1. Introduction

Non-linear Raman imaging techniques, such as Coherent anti-Stokes Raman scattering (CARS), are important tools for performing optical biopsies for diagnosing tumours [1,2], as well as for label-free imaging of lipid-rich tissue structures such as myelinated axons [3,4] and lipid droplets [5]. For imaging deep inside tissue, endoscopic probes are needed since the tissue is highly turbid [6,7]. The diameter of standard lens-based endoscopes is, however, too large for use inside sensitive tissue, and multimode fibres (MMF) have emerged as a possible alternative. Such fibres can have a high numerical aperture (NA) [8], giving comparable image quality to other endoscopes and, because of their small diameter ($\approx 100\,\mathrm {\mu }\textrm {m}$), they allow imaging deep in tissue with minimal damage. Development of wavefront shaping techniques [9,10] has allowed focusing and scanning of coherent light through a MMF [11], and following that breakthrough, methods relevant for bio-imaging, such as fluorescence imaging, were demonstrated through a single MMF [12,13], and applied in-vivo [14,15]. Multiple other imaging methods have been successfully implemented through a single MMF including wide-field [16], confocal [17], and two-photon excitation fluorescence (TPEF) imaging [18–20], as well as label-free imaging methods such as linear Raman [21,22] and second-harmonic generation (SHG) imaging [23]. In this article, we demonstrate the use of a high-bandwidth, composite MMF probe which has a reduced non-linear background from the fibre, and therefore allows CARS imaging of biologically relevant tissue using a femtosecond laser source.

When fibres are used in endoscopes for CARS imaging, an unwanted background signal can be generated inside the fibre itself due to four-wave mixing (FWM) and other non-linear processes [6,7,24,25], in particular when femtosecond excitation sources are used. This non-linear background results in strongly reduced contrast in the images. Since the background overlaps spectrally with the CARS signal generated in the sample, it cannot be filtered out when the same single fibre is used for both the excitation and the detection [26], making it challenging to construct a thin endoscope.

Notably, successful attempts have been made to suppress the FWM background in CARS endoscopes, using an air core photonic crystal fibre [6], and a double core-double clad fibre [27] for laser pulse delivery. The resulting endoscopes, however, have diameters of $4.2\,\mathrm {mm}$ and $2.4\,\mathrm {mm}$, respectively. In Fig. 1(E) we illustrate the problem with using such wide probes. The figure shows a to-scale image of a mouse brain, the probe from [27] and the probe discussed in this manuscript. Clearly, the probe we describe here is more suitable for imaging inside the brain of an animal model. In fact, the type of endoscopes described in [6,27], cannot be made particularly thin. The reason is that the delivery fibre is necessarily single mode. Single mode fibres cannot be turned into an imaging device with the wave-front shaping techniques used in this manuscript and for other MMF endoscopes. Thus, some mechanical scanning mechanism is needed. In both [6] and [27] this is accomplished by moving the fibre along a spiral trajectory using piezo electrodes. This limits how much the endoscope can be reduced in size, and the fraction of the endoscope that will correspond to a field of view. While such endoscopes are highly suitable for CARS imaging for e.g. gastrointestinal endoscopy, their diameters are prohibitively large for imaging inside sensitive tissue. Although our MMF endoscopic probe has a somewhat lower NA compared to [6,27], this could be overcome with future development of higher NA graded-index (GRIN) fibres, and it, nevertheless, offers meaningful levels of resolution for imaging biological structures.

 

Fig. 1. Experimental setup used for fibre characterization and imaging. A. Setup (see the Supplement for detailed drawing). IPC – input polarization control. The inset shows the distal end of the spliced probe. B. Illustration of the principle of spectral focusing CARS, where the time delay between the pump and Stokes tunes the addressed Raman shift. $\nu _\mathrm {inst}$ – instantaneous frequency. C., D. Calculated optical electric fields inside a pure GRIN fibre probe (C) and a composite probe with a $2\,\textrm {mm}$ long splice (D). E. Size comparison of the proposed probe, an endomicroscopic imaging probe based on a double-core double-clad fibre [27] (probe in black, FOV in yellow), and a mouse brain. The brain image is adapted from the Allen mouse brain atlas [33,34]. The scalebar is $1\,\mathrm {mm}$.

Download Full Size | PDF

An ideal MMF probe for performing CARS endoscopy using a femtosecond laser would have two key characteristics. First, it should have a bandwidth high enough to focus all the wavelengths in the pulse to the same point in time and space. Second, the FWM background generated in the fibre should be as low as possible. The first requirement rules out the use of step-index fibres which have a low bandwidth and, in fact, focus different wavelengths to different distances from the fibre facet. This results in a notably elongated focus [20] and thereby a reduction of the sectioning capability normally associated with non-linear imaging, as well as a lower excitation intensity and thus signal. In contrast, GRIN fibres have sufficient bandwidth to focus a femtosecond laser pulse. However, they also exhibit high levels of FWM background; because of the self-imaging property of these fibres, the focused point that is used to excite the signal at the sample plane is re-imaged multiple times along the probe (similarly to a GRIN lens). This creates high-intensity points with overlap in time and space of the pump and Stokes beams inside the fibre, generating a strong FWM signal.

The background generated in GRIN fibres could be significantly reduced by modifying the distal end, i.e., the end going into the sample, in order to prevent that the self-imaged fields form tight foci inside the fibre. In essence, if the field at the output end of the GRIN fibre is a random speckle pattern, the foci that are still formed through the self-imaging effect will be broken up into low-intensity spots. To achieve this, we propose using a composite probe that consists of a relatively long GRIN fibre, so that the bandwidth of the probe is still sufficient for focusing a broadband pulse, and a short SI fibre spliced on at the distal end (see inset in Fig. 1) in order to create a random speckle pattern at the interface to the GRIN fibre.

We demonstrate that for this composite probe the background generated in the GRIN fibre is reduced by more than an order of magnitude compared to a pure GRIN fibre. This was crucial for implementing CARS in tissue using a fs laser. Although the authors have recently demonstrated CARS imaging through a MMF [28] using a ps laser, this was only demonstrated on high contrast polymer beads on a surface. A fs laser is also advantageous for combining CARS with other non-linear imaging methods, such as SHG or TPEF imaging [3,29]. The lateral and axial size of the point spread function, the power ratio, i.e., the fraction of the power in the sample plane contained in the focused point, and the range over which the wavelength can be tuned after the initial calibration are retained. Furthermore, since the splice only has a small effect on the dispersion of the probe, the spectral resolution and chemical contrast of spectral focusing (SF) CARS [30–32] are not affected.

This composite probe, which we characterize in this paper, allows CARS imaging of tissue structures with epi-detection, as appropriate for an endoscope, whereas with the bare GRIN fibre, no tissue structures could be imaged because of the low contrast due to the high FWM background. We demonstrate this by imaging myelinated axons in a sciatic nerve and in the white matter of the brain. Moreover, we show simultaneous CARS and TPEF imaging of the cerebellum of a fixed brain expressing GFP in the mossy fibres. This is the first demonstration of CARS imaging through a single multimode fibre on relevant tissue structures.

2. Methods

2.1 Experimental setup

The schematic of the setup is shown in Fig. 1(A) and (a) detailed schematic can be found in Fig. S1. We used a femtosecond laser and OPO (Coherent Chameleon Discovery) producing two synchronized pulse trains at different wavelengths ($1040\,\mathrm {nm}$, used as a Stokes beam, and a tunable beam, used around $800\,\mathrm {nm}$ as a pump beam) as an excitation source. Each beam from the laser was stretched using $220\,\mathrm {mm}$ of SF57 glass, in order to implement SF-CARS [31,32], which increases the spectral resolution when using a broadband laser. The glass blocks, together with the optics in the setup and the GDD precompensation integrated into the laser added $4.7\cdot 10^4\,\mathrm {fs^2}$ to both beams. The beams were then overlapped with an adjustable time delay, and expanded to fill the active area of a spatial light modulator (SLM) (Meadowlark HSP1920-1064-HSP8), which was used in an off-axis configuration. The beam modulated by the SLM was Fourier transformed using an achromatic doublet, allowing separation of the diffraction orders. The Fourier plane of the first-order diffraction was then demagnified onto the proximal facet of the fibre using a pair of achromatic doublets.

