1. Introduction

Coherent anti-Stokes Raman scattering (CARS) microscopy is an emerging method in biomedical study along with similar method of stimulated Raman scattering microscopy. Unlike conventional fluorescence microscopy, it can obtain rich information about chemical species in a sample through Raman spectra without invasive labeling process [1–3]. Because a light source mainly limits performance of a CARS microscope, it has been extensively developed since emergence of CARS microscopy [4–19]. Main requirements for CARS light source suitable for biomedical study are (i) spectral bandwidth covering fingerprint region (500-1800 cm⁻¹), (ii) several 10 cm⁻¹ spectral resolution, and (iii) signal levels that yield sufficient signal-to-noise ratio with exposure time less than 100 ms.

These requirements have already been satisfied by several recent implementations such as supercontinuum-based ultrabroadband multiplex CARS systems [4, 5]. On the other hand, practical aspects such as (iv) compactness, (v) high stability (or being alignment free), and (vi) low cost are essential for further dissemination of CARS microscopy, which has been mostly overlooked. Actually, most of the milestones of CARS microscopy have been achieved by elaborate systems that include large-scale, expensive lasers such as a modelock laser installed on an optical table and require careful alignment [19]. This fact prevents spread of CARS microscopy to potential users such as biologists, biochemists, and medical scientists. Moreover, endoscopic implementation necessitate these requirements [20, 21]. Recently, fiber-based light sources for CARS microscopy that aim at these requirements were reported [6–10]. However, at this moment, they do not fully satisfy the requirements of (i)-(iii). Single-pulse CARS systems may have potential to satisfy (iv)-(vi) because of their collinear configuration [11–14]. However, they include some bulky components such as a prism compressor and a pulse shaping system, and thus no compact implementation has been reported [19]. Very recently, we have developed a compact light source of CARS microscopy that satisfies all requirements of (i)-(vi) [19]. However, it still includes beam splitting and beam combining parts and thus leaves complexity of optical setup and requires careful adjustment of beam paths. In this paper, we report a new implementation of light source for CARS microscopy that satisfies all the above requirements of (i)-(vi), having a surprisingly simple configuration. The desired light is obtained from a short photonic crystal fiber pumped by a microchip laser output. The configuration is fully collinear, that is, it does not include beam splitting/combining part. We investigated suitable length of the fiber and incident laser power to the fiber, where supercontinuum has broad bandwidth and the incident light pulse does not seriously collapse. We found that the best length of the photonic crystal fiber was several tens of centimeters, which was much shorter than that of previously reported light sources of CARS having similar setups [4, 22].

2. Basic design

A schematic of a whole system of CARS microscopy with the proposed light source is shown in Fig. 1. A microchip laser (HorusLaser HLX-I-F040, ~1.1 ns pulse duration, 1064 nm center wavelength, ~495 mW average power, ~27 kHz repetition rate) output is incident on a polarization maintaining nonlinear photonic crystal fiber (PCF, NKT Photonics SC-5.0-1040-PM, length: 30-50 cm) after power attenuation. The zero dispersion wavelength of the PCF is 1040 nm and longer wavelengths corresponds to anomalous dispersion region. The polarization direction of the incident light is set to be the same as one of propagation axes of the fiber. Two longpass filters (cutoff wavelength: 1025 nm, OD4) are inserted in the beam path of the output of the fiber, which block shorter wavelength component. The residual light includes supercontinuum and the original laser wavelength components that are used as Stokes and pump light components for CARS, respectively. The light beam is focused on a sample by an objective (Nikon Plan Apo IR 60x NA1.27WI). Generated anti-Stokes light from the sample is collected by another objective (Nikon S Plan Fluor ELWD 40x NA0.60), passes through a shortpass filter (cutoff wavelength: 1025 nm, OD4) and a notch filter (center wavelength: 1064 nm, OD6) so that they reject pump and Stokes light components, and is incident on a spectrometer (Princeton Instruments Acton SP-2358). A CCD camera (Princeton Instruments PIXIS 400BR) attached to the spectrometer records CARS spectra. In the experiment shown below, we insert an isolator in front of the laser output to reject returned light from end facets of the PCF, and a half waveplate right after the isolator to adjust polarization angle of the incident light so that it coincides with one of the propagation axis orientations of the PCF. However, they can be omitted by applying angled endfaces and rotational adjustment of the PCF, respectively. Note that the part of the setup after the first objective is a standard forward CARS setup [4].

 

Fig. 1 A schematic of experimental setup. Attn., attenuator; PCF, photonic crystal fiber; LPF, longpass filter; Obj., objective; SPF, shortpass filter; NF, notch filter.

Download Full Size | PDF

In order to validate the feasibility of the proposed setup, we measured CARS signal levels for different fiber lengths and incident light power levels to the fiber. The sample was methanol that exhibited C-O stretch resonance at ~1033 cm⁻¹ [23]. Actually, nonresonant background was superimposed to the resonant signals in a CARS spectrum, thus we observed the signal levels at 1022 cm⁻¹ (nearly maximum of the spectrum). The results are summarized in Fig. 2(a), which indicate the followings;

 

Fig. 2 PCF input power dependence of CARS intensity of methanol at (a) 1022 cm⁻¹, and (b) 1444 cm⁻¹. The inset shows lengths of PCF.

Download Full Size | PDF

Similar data obtained by the signal levels at 1444 cm⁻¹ (corresponding to the CH₃ deformation resonance at 1448 cm⁻¹) are shown in Fig. 2(b), which show the same tendencies as above. By comparing the results of Fig. 2(a) and 2(b), we observed that signal levels at larger wavenumber saturated at higher incident light powers to the fiber. This indicates that broader CARS spectrum may be achieved by higher incident light power. Throughout the above experiment, spectral resolution were several 10 cm⁻¹.

Figure 3(a) shows temporal waveforms of the incident light component at the output of the fiber for several incident light power levels observed by a 3-GHz bandwidth photodetector. We inserted a bandpass filter (bandwidth: 10 nm) to the output beam of the fiber to extract the pump light component. Obviously the waveforms gradually collapsed and the power levels at ~2 ns decreased as the incident light power level increased. This may be caused by modulation instability in the PCF [22, 24]. This result indicates that the collapse of the waveform degrades temporal overlap of the pump and the Stokes light components and thus causes saturation or decrease of the efficiency of CARS process observed in Fig. 2. We also observed spectra of Stokes light component for several incident light power levels. The result shown in Fig. 3(b) indicates that spectrum of Stokes light generated by the modulation instability in the PCF broadened as incident light power increased. This broadening may be caused by multiple nonlinear effects such as soliton phase frequency shift and stimulated Raman scattering [22, 24]. We found that, for the photonic crystal fiber with length of around 30-50 cm and proper incident light power settings of the microchip laser, the Stokes light spectrum broadened so that it covered the entire fingerprint region (500-1800 cm⁻¹) and the pump waveform does not completely collapse so that the pump and the Stokes light components effectively excite CARS process in a target sample. The above results suggest that we should choose shorter fiber length and higher incident light power to achieve broader and more intense CARS signals. However, the maximum incident light power should be carefully chosen by considering the limitations such as the maximum laser output power, damage of the PCF, and sample damage. Therefore, in practice, the incident light power should be determined by these limitations and then the best length of the fiber should be determined according to the light power. In the experiments presented here, the fiber damage limited the incident light power at ~300 mW. Because the fiber damage is mainly caused by tightly focused beam on the surface of the PCF, the damage threshold may be improved by at least several times by creating collapsed areas in the end facet or splicing the PCF with an end cap that avoid tight focus at the surface [25].

 

Fig. 3 (a) Waveforms of pump light component at the output of PCF. (b) Power spectra of PCF output. Wavenumbers were measured from 1064 nm. Both of the data were taken by a PCF with the length of 33 cm.

Download Full Size | PDF

From the above results, it is likely that commercially available microