state, the maximum correlation coefficient between lines 1 and 2 is approximately 0.24, and occurs for the lowest sampling speed (DC). For all higher sampling frequencies, however, the correlation coefficient decreases to approximately 0.01, indicating highly uncorrelated intensity noise between comb lines. We expect a similar behavior for the the classical noise of nearly all incoherent sources, including for SLD sources.

As mentioned earlier, given that the ultimate limit of the noise properties of frequency domain OCT is set by the degree of correlation of intensity noise between various spectral channels Yun et al. (2004), and therefore different optical frequencies, the DKS state can offer significant advantages, in terms of noise, as compared to the MI state. In view of these differences in noise performances, as well as the DKS’s superior nonlinear efficiency and bandwidth, we chose to use a DKS source for the OCT experiments presented here.

Spectral characteristics of microcombs and their implications for OCT imaging and circular ranging. In frequency domain OCT, depth-resolved information about the sample is conveyed through the amplitude and frequency of an inteferogram. A reflectivity profile is obtained through a Fourier transform of the recorded spectrum on the spectrometer. From sampling theory, the maximum imaging depth obtainable $z_{\mathrm{max}}$ is therefore dictated by the spectrometer’s spectral resolution $\delta k_{\mathrm{CCD}}$ as Izatt and Choma (2008):

[TABLE]

As such, OCT systems designed for high axial resolution and deep penetration imaging require a detection capable of registering a broadband spectra at a fine spectral resolution. In practice, combining these two features is cumbersome in SD-OCT due to the limited length of current array detectors (typically between 1024 and 2048, and exceptionally 8196 pixels Lichtenegger et al. (2018)), ultimately limiting either the effective resolution or the available imaging range.

When comb-like sources, such as Kerr combs, are employed instead of a continuous spectrum, the discrete set of frequencies will generate a periodicity in the tomogram if the frequency/time difference between the combs is sampled by the detector Siddiqui et al. (2018). The frequency of this periodicity, called the ambiguity range, is determined by the source’s repetition rate $f_{\mathrm{rep}}$ (which also corresponds to the temporal separation between the individual pulses). For Kerr combs, the repetition rate is given by the micro-resonator FSR ($\mathrm{D_{1}/2\pi}$):

[TABLE]

with the speed of light c and the tissue refractive index $n_{\mathrm{tissue}}$. For the imaging experiments carried out below, we used micro-resonators with a $1\,\mathrm{THz}$ FSR, leading to an ambiguity range of $\sim 71\,\mathrm{\mu m}$ compared to a maximum imaging range of $\sim 2\,\mathrm{mm}$ offered by the spectrometer. In contrast, the lower FSR DKS sources shown in Fig. 1 e) offer repetition rates down to 100 GHz, corresponding to an increased ambiguity range of $\sim 710\,\mathrm{\mu m}$.

In addition to their discreteness in frequency, DKS sources also possess interesting temporal coherence properties. Although the overall coherence length of the source is dictated by its bandwidth, the coherence length of each comb tone of the DKS source equals that of the driving pump laser and thus amounts to several kilometers for a pumping linewidth around $100\,\mathrm{kHz}$. As mentioned earlier and highlighted in Equation 1, the attainable imaging range in FD-OCT is typically dictated either by the spectral resolution of the spectrometer or by the width of the swept spectral line (for spectral-domain and swept-source respectively). When combining DKS sources with an SD-OCT system, a mismatch can therefore occur between the imaging range (given by the spectrometer, here $\sim 2$ mm) and the coherence length of each comb tone (here $>2$ km). As such, the coherence lengths reached here largely exceed the imaging ranges of typical OCT systems, entailing novel advantages and disadvantages for imaging, which will be discussed below.

OCT imaging with a DKS microcomb. The difference in performance between the SLD and the DKS as sources for OCT imaging was qualitatively assessed by imaging a $\sim 50\,\ \mu$m thick slice of a mouse brain tissue. The OCT was equipped with a 40$\times$ 0.8 NA objective (Olympus) to obtain a lateral resolution of $\sim 1.5\,\ \mu$m and a depth-of-field shorter than the source’s ambiguity range. In a first step, we imaged the slice using the SLD source, providing an axial resolution of $\sim$ 6 $\mu$m in air. Figure 4 a) presents en-face views over a $200\times 200\,\ \mu$m2 area at specific depths, whereas panel b) shows the cross-section images of the SLD based OCT tomogram. These views present similar features as other OCT observations of cerebral tissues Srinivasan et al. (2012), such as neural fibers (pointed by white arrows), which appear as directional, bright, and fine structures over dim neuropil. Within the neuropil, darker circular structures seemingly point to the presence of neuronal cell bodies, as already observed in high resolution OCT Srinivasan et al. (2012); Assayag et al. (2013).

Secondly, without modifying any imaging parameters nor touching the sample, the SLD was disconnected from the system and replaced with the DKS source, providing an axial resolution of $\sim$ 10 $\mu$m in air. Figures 4 c), d) and e) show the OCT tomogram of the same sample with the DKS light source. The neural fibers can be clearly observed in the en-face views with higher contrast. However, the neuropil appears darker and fewer details can be discerned. Additional artifacts, as indicated by the red arrows, are present in some of the DKS views and are likely caused by the combination of two characteristics of the DKS source: its discrete set of frequencies and its narrow linewidth. Overall, the dynamic range obtained in the DKS images is reduced by $\sim$ 19 dB compared to the SLD. This discrepancy could originate from the significantly lower power provided by the DKS source (estimated to be up to a fourth of the SLD power) and from the presence of the spurious back-reflections, ultimately drowning the collection of weakly scattering features. For both sources, the A-scan rate was maintained at 46 kHz. The images presented in panels c) and d) were obtained by selecting only the comb tones from the interferograms, dismissing non-illuminated pixels. Conversely, for panel f) the entire recorded interferogram was used. More details on the processing are available in the S.I. The axial resolution of the SLD and the DKS were extracted using a reflective mirror (as shown in S.I. Fig. 6) and are $\sim$ 6 $\mu$m and $\sim$ 10 $\mu$m respectively.

As mentioned earlier, the frequency discretization of the source will lead to a periodic image folding along the axial dimension, as similarly observed by Siddiqui et al. Siddiqui et al. (2018). The ambiguity range of the source can be observed in the cross-section (Fig. 4 e) and manifests itself as an axial periodicity of the structures (orange arrows in Fig. 4 e). As the comb width is significantly narrower than the spectrometer’s spectral resolution, the coherence length of the DKS comb tones exceeds both the ambiguity range and the spectrometer’s imaging range (Fig. 1 b). The aforementioned image folding and extended coherence length thus allows reflections within the optical path to interfere with the reference arm, and will ultimately be superimposed with the features under investigation. As a result, some of the artifacts in the DKS images might stem from the folding of structures beyond the DKS’s ambiguity range, such as reflections from optical components and the coverslide (illustrated in Fig. 1 a) or from the back-scattering of cerebral structures. Some of the artefacts pointed by red arrows in Fig. 4 c) can be observed at deeper locations in the SLD’s tomogram, highlighted by green arrows in Fig. 4 b). Typically, these strong reflections will occupy a significant portion of the spectrometer’s dynamic range and could ultimately drown the fine details of the image, as previously observed in OCT Villiger et al. (2010); Blatter et al. (2011).

Future direction of the DKS based SD-OCT. In this manuscript, we have demonstrated for the first time the use of a DKS source for SD-OCT. We show that such soliton sources (DKS) are an interesting candidate for SD-OCT imaging through their low-noise, discrete set of frequencies and large bandwidths. Our work highlights the high noise performance of the source: specifically, the DKS (a coherent broadband source), equals and even outperforms an SLD (fully incoherent source) in its relative intensity noise (RIN). Equally important, DKS feature a unique property, in our knowledge previously unseen in OCT sources: the noise between the comb tones comprising the soliton frequency combs shows an unprecedentedly high degree of correlation. This feature is particularly important in OCT, as images are obtained through a Fourier transform of the spectrum. As such, noise common to all comb tones does not degrade the dynamic range, whereas relative uncorrelated fluctuations from pixel to pixel contribute to a significant dynamic range reduction Yun et al. (2004). With the noise of the source characterized, we imaged ex vivo mice fixed brain slices, and found that the novel source allows for visualization of similar features to an SLD source, although with an overall reduced dynamic and imaging range. Overcoming these pitfalls can be achieved by optimizing both the OCT instrument and the source. First, the artefacts present in Fig. 4 c)-e) could be suppressed either by using solely reflective optical elements Amirsolaimani et al. (2017) or through a dark-field implementation Villiger et al. (2010); Blatter et al. (2011). As these spurious reflections can occupy a significant portion of the dynamic range of the camera, eliminating these features could help further enhance the system’s imaging capabilities. Second, the $\sim 71\,\mathrm{\mu m}$ ambiguity range available with 1 THz DKS is too short for most imaging applications. It is however sufficient for imaging of thin flat tissues and for certain optical biopsy applications Belykh et al. (2018a); Sanai et al. (2011); Schlosser and Bojarski (2011); Charalampaki et al. (2015); Zehri et al. (2014); Fugazza et al. (2016); Wellikoff et al. (2015); Bui et al. (2015); Fuks et al. (2018); Krafft et al. (2018); Belykh et al. (2018b); Lombardini et al. (2018), wherein there is need for a real-time assessment of brain and tumor tissue on a cellular level, as patient survival has been shown to be correlated to the extent of tumor resection Stummer et al. (2006). Moreover, as shown in Fig. 1 e), DKS sources with shorter FSRs down to 100 and 200 GHz are already available with similar noise profiles as the one used here for imaging. These sources would enable reaching ambiguity ranges up to $\sim 0.7$ mm, which are compatible with most in vivo imaging applicationsSiddiqui et al. (2018). Third, fully exploiting the circular ranging capabilities of the source requires reading the interferograms in a complex-valued form Siddiqui and Vakoc (2012), which can be attained by adding acousto-optic frequency shifters to the system Lippok et al. (2019); Bachmann et al. (2006). Lastly, the central wavelength of 1300 nm used here is not suitable for all in vivo applications, especially human ophtalmology. Nevertheless, the source’s design can be modified, enabling shifting of the central wavelength to shorter spectral ranges, such as 1 $\mu$m, as demonstrated previously Karpov et al. (2018); Lee et al. (2017).

Overall, in addition to the unprecedented noise performance of the DKS source and the increased imaging efficiency available through optical-domain subsampling, frequency combs could potentially alleviate certain shortcomings of SD-OCT detection schemes by facilitating $\lambda$-to-k mapping and reducing depth dependant sensitivity roll-off Bajraszewski et al. (2008); Tsai et al. (2009). DKS sources could also lead to higher axial resolutions at 1300 nm: as highlighted in Fig. 2 c), the power spectral density of the DKS source exceeds the SLD’s from $\sim$ 1250 nm to 1500 nm. As such, using spectral shaping, the DKS could provide a bandwidth comparable or larger than current broadband SLDs used for 1300 nm imaging.

The high performance of the DKS source could lead to a significant miniaturization of the OCT system. The optical-domain sub-sampling capabilities of the source, highlighted in Fig. 4 c), already indicate a potential shortening of the reference arm of $\sim 2$ mm. Furthermore, although not demonstrated here, the long coherence length of the DKS combs could enable further shortening of the length of the reference arm, reducing the instrument’s footprint. In traditional SD-OCT systems, the path delay difference between the reference and sample arms needs to be smaller than the maximum imaging range of the spectrometer to record an interference. In the case of a frequency comb, this condition is alleviated through optical sub-sampling, so long as the path delay difference is within the coherence length of each line of the source. As the DKS source used in this study has a theoretical coherence length for each comb line beyond a kilometer, the reference arm length could be significantly shortened, ultimately paving the way to future miniaturized and potentially more efficient high-resolution OCT imaging systems. Lastly, the optical-domain subsampling properties of our source would be highly valuable in human in vivo imaging, wherein the sample geometry is often non-planar and features could exceed the imaging range, such as in ophtalmology and intra-operative OCT. Altogether, the aforementioned noise and spectral properties of DKS microcombs hint to their significant unexplored potential for future exploitation in SD-OCT.

Data availability statement

The data and code used to produce the results of this manuscript will be available on Zenodo upon publication.

Authors contributions

P.J.M. and M.H.P.P. performed the experiments. P.J.M., J.-J.H., M.H.H.P. analyzed the data. J.L. designed and fabricated the \ceSi3N4 chip devices. P.J.M. designed and constructed the OCT setup. T.J.K., T.L., and C.H. supervised the project. J.C.S and J.R performed and analyzed all noise-related measurements. All of the authors contributed to the manuscript.

Acknowledgements

This work was supported by an industrial grant with Carl Zeiss AG "Optical coherence tomography with chip-scale Kerr soliton frequency combs". \ceSi3N4 micro-resonator samples were fabricated in the EPFL Center of MicroNanotechnology (CMi). This publication was supported by the Swiss National Science Foundation under grant agreement No. 165933. P.J.M. acknowledges that part of his contribution was undertaken thanks to funding from the Canada First Research Excellence Fund through the TransMedTech Institute and from the Swiss National Science Foundation.

Methods

Here we describe the experimental realization of Kerr comb based SD-OCT. Figure 2 shows the experimental setting consisting of two distinct setups located in buildings spaced by about $\sim 700\,\mathrm{m}$. A fiber link connects the setup for DKS generation and the SD-OCT setup between the two laboratories.

Micro-resonator fabrication. The samples employed are $1\,\mathrm{THz}$ FSR micro-resonators formed by \ceSi3N4 waveguides. Figure 1 e) shows the micro-resonator used in this work. These resonators were fabricated using the photonic Damascene process which avoids common processing challenges of thick \ceSi3N4 films Pfeiffer et al. (2016, 2018a) and has recently allowed for micro-resonator Q factors exceeding 10 million Liu et al. (2018a). The continuous wave pumping light is coupled into the \ceSi3N4 chips via a double inverse taper Liu et al. (2018b). For the 1 THz DKS comb, the cross section is $1.45\times 0.78\,\mathrm{\mu m^{2}}$ while for the chaotic comb, it is $1.425\times 0.73\,\mathrm{\mu m^{2}}$. The bus waveguides (design width $0.55\,\mathrm{\mu m}$ for DKS and $0.525\,\mathrm{\mu m}$ for the chaotic comb) couple the light into the ring resonators ($22.71\,\mathrm{\mu m}$ radius) are mode matched to excite the fundamental $\mathrm{TM_{00}}$ mode. The resonance linewidth is below $100\,\mathrm{MHz}$ as has been measured in the recent publication Pfeiffer et al. (2018b). The waveguide cross-section of the 100 and 200 GHz FSR DKS microresonators is $1.52\times 0.82\,\mathrm{\mu m^{2}}$. The simulated dispersion profiles, including the modal dispersion ($D_{\mathrm{2}}$) and the modal deviation from the resonance frequency of the nearest mode ($D_{\mathrm{int}}$) can be found in figure 5.

Kerr comb generation. The DKS light source is pumped by a $1300\,\mathrm{nm}$ external cavity diode laser, which is amplified up to $\sim 650\,\mathrm{mW}$ power using a semiconductor optical amplifier (SOA). The amplified light is coupled to the silicon nitride micro-resonator chip via lensed fibers. The pump polarization can be adjusted via a paddle controller and both the power before and after the chip are monitored via power meters (PM). We estimate a soliton excitation power in the bus waveguide of $\sim 290\,\mathrm{mW}$. An arbitrary function generator (AFG) provides the voltage ramp signal driving the laser frequency tuning. A standard voltage ramp tuning method Guo et al. (2016): through the voltage-ramp tuning, a multi-soliton state is excited which is then converted into a single soliton through backward tuning. A tunable fiber Bragg grating (FBG) is used to attenuate the residual pump light while the back-reflected pump is detected through a fast photodiode and shown on an oscilloscope (OSC) to monitor the laser tuning. The generated DKS spectrum is free of avoided mode crossings causing strong local power deviations typically originating from the multimodal nature of the waveguide.

OCT imaging. The generated DKS comb is then sent to a custom-built OCT setup through a $\sim 700\,\mathrm{m}$ long optical fiber link (to connect the source from one laboratory to the OCT setup in another laboratory). The SD-OCT setup was designed for a commercial SLD with a central wavelength $\lambda_{0}=1310\,\mathrm{nm}$ and bandwidth $\delta\lambda=150\,\mathrm{nm}$ (LS2000C, Thorlabs) and its detection is based on a highly sensitive spectrometer, as described previously Marchand et al. (2018). Both the source and detection are connected via a broadband fiber beam splitter (TW1300R5A2, Thorlabs) with a dispersion compensated reference arm and a sample arm comprising a galvo mirror scan unit (6210H, Cambridge Technologies), a high NA objective (LUMPLFLN-40XW, Olympus) and imaging optics. The scanner control and data readout are performed by a connected computer with a high-speed input. The output optical power of the SLD source is $\sim 9.15\,\mathrm{mW}$ while the DKS comb is $<2.3\,\mathrm{mW}$. All images were acquired at an A-scan rate of $46\,\mathrm{kHz}$. The post-processing steps, including k-space resampling and Fourier transformations were performed using a custom software implemented in MATLAB (Mathworks). The axial resolution of the SLD and DKS systems were characterized by placing a mirror in the front focal plane of the objective and were measured as $\sim 6\,\mathrm{\mu m}$ and $\sim 10\,\mathrm{\mu m}$ in air respectively.

