Fiber-integrated Brillouin microspectroscopy: towards Brillouin endoscopy

Brillouin imaging (BI) for micromechanical characterization of tissues and biomaterials is a fast-developing field of research with a strong potential for medical diagnosis of disease-modified tissues and cells. Although the principles of BI imply its compatibility with in vivo and in situ measurements, the integration of BI with a flexible catheter, capable of reaching the region of interest within the body, is yet to be reported. Here, for the first time, we experimentally investigate integration of the Brillouin spectroscope with standard optical fiber components to achieve a Brillouin endoscope. The performance of single-fiber and dual-fiber endoscopes are demonstrated and analyzed. We show that a major challenge in construction of Brillouin endoscopes is the strong backward Brillouin scattering in the optical fiber and we present a dual-fiber geometry as a possible solution. Measurements of Brillouin spectra in test liquids (water, ethanol and glycerol) are demonstrated using the dual-fiber endoscope...


Introduction
Micromechanical properties of tissues, cells and biomaterials present key information for understanding their physiological function and for evaluation of pathologic conditions. Brillouin imaging (BI) is an optical technique for mapping elasticity in three-dimensions with micron resolution, 1 and as such it can be used to assess micromechanical properties of biomaterials. BI is based on the principle of inelastic scattering of light by thermally activated acoustic vibrations or phonons. 2 Since the technique uses a light beam of relatively low power and requires no physical contact with the tissue, the measurement is thought to be noninvasive and suitable for in vivo diagnosis. [3][4][5] Application of Brillouin spectroscopy to biology and biomedicine was suggested in 1966, 6 but thē rst experimental observation of Brillouin spectra in a biologically-relevant material, gelatin gels, was achieved in 1976. 7 Shortly after, elastic constants were determined for collagen¯bers and muscle tissues. [8][9][10] In early work, Brillouin spectra were acquired using multi-stage scanning Fabry-Perot interferometers, 11 since high sensitivity and megahertz resolution are required for isolation of the Stokes and anti-Stokes (AS) signals from the unshifted elastically scattered light. Although tandem scanning Fabry-Perot interferometers yield spectra with high signal-to-noise ratios, the measurement at a single point inside the sample typically takes long time, from minutes to hours, restricting the applicability of the method for point measurements only. In 2005, Koski and Yarger introduced a new spectrometer-based on a tilted Fabry-Perot etalon. 1 The angle-dispersive spectrometer simpli¯ed the system signi¯cantly and allowed to accomplish Brillouin measurements at faster rates, compared to scanning Fabry-Perot interferometers. An addition of high-re°ectivity coatings and a transmission window on the surface of the etalon improved the power e±ciency of the spectrometer and resulted in a new instrument, so-called virtual-image phasearray (VIPA). 12 Use of VIPA spectrometers makes it possible to acquire a three-dimensional map of object's elasticity in just a few minutes, a timescale better suited for biomedical imaging. 3 The main technological developments in the¯eld of BI to date have concentrated on improvements in the spectral resolution, the sensitivity or the acquisition time of the spectrometers used for isolation of Brillouin signals. [14][15][16][17] The sample-facing part of the optical setup was kept unchanged, heavily relying on bulk optical elements or being integrated with a confocal microscopy system. 3 Thus, so far Brillouin microscopes and spectroscopes are mostly con¯ned to research laboratory environment, with the system footprint requirement of a full optical table and constant adjustment of the optical alignment. To realize the technique's potential for in vivo and in situ biomedical imaging, for which Brillouin signals need to be measured at a given location inside the body calls for system integration with°exible and miniature probes to replace bulk optics and reduce the system complexity and footprint. Fiber-optical probes compatible with standard size°exible endoscopes (1-2 mm) could be an ideal solution for achieving scalable and portable Brillouin imaging systems.
In this paper, we discuss the route taken towards integration of the Brillouin spectroscope with ā ber-optical probe for the laser light delivery and the scattered signal collection. We present, for thē rst time to our knowledge Brillouin scattering measurements obtained by means of a¯ber-optical probe. We demonstrate Brillouin microspectroscopy measurement using two types of probe, based on single-and dual-¯ber geometries. We¯nd that construction of a¯ber probe for Brillouin endoscopy presents similar challenges to that of Raman endoscopy. 18 Brillouin scattering from inside the optical¯ber results in a strong signal that swamps that from the sample under test. We show that this can be overcome by using a dual¯ber probe which uses separate¯bers for delivery and collection of the scattered signal. The e±ciency of the scattered light collection in our¯ber probes is analyzed numerically and an optimization strategy is presented.

Brillouin Scattering
In Brillouin scattering, an incoming photon (at the angular frequency ! i and the wave vector k i ) interacts with a traveling thermal phonon (, K), resulting in the photon being scattered at an angle Â ¼ \ðk i ; k s Þ ( Fig. 1) and with a Doppler frequency shift equal to the phonon frequency, i.e., ! s ¼ ! i AE . The sign AE corresponds to creation or annihilation of a phonon in the scattering event. Energy and momentum conservation laws lead to the relationship for the frequency shift ¼ 2nv sinðÂ=2Þ, where n is the refractive index of the scattering medium, v is the acoustic speed and is the optical wavelength (in vacuum).
The scattering angle Â is usually determined by the geometry of the experiment, and in this work we examine only situations when the scattering angle is 180 or backscattering geometry. In this situation, the frequency shift is simply ¼ 2nv . We can immediately see that the Brillouin frequency shift is proportional to the acoustic speed v, that in turn is a function of the longitudinal elastic modulus M ¼ v 2 , with being the material density. In general, di®erent acoustic modes propagate with di®erent velocities. However, in the backscattering geometry (Â ¼ 180 ) and in isotropic media, only longitudinal acoustic waves can be probed, resulting in a single Brillouin doublet arising in the spectrum of scattered light. Biological tissues and cells generally have high liquid content (> 70%) and Poisson ratios % 0:5 and are therefore very di®erent in their mechanical properties compared to solids. The mechanical response of biomaterials and tissues can be approximately described by visco-elastic theory, according to which the longitudinal modulus becomes a complex number M ¼ M 0 þ iM 00 . In this description the acoustic velocity is related to the real part of is associated with the acoustic damping and is proportional to the imaginary part of the modulus. The relation between the Brillouin frequency shift and the more conventional mechanical parameters such as Young's modulus and shear modulus in biological tissues remains the subject of discussion. 19,20 That is, however, beyond the scope of this paper.

Experimental Methods
Three optical con¯gurations as shown in Fig. 2 were used to record Brillouin spectra of liquid samples: two¯bre-probe arrangements, and for reference purposes, a bulk-optic arrangement. Figure 2(a) is a bulk-optic confocal arrangement as previously reported. 13 A frequency-doubled,¯ber-coupled Nd: YVO 4 diode-pumped solid-state laser (CNI, MSL-FN-671), operating at ¼ 671 nm was used to probe the liquid. The spectrometer comprised an interference¯lter, a VIPA etalon (LightMachinery) with a free spectral range of 30 GHz and a EMCCD camera (Andor iXon DU888). The laser output was collimated into an 8 mm diameter beam to¯ll the back focal plane of a 20Â microscope objective (Olympus, NA ¼ 0.5). The quarter-wave plate inserted just before the objective was used to transform the laser linear polarization into a circular polarization. Scattered light was collected by the same objective, separated from the incident light by the polarization beam-splitter and coupled into a single mode¯ber (SMF), which also acted as a confocal pinhole. The¯nal imaging resolution of this setup was calculated to be 1 Â 1 Â 5 m. Figure 2(b) is a single-¯ber probe arrangement. The light from the laser was passed through a 50/50 ber coupler (C), one output port of which was connected to a¯ber ferrule and a graded-index lens (Thorlabs GRIN2906). The ferrule and the¯ber lens both had diameter of 1.8 mm, were aligned along the optical axis, brought in contact and glued together. The length of GRIN lens was chosen to be 0.29 of its pitch, resulting in the laser beam being focused at a distance of 1.35 mm after the lens. The diameter of the focal spot was measured to be approximately 3 m, corresponding to the imaging resolution of 3 Â 3 Â 10 m for this system. The scattered light was collected by the same GRIN lens, coupled back to the SMF through the samē ber ferrule, and redirected to the spectroscope by the¯ber coupler. Thus, the measurement system B presents the direct analog of the bulk-optical system A and, as A, has a confocal nature. Figure 2(c) is a dual-¯ber arrangement in which two optical¯bers brought in contact with each other side-by-side, aligned symmetrically with the center of the GRIN lens and placed at the focal distance from the lens in a v-grove. In this instance, no¯ber coupler was used: the light from the laser was sent directly to the sample through the GRIN lens, and the scattered signal was collected by the second¯ber and directed towards the spectrometer.