Unlike SI fibres, the GRIN fibres used for this experiment do not maintain circular polarization, and to overcome this, we simultaneously controlled the spatial phase and polarization distribution at the fibre input facet as described in [23,28].

Since the holograms displayed on the SLM have the form of sums of diffraction gratings, the phase modulation is wavelength-dependent, and this effect was compensated for using a prism placed in a plane conjugate to the SLM plane [35].

To characterize the light transport through the fibre, which is a necessary step to convert the fibre into a point-scanning imaging device, the imaging plane (located $30\,\mathrm {\mu }\textrm {m}$ from the distal facet of the fibre) was imaged onto a CMOS camera using an objective lens (Olympus MPlanFL N 20$\times$/0.45NA) and an achromatic doublet lens. The image was overlapped with a plane wave reference beam, which was also generated on the SLM. The whole calibration module could be removed during the imaging phase. The calibration method is described in Section 2.3.

During imaging, the CARS and TPEF signals collected through the fibre were first reflected off a dichroic mirror (DM) (Thorlabs M254C45), passed through a second DM (Thorlabs DMLP505) and then split by a third DM (Thorlabs DMLP605) into CARS and TPEF channels. Both signals were then filtered using a short-pass filter (Thorlabs FESH0700) and a band-pass filter (Thorlabs FBH650-40 for CARS, Semrock BrightLine 525/39 for TPEF) and focused onto a PMT (Hamamatsu H10723-20). For transmission imaging, after removing the mirror in the calibration module, the signal was filtered (Thorlabs FESH0700 and FBH650-40) and focused onto a PMT (Hamamatsu H10723-20). The signals from the PMTs were $10\,\mathrm {kHz}$ low-pass filtered and sampled by a data acquisition card (NI USB-6343).

2.2 Probe preparation

The fibre was first stripped from its acrylate coating and cleaved to a flat facet at both ends. For the spliced probes, a SI fibre was tapered to have the same core diameter as the respective GRIN fibre. The taper was subsequently spliced onto one end of the GRIN fibre (which becomes the distal end of the probe) and cleaved to the desired length (see the inset in Fig. 1).

The length (including the splice) of all probes was $(30\pm 2)\,\mathrm {mm}$. In order to hold the fibre in the setup, the probes were glued into ceramic ferrules (Thorlabs CF128) with UV curable glue (Norland Products NOA 65). The ferrule is $10.5\,\mathrm {mm}$ long and the proximal end of the fibre is flush with the face of the ferrule. The remaining ca $20\,\mathrm {mm}$ can be thus used for insertion into the tissue. This is sufficient for imaging, for example, throughout the entire depth of a mouse brain. Longer GRIN fibres could be used, although, ideally, not longer than that they still support the bandwidth of the femtosecond laser [35].

The tapering and splicing were performed on Large Diameter Splicing System 2.5 (3SAE Technologies).

For the demonstration in Fig. 2(A), we used three different silica glass GRIN fibres: Prysmian DrakaElite ($50/125\,\mathrm {\mu }\textrm {m}$ core/cladding diameter, NA of $0.29$), YOFC GI2017-C ($80/125\,\mathrm {\mu }\textrm {m}$, $0.30$ NA) and a custom drawn GRIN fibre ($62.5/125\,\mathrm {\mu }\textrm {m}$, $0.30$ NA, drawn at CREOL, the College of Optics and Photonics, University of Central Florida). For the splices, we used a commercially available SI silica multimode fibre CeramOptec Optran Ultra WFGE ($100/110\,\mathrm {\mu }\textrm {m}$, $0.37$ NA).

 

Fig. 2. CARS imaging of beads through MMFs with and without step-index splices. A. CARS images of $2\,\mathrm {\mu }\textrm {m}$ polystyrene beads on glass slide taken at $3059\,\mathrm {cm^{-1}}$ in transmission and images taken with the sample moved out of the focus. B. Difference between the images in A with and without sample for the Prysmian fibre. C. Images of PS beads with epi-detection for the custom GRIN fibre. D. Bead to background intensity ratio as a function of the splice length for the custom GRIN fibre.

Download Full Size | PDF

2.3 Fibre calibration procedure

In order to obtain a focus at the distal end of the multimode fibre, the light transport through the fibre must first be characterized in a calibration procedure. This was performed using the calibration module described above. The procedure consists of the three following steps: finding the extent of the fibre core, finding the correct time overlap with the reference and finally finding the transmission matrix.

In the first step of the calibration procedure, a point was raster-scanned across the input of the fibre using the SLM and the transmitted power was measured by integrating the image on the camera. Only the input points that gave at least $20\,\%$ of the maximum intensity (their position corresponded to the core) were selected for the subsequent steps.

In the second step, the input of the fibre was filled with a random speckle pattern in order to excite a large number of fibre modes. The image on the camera was overlapped with the reference beam, while its phase was shifted using the SLM. The length of the reference delay line was then adjusted to obtain maximal average oscillation amplitude over the field of view, thereby achieving optimal overlap of the signal and reference pulses. This procedure resulted in maximum intensity in the focused points after the calibration. Exciting a large number of modes was found to be particularly important for the probes with longer splices. For the spliced fibres and particularly for pure SI fibres, we also observed that the reference delay line must be set to within ca $\pm 30\,\textrm {fs}$ of the optimal position, to maximize in the quality of the focus. For GRIN fibres, however, which have a much larger bandwidth than the spectral width of the laser [35], this alignment is not critical, since slightly changing the delay of the reference beam, within the length of the laser pulse, or exciting a smaller subset of the modes produced the same quality of the focused spot.

In the final step, we measured the transmission matrix, i.e., the relation between the optical field at the input fibre facet and the field in the sample plane. The selected input points were projected one by one, and the amplitude and phase of the output were measured using phase-shifting interferometry. Once the calibration procedure was finished, access to the distal end of the fibre was no longer necessary.

To create a single point at the output, all the selected input points were projected simultaneously, with their phases are set to the values of the measured phases, but with the opposite sign, and relative amplitudes set to the measured values of the amplitude. This ensured that all the light interfere constructively in the output point, while the relative amplitude information suppressed those input points which would not contribute to the focus and only increase the background.

For CARS imaging, two wavelengths separated by ca $3000\,\mathrm {cm^{-1}}$ must be focused simultaneously in the same point at the sample plane. Since the bandwidth of the system was not large enough to cover such a large wavelength range, we performed the calibration separately for the Stokes and pump beams. To create the joint focal points for imaging, we simply summed the two transmission matrices [28]. The temporal overlap of the pulses was set by the delay line in the Stokes beam.

The whole calibration procedure took about a minute per wavelength and input polarization, and thus the whole calibration necessary for CARS imaging took less than five minutes.

2.4 Sample preparation

The sciatic nerve and brain used for the imaging experiments were harvested from paraformaldehyde (PFA) perfusion fixed adult mice (wild type and STOCK Tg(Thy1-EGFP)MJrs/J). The sample acquisition procedures followed the Czech guidelines for animal experimentation and were approved by the Branch Commission for Animal Welfare of the Ministry of Agriculture of the Czech Republic (permission No 47/2020). The samples were further post-fixed in $4\,\%$ PFA in phosphate-buffered saline (PBS) for 24–48 hours and then stored in PBS until imaging.

3. Results

3.1 Suppression of signal generated in the fibre

Figure 2(A) shows the effect on the non-linear background of adding a short SI splice to a GRIN fibre. Here, we imaged $2\,\mathrm {\mu }\textrm {m}$ polystyrene (PS) beads on a glass slide in transmission using different probes. To demonstrate what fraction of the collected signal comes from the fibre, we moved the sample $100\,\mathrm {\mu }\textrm {m}$ away from the focal plane and acquired another image. Since $100\,\mathrm {\mu }\textrm {m}$ was much further than the length of the point spread function ($\approx 25\,\mathrm {\mu }\textrm {m}$), no signal was generated in the sample, and all the measured signal was an undesirable background generated in the fibre.

The SI splices clearly reduced the signal generated in the fibre by at least an order of magnitude. The residual background in the images of the PS beads captured using the spliced probes made from the custom GRIN fibre and the YOFC fibre is the non-resonant signal generated in the glass slide. The reason for the reduction of the background is two-fold: The speckle pattern at the GRIN fibre interface, generated by the step-index fibre splice, results in broken up, low-intensity foci in the GRIN fibre (Fig. 1(D)). The foci are still present since the self-imaging property is still present. Furthermore, because of the strong wavelength dependence in the step-index splice, the speckle patterns are different for the pump and the Stokes beam, and thus the foci for the two wavelengths in the GRIN fibre do not have a strong spatial overlap, resulting in an inefficient FWM generation. The maximum value of the calculated distribution for $I_\textrm {pump}^2 I_\textrm {Stokes}$ (Figs. S2, S3), which is proportional to the FWM signal, was 100 times smaller for the fibre probe with a $2\,\mathrm {mm}$ splice, similar to the measured degree of background suppression. Further details on the model and the distribution of the individual fields in the fibre are presented in the Supplement (Figs. S2–S4).

Depending on the type of fibre, however, an order of magnitude suppression of the background might not be sufficient. While the spliced probes based on the custom GRIN fibre or the YOFC fibre allowed imaging with high contrast, the signal generated in the Prysmian fibre was so strong compared to the CARS signal from the PS beads, that even with the splice, the contrast of the images was poor. Consequently, care must be taken when choosing a suitable fibre. For the data presented in Figs. 4–5, we used the custom drawn GRIN fibre.

Figure 2(B) shows that the image without the background signal could not be obtained simply by subtracting the image taken without the sample. In addition, during tissue imaging with epi-detection, where the detected signal is back-scattered from the tissue, the level of the background signal would be significantly affected by the scattering properties of the tissue and could therefore change depending on the position in the sample.

As seen in Fig. 2(C) the reduction of background was actually larger when using epi-detection for a sample in air. This, since a substantial fraction of the light generated in the fibre is internally reflected off the output facet back to the detector due to the high refractive index difference at the glass/air interface, significantly increasing the background signal and reducing the contrast in the images. During tissue imaging, however, the output of the fibre would be in contact with the tissue or a liquid with a refractive index close to $1.33$, significantly weakening the back-reflection. Hence, the transmission images of the beads are more representative of the degree of background reduction when imaging tissue using epi-detection.

Figure 2(D) shows a comparison of the contrast between the beads and the background for different lengths of the splice (the total length of all probes was the same). We calculated the ratio of a signal measured across a bead and a signal measured in the same pixels without the sample. The error bars show one standard deviation of this quantity across all the beads in the field of view. The contrast was largest for a splice length of about $1\,\mathrm {mm}$, indicating that this was an optimal splice length. The graph also shows two data points taken with the SF57 glass blocks removed from the setup and therefore without spectral focusing. Although the signal level from the beads as well as from the background was higher by a factor of 10 for the short unchirped pulses, the contrast of the image was lower due to the inefficient excitation of the PS Raman resonance with a broadband pulse. The background reduction due to the splice was, however, similar to the reduction observed for the chirped pulses.

To characterize the wavelength dependence of the signal generated in the probe, we captured CARS images in transmission at different Raman shifts using probes with different splice lengths without a sample present (Fig. 3(A)). The range of the Raman shifts was selected to cover typical values used for bio-imaging (of lipids, proteins, and DNA) and the fibre was calibrated for each Raman shift separately. The signal depended on both the Raman shift and the length of the splice. For all the splice lengths for this fibre the background was lower for higher Raman shifts.

 

Fig. 3. Dependence of the signal generated in the fibre on Raman shift and the length of the splice. A. CARS images taken in transmission without a sample at different Raman shifts using probes based on the custom GRIN fibre with and without the splice. B. Intensity of the signal generated in the fibre as a function of the splice length averaged over all measured Raman shifts. The graph shows both the measured data and data corrected for the quality of the focused points. The error bars show the minimal and maximal values across the range of Raman shifts used for the experiment. C. The intensity of the signal generated in the fibre as a function of the Raman shift for 3 different GRIN fibres.

Download Full Size | PDF

 

Fig. 4. Evaluation of the quality of the focused points. A. Point spread functions for three different probes measured at $789\,\mathrm {nm}$. B. Spot quality (power ratio, spot size and spot length) for two wavelengths as a function of the length of the splice. The value $0\,\mathrm {mm}$ corresponds to a GRIN fibre without a splice, the value $30\,\mathrm {mm}$ to a SI fibre. The dashed lines denote the diffraction-limited size given by the numerical aperture of the fibre. The error bars show the range of values across a $50\,\mathrm {\mu }\textrm {m}$ field of view with the highest power ratio and the smallest spot size being on the axis and the lowest power ratio and the highest spot size at the edge of the field of view. This spot size and power ratio distribution is characteristic of GRIN fibres [35] and an effect of their refractive index profile. C. An example of the distribution of the power ratio across the field of view, measured at $789\,\mathrm {nm}$ for two different probes. D. Wavelength tuning range (FWHM of spot intensity as a function of wavelength) as a function of the length of the splice. The error bars show one standard deviation across a $50\,\mathrm {\mu }\textrm {m}$ field of view.

Download Full Size | PDF

 

Fig. 5. Demonstration of the chemical contrast achieved by spectral focusing. A. CARS images (signal detected in transmission) of a mix of PS and PMMA beads on a glass slide taken with two different time delays between the pump and Stokes beams. B. Overlay of the images. C. Intensity of the beads as a function of the time delay between Stokes and pump beams.

Download Full Size | PDF

In Fig. 3(B) the background intensity, averaged over all Raman shifts, is plotted. The focus quality at the sample plane (Fig. 4), and thus also the focus quality in the fibre changed with the length of the splice. In Fig. 3(B) the data is thus also plotted corrected for the change in power ratio and spot size, assuming that the background depended quadratically on the pump intensity and linearly on the Stokes intensity (which we also confirmed). The corrected data exhibit a minimum for a splice length of $2\,\mathrm {mm}$, indicating that this is an optimal splice length. This is close to the optimum obtained from Fig. 2(D). The difference could be due to the effect of the splice on the pulse shape (see the Supplement and Figs. S5 and S6 for a discussion of this). Moreover, the values in Fig. 2(D) depend on the distribution of the beads across the field of view and were measured for a single Raman shift only. Nevertheless, in terms of the signal to background ratio, the optimal length of the splice for the chosen GRIN fibre is somewhere between 1 and $2\,\mathrm {mm}$.

The strong dependence of the background signal on the pump wavelength, with a factor of 10 decrease for $10\,\mathrm {nm}$ decrease in the pump wavelength (Fig. 3(A)), and the large difference in intensity of the background for different fibres (Fig. 2(A)) indicate that the background was not solely from non-resonant FWM, but from resonant processes, which might depend on the details of the fibre doping and thus result in different background levels for different fibres. We also note that the pump wavelength dependence was different for different fibres, as shown in Fig. 3(C). Furthermore, the details in the refractive index profile of the fibres also affect the pulse propagation, the polarization scrambling and quality of the spatial overlap of the two wavelengths, which affects the absolute level of the background.

3.2 Focusing performance

To evaluate the effect of the splices on the focusing performance of the fibre probes, we calibrated the system at two wavelengths ($789\,\mathrm {nm}$ and $1040\,\mathrm {nm}$, corresponding to a Raman shift of $3059\,\mathrm {cm^{-1}}$, optimal for exciting PS). For each position of the focused point in the sample plane, we captured an HDR image of the point spread function in the focal plane. Moreover, by focusing using the objective lens behind the fibre, we measured the 3-dimensional point spread function (Fig. 4(A)).

The HDR images were fitted by an Airy disk, allowing the calculation of the spot size and the power ratio, see Fig. 4(B). The power ratio showed only a small change with increasing splice length. The size of the focused spots was close to the diffraction limit given by the numerical aperture. For the spliced probes which consist of two fibres with different numerical apertures, the numerical aperture of the composite probe is given by the lower of the two, here the GRIN fibre. In conclusion, the splice did not have a detrimental effect on the focus quality.

For comparison, Fig. 4(A),B also shows the quality of the foci achieved with just the SI fibre. As expected, since the bandwidth of such a probe is much smaller than the bandwidth of the laser, the achieved power ratio was very low. Moreover, since different wavelengths are focused at different distances from the facet, the resulting spot is much longer than expected from the numerical aperture. This would not only significantly limit the sectioning capability, but, more importantly, would result in different Raman shifts being excited at different distances from the facet. This, since the magnitude of the focus shift is wavelength dependent, resulting in a different spatial chirp for the pump and the Stokes beam. Hence, a pure step-index fibre is not suitable for CARS imaging with a fs laser.

Figure 4(C) shows the distribution of the power ratio across the focal plane. This shows that the size of the field of view is not affected by adding the splice.

3.3 Probe bandwidth

For CARS imaging, it is useful to be able to tune the wavelength after the system has been calibrated to image at different Raman shifts without re-calibrating. In order to evaluate the effect of the length of the splice on the bandwidth of the fibre probe (and hence the tuning range), we measured the quality of the focus (in terms of power ratio and spot size) across the field of view after the wavelength was tuned from the value used for the calibration. When the wavelength is tuned away from the calibration wavelength of $789\,\mathrm {nm}$, the power ratio drops and the size of the focus increases [35]. Figure 4(D) shows the tuning range for different splice lengths. It is clear that adding a few millimetres long SI splice to the high-bandwidth GRIN fibre has only a small effect on the wavelength tuning range. Hence, such probes can be used for imaging when excitation wavelength tuning is required.

3.4 Spectral focusing

Implementing spectral focusing allows targeting different Raman shifts and imaging with chemical contrast by tuning the relative time delay between pump and Stokes pulses. We demonstrate that this technique can be implemented through a MMF by imaging $2\,\mathrm {\mu }\textrm {m}$ PS and $2.5\,\mathrm {\mu }\textrm {m}$ polymethyl methacrylate (PMMA) beads on a cover glass (Fig. 5(A),B). Here, we used a fibre probe with a $2\,\mathrm {mm}$ long splice. The excitation wavelengths corresponded to a Raman shift of $3059\,\mathrm {cm^{-1}}$ (optimal for exciting PS). By changing the time delay by $1\,\mathrm {ps}$, we excited PMMA at $2947\,\mathrm {cm^{-1}}$ instead. The acquired spectra are plotted in Fig. 5(C). Figure 5(C) confirms that, since we obtained the correct spectrum for PS and PMMA after converting the time delay to a Raman shift, the SI splice did not result in errors in the spectral focusing because of the wavelength dependence of SI splice discussed above. In Fig. S7 we, furthermore, show that the chemical contrast is correct also slightly above and below the focus, indicating that the correct Raman shift was excited throughout the focal volume. This is the result of that the splice is kept short, which limits the magnitude of the wavelength-dependent focal shift.

3.5 Tissue imaging

The bio-imaging capabilities of the spliced probes were tested on a fixed sciatic nerve and fixed brain tissue from a mouse. Figure 6(A) shows a CARS image of a transverse cut of a sciatic nerve, where the myelin sheaths surrounding the axons are visible as rings. Figure 6(B) shows an image taken with the composite fibre probe inserted $1.5\,\mathrm {\mu }\textrm {m}$ into a mouse brain and shows myelinated nerve fibres in corpus callosum. In Fig. 6(C), we show simultaneous CARS and TPEF imaging of myelinated axons in the cerebellum of the brain from a mouse model expressing GFP in a subset of the mossy fibres (Thy1-EGFP).

 

Fig. 6. Tissue imaging. A. CARS image of myelin sheaths in a transversal cut of a sciatic nerve taken on the surface. B. CARS image of myelin sheaths in the corpus callosum of mouse brain taken with the probe inserted $1.5\,\mathrm {mm}$ inside the tissue. C. Multimodal CARS (red) and TPEF (green) image of myelinated nerve fibres in the cerebellum.

Download Full Size | PDF

The excitation wavelengths were $802\,\mathrm {nm}$ and $1040\,\mathrm {nm}$ for the pump and Stokes beams, respectively, corresponding to a Raman shift of $2853\,\mathrm {cm^{-1}}$ (optimal for exciting lipids). The time overlap of the two beams was set to maximize the contrast of the CARS images. This delay line position also maximized the intensity of the TPEF signal, indicating that this was the maximum time overlap of the pulses and thus addressing the intended Raman shift. Note that the GFP was excited using the excitation from the combination of the 2 pulses, thus effectively at $905\,\mathrm {nm}$, which is quite optimal for GFP. For the images in Fig. 6(A),B, the power in the focused point was $70\,\mathrm {mW}$ and $25\,\mathrm {mW}$ for the pump and Stokes beams, respectively. For multimodal CARS and TPEF imaging (Fig. 6(C)), we used $50\,\mathrm {mW}$ in each beam in order to increase the GFP signal. The per-pixel integration time was in all cases $2\,\mathrm {ms}$. The signals were collected through the fibre. Since CARS is generated predominantly in the forward direction, epi-detection relies on back-scattering from the tissue. Despite the low collection efficiency due to the small cross-sectional area of the fibre, the signal to noise ratio in the images was sufficient for imaging biologically relevant structures. For images Fig. 6(A),C, we used the custom GRIN with a $2\,\mathrm {mm}$ splice, for image Fig. 6(B), we used the YOFC GRIN fibre with a $1\,\mathrm {mm}$ splice. When imaging with a GRIN fibre without a splice, no structures were visible in the images, because of the strong background generated in the fibre. Thus, adding the SI splice to the GRIN fibre reduced the background sufficiently to make label-free CARS imaging of biologically relevant structures through a MMF is possible, and Fig. 6(B) shows that the MMF endoscope can be used for CARS imaging deep inside tissue.

4. Summary

In summary, we have shown that by using a composite fibre probe, the strong FWM background generated in GRIN fibres can be suppressed sufficiently to allow label-free CARS imaging with a femtosecond laser as an excitation source, through a MMF endoscope. The background was reduced by splicing a short ($0.25$ to $5\,\mathrm {mm}$) piece of SI fibre onto the GRIN fibre, in order to prevent the self-imaging effect from resulting in tight foci, thereby reducing the peak intensity inside the fibre, and thus the FWM intensity. We also verified that the splice had this effect by modelling.

The reduction of the background signal was at least an order of magnitude, while the short splice had only a minor impact on the bandwidth of the probe and the quality of the foci at the sample plane. We imaged $2\,\mathrm {\mu }\textrm {m}$ PS beads and, based on the bead to background ratio, and the average background across the spectral interval relevant for bio-imaging (here, $2807$ to $3107\,\mathrm {cm}^{-1}$), we estimated the optimal length of the splice to be $1$ to $2\,\mathrm {mm}$ for the fibres used here. Furthermore, by imaging a sample with both PMMA and PS beads, we showed that spectral focusing CARS can be implemented through a MMF, achieving imaging with chemical contrast using a fs laser, and that the SI splice did not have a detrimental effect on the chemical contrast.

The optimized composite fibre probe was used to acquire CARS images of the myelin sheaths in a sciatic nerve from a mouse and inside the corpus callosum of mouse brain. Moreover, we showed that these probes allow multimodal imaging, by performing simultaneous CARS imaging of myelin in the white matter, and TPEF imaging of mossy fibres expressing GFP in the cerebellum of a fixed brain from a mouse.

To turn this endoscope into a viable tool for tumour diagnosis inside sensitive tissue in animal models, it would be necessary to pinpoint the appropriate combinations of non-linear imaging method for reliable diagnosis for the targeted tumor type, and, preferably, improve the scan speed (see [23,28] for a discussion on possible approaches for that). Nevertheless, we believe that this demonstration is a substantial and key step towards enabling acquisition of optical biopsies inside sensitive tissue.