Image processing. The images presented in Fig. 4 were obtained after Fourier transform of the spectral interferograms recorded by the spectrometer. Prior to visualization, the dynamic range of the data was reduced using first a logarithmic operation (10$\times$log10()) and a clipping operation (same operations for both DKS and SLD images). The data was then spatially smoothed using a median filter in MATLAB Schindelin et al. (2012), planes at different depths were selected. The clipping limits were obtained by taking the 0.01% and 99.9% intensity values of the imaged planes, after median filtering.

Background subtraction was performed, prior to Fourier transforming, by averaging each B-scan into a single background vector, which was then subtracted to the entire B-scan. This step was repeated for each B-scan of the volume. This step was performed for both SLD and DKS data. The DKS data was processed in two separate ways, as shown in Fig. 4, either by considering the entire interferogram or by selecting only the comb peaks. In the first processing method, the entire interferogram was considered, including non-illuminated pixels. The obtained A-scan for each position is of the same length as the spectral interferogram (Fig. 4 e). In the second method, the comb tones positions on the interferogram were first identified by computing the local maximal value around the tone. Using their positions, shorter interferogram were obtained by eliminating all other pixels (non comb tone pixels). The resulting A-scan were therefore significantly shorter than those obtained in the first method, and include solely one ambiguity range (Fig. 4 c-d). The dynamic range of the planes for both methods was 19 and 23 dB for the first and second processing pipelines respectively.

Brain tissue preparation. All animal procedures were carried out according to Swiss regulations under the approval of the veterinary authority of the canton of Vaud (protocols VD3056 and VD3058), are in-line with the 3Rs and follow the ARRIVE guidelines. After transcardiac perfusion, the brains of B6SJL/f1 mice were extracted, placed into 4% PFA overnight and then placed in a solution of 30% glucose. The brains were finally cut into slices of $\sim$ $50\,\mathrm{\mu m}$ using a microtome and placed on a glass coverslide. These samples had been prepared for previous studies Marchand et al. (2017); Nguyen et al. (2017), no new samples were prepared for this manuscript.

Supporting information

The dispersion profiles, including the GVD parameter ($\mathrm{D_{2}/2\pi}$) and the integrated dispersion ($\mathrm{D_{int}/2\pi}$) (Fig. 5) of the micro-resonators were simulated using COMSOL multiphysics® with the 2D axial symmetric model. The cross-section dimension is $1.45\times 0.78\,\mathrm{\mu m^{2}}$ for the DKS comb and $1.425\times 0.73\,\mathrm{\mu m^{2}}$ for the chaotic comb. Both the radius of the resonators were $22.71\,\mathrm{\mu m}$. The $\mathrm{TM_{00}}$ mode in the waveguides were calculated.

To characterize the imaging performance of the DKS as a light source for OCT imaging, we used a highly reflective mirror substrate as a sample. Figure 6 a) shows the DKS signal as recorded by the spectrometer’s image sensor without the mirror (i.e. signal from the reference arm). Interestingly, although the line width is significantly shorter than the camera’s spectral sampling, the comb line can be sampled on adjacent pixels as can be seen in the inset. From the Gaussian-like shape of the recorded peak, we believe this effect to be likely caused by the diffraction limited size of the spot on the camera. Other potential causes include electronic cross-talk between the pixels and a sub-optimal matching between comb teeth spacing and the detector pixel pitch. The apparent divergence between this observation and the conceptual illustration presented in Fig. 1 a) does not however impact the circular ranging abilities of the source, as the coherence length of a single comb tone still exceeds the imaging range of the spectrometer. After placing the mirror, we obtain the tomogram shown in logarithmic scale in Figure 6 b) by Fourier transforming the signal obtained from the spectrometer. A resolution of $\sim 10\,\mathrm{\mu m}$ and an ambiguity range of $\sim 71\,\mathrm{\mu m}$ are derived in the air. The signal attenuation over depth observed, also termed roll-off, is caused primarily by the spectrometer’s finite spectral sampling despite the fine width of the comb’s lines.