Measurement Results
Samples of water, ethanol and glycerol were measured using the two¯ber setups [Figs. 2(b) and 2(c)]. For reference purposes, we¯rst used the confocal microscope arrangement [ Fig. 2(a)] to record the spectrum for distilled water. The Brillouin frequency shift in water is well-documented as water is often used in Brillouin imaging experiments for calibration purposes. The central elastic (Rayleigh) peak and the Brillouin doublet of water are clearly visible in Fig. 3(a). The curve, shown with blue crosses, represents averaged results over 20 measurements, with each measurement taken using a 1 s acquisition time on the camera. The Brillouin peaks were¯tted with a Lorentzian lineshape (in agreement with the mechanical oscillator model), to obtain a measured Brillouin frequency shift of w ¼ 5:81 AE 0:01 GHz. This frequency corresponds to an acoustic velocity of 1465 AE 2:5 ms À1 (taking the refractive index of water as n w ¼ 1:332), the result expected for distilled water at 20 C.
Next, we tested the single-¯ber Brillouin probe arrangement [ Fig. 2(b)] with three types of liquids: water, ethanol and glycerol. The results of this test are presented in Fig. 3(b). Since all three measurements overlap, we use di®erent symbols for water (blue cross) and glycerol (yellow open circle), and a solid red line for the measurement of ethanol. In all three measurements, the Brillouin signal is shifted by 5:07 AE 0:01 GHz from the central Rayleigh peak. It is known that the Brillouin frequencies of the three liquids are, in fact, substantially di®erent. 21,22 The explanation of the measurement results becomes clear if we recall that the probe light, before being focused into the sample, passes through a 1 m-long piece of a single-mode optical¯ber (from the¯ber coupler). In the¯ber core, the light undergoes spontaneous Brillouin scattering in the backward direction, that is combined with the signal scattered by the sample. In fact, the amount of light scattered inside the¯ber is expected to be several orders of magnitude larger compared to the  Fig. (2) for water, ethanol and glycerol. The spectra all comprise a central elastic peak (R) and a Brillouin doublet, Stokes (S) and AS peaks, respectively. All three measurements in B are overlapping; hence it is likely that the Brillouin signal originates not from the liquid samples, but from the¯ber itself. Note that the spectra in C are separated vertically by 10 units for better visibility.
light scattered inside the liquid sample. The ratio of these two signals scales roughly with the ratio of the scattering volumes in the¯ber and liquid, respectively, so V f =V l % L f =L l ¼ 1 m/10 m ¼ 10 5 , since the mode¯eld diameter of the single-mode¯ber and the GRIN lens' focal spot diameter are approximately the same (d ¼ 3 m). The di®erence between the scattering e±ciency in liquids and solids is not taken into account in this calculation, although the scattering e±ciency of solids is typically higher. Thus, the factor 10 5 is likely to be a conservative estimate for the ratio of light scattering in a glass ber versus liquid samples. This indicates that the side-lobes in Fig. 3(b) correspond to Brillouin scattering from the¯ber rather than the liquid samples, which is expected to dominate the spectral measurements based on our estimates. Indeed, if we consider that the Brillouin signal to the right of the Rayleigh peak in Fig. 3(b) is, in fact, the Stokes signal (this is possible if the frequency shift is larger than half of the free spectral range of the VIPA), and that to the left of the Rayleigh peak is the AS signal, then we arrive at glass ¼ 24:93 AE 0:01 GHz, in close agreement with previously reported value for Brillouin frequency shift in fused silica glass, corrected for the wavelength of the probing laser. 23 The results of using a single-¯ber Brillouin probe demonstrate that the scattering by the¯ber itself presents a signi¯cant challenge in Brillouin endoscopy. This situation closely resembles the problem encountered in the¯eld of Raman endoscopy. 24,25 Over the last 20 years of technological and scienti¯c development, the¯eld of Raman endoscopy has reached relative maturity with clinical applications being possible today. 18 The process of designing a sensitive and e±cient Raman¯ber probe, however, has to overcome the same hurdle, i.e., the signal coming from the sample has to be isolated from thē ber background. The main di®erence between Brillouin and Raman light scattering is their spectral properties, namely the Brillouin signals being narrow-band (tens of MHz) and shifted by only 1-30 GHz from the elastic peak (a narrow band that renders traditional¯lters, e.g., dielectric¯lters useless), which imposes additional, non-trivial challenges in overcoming this hurdle. On the other hand, unlike Raman scattering, spontaneous Brillouin scattering in optical¯bers only occurs in backward direction, since forward scattering does not satisfy the phase-matching condition between optical and acoustic waves. This fact becomes very important in construction of the Brillouin endoscope in a backscattering geometry using multiplē bers. Choosing separate¯ber channels for the probe and return beam prevents back-scattering of the probe in the¯ber from reaching the spectrometer.
The results of using this dual-¯ber approach [ Fig. 2(c)] are presented in Fig. 3(c). In contrast with the spectra obtained with the single-¯ber arrangement, it can be seen that the side-peaks are di®erent for each sample. After¯tting the measurement data with Lorentzian curves (Fig. 4), we arrive at the following Brillouin frequency shifts

Numerical Modeling of Light Propagation in Fiber Endoscopes
Although the single-¯ber arrangement of Fig. 2(b) is confocal in nature, this is no-longer the case for the dual-¯ber arrangement of Fig. 2(c), since the two¯ber cores are separated laterally by 250 m, the diameter of the¯ber cladding. The apertures of the illumination and collection¯bers are located 125 m away from the center of the GRIN lens, which can lead to optical aberrations and poor overlap between illumination and collection volumes. To understand the e®ect of this¯ber core separation on the collection e±ciency of the scattered light, we approximate the light exiting from the optical¯ber by a Gaussian beam with the full width at half maximum of 3 m (equal to the modē eld diameter of the SMF) and use the beam propagation method to study numerically the light propagation in our¯ber probes. 26 Since the¯ber and the lens have cylindrical symmetry, we restrict our model to the twodimensional plane (X; Z) with z-direction being along the optical axis of the lens. The GRIN lens (Thorlabs) is modeled as a variable refractive index pro¯le Here, l 0 ¼ 1:35 mm is the focal distance of the lens, L ¼ 5:38 mm is the lens length, n 0 ¼ 1:607 is the maximum refractive index of the lens and A ¼ 0:1149 is the coe±cient of the refractive index curvature.
Results of the beam propagation modeling are shown in Fig. 5. Figure 5(a) shows a single Gaussian beam, placed at the focal plane of the GRIN lens, and traveling through the lens. As expected, the beam is focused behind the lens at the focal distance and forms a good image of the initial beam. This situation corresponds to the single-¯ber probe setup. Figure 5(b) shows two Gaussian beams, placed in the focal plane in front of the lens and separated vertically by 250 m. This corresponds to the dualber arrangement. In the experiment, only one¯ber delivers light, whereas the second¯ber collects the scattered signal. Since the system is time-reversible, it does not matter if the source is placed from the left or the right side of the lens. Thus, by putting both sources at the apertures of optical¯bers, we can trace the beams trajectories and analyze their overlap at the image plane. The collection e±ciency is dependent on a good overlap between the illumination and collection beams at the scattering sample. As is clear from Fig. 5(b), the illumination and collection volumes do not overlap signi¯cantly at the sample plane, meaning that negligible amount of light scattered by the sample can be coupled back into the collection¯ber. Although we have measured Brillouin signals in transparent and homogeneous liquids (water, ethanol and glycerol) using a dual-¯ber probe, the poor overlap between illumination and collection volumes, characteristic for this endoscope design, makes its use in conditions of highly scattering biological tissues problematic.
Next, we explored di®erent separation distances between the¯bers and the¯ber tilt angles in an attempt to seek an optimal scenario, where the images of the two Gaussian beams overlap. Figure 5(c) illustrates one such con¯guration, in which the¯ber cores are separated by 50 m and each core is tilted by 6 away from the optical axis. One can see that although the images of the beams show signi¯cant aberrations they do in fact overlap, which suggests higher collection e±ciency for this probe design compared to the previous one. The tilt angle and the core separation conditions can be realized through polishing of separate single-mode optical¯bers, or possibly through fabrication of a dual-core¯ber, which facet is polished to a conical shape with the required angle. An alternative solution could be a complex¯ber lens, fabricated by the direct laser writing method on the tip of a multicore optical¯ber that would enable light focusing in a single aberration-free spot. 27

Discussion and Conclusions
Based on our experiments using a single-¯ber and a dual-¯ber Brillouin probes, we have concluded that the former design is not suitable due to signi¯cant backscattering inside the¯ber, whereas the latter is suitable for Brillouin endoscopy after improvements in the collection e±ciency. Our simulation results suggest that a simple arrangement of a two optical bers and a GRIN lens does not have the desired overlap between the illumination and the collection volumes at the image plane of the lens. Improvements on the dual-probe design, namely reduction of the core separation and the tilt, may help to increase the collection e±ciency by allowing for a better optical overlap between the illumination and collections volumes.
It is worth noting that, despite the poor collection e±ciency, we were able to obtain clear Brillouin signals from the three liquids using the dual-¯ber probe. In fact, the signal-to-noise ratio (SNR) in the measured spectra was similar to that using a conventional arrangement with a confocal microscope [setup illustrated in Fig. 2(a) and the measurement result shown in Fig. 3(a)]. This might seem surprising at¯rst because of the much lower collection e±ciency, however it should be noted that other factors a®ect the SNR, such as stray re°ections from optical elements and Rayleigh scattering. Although, we have utilized polarization rotation and¯ltering in the setup 2(A) in order to reduce amount of stray light being sent to the spectrometer, the¯nite extinction ratio of the polarization beam splitter and imperfections of the alignment could explain a small fraction of stray light being added to the Brillouin signal. In the dual-¯ber geometry, however, the chance of stray light reaching the detector is negligible since the illumination and the collection paths are completely separated. Also, the lower collection e±ciency also reduces the amount of Rayleigh light being collected and sent to the spectrometer.
In summary, we have demonstrated Brillouin microspectroscopy measurements of liquids (water, ethanol and glycerol) using¯ber-integrated BI system. We have designed a single and a dual-¯ber probes using o®-the-shelf components and demonstrated Brillouin spectroscopy measurements using the dual-¯ber probe. The main challenge of the Brillouin endoscopy using optical¯bers has been identi¯ed, namely the detrimental e®ect of the inelastic scattering in¯bers. We believe that this work represents an important¯rst step towards simplied,¯ber-integrated BI setups, that in the future could¯nd applications in biomedical research and clinical trials. The next step in the development of¯ber probes for Brillouin endoscopy could be the optimization of the dual-¯ber design in order to provide better coupling e±ciency of the scattered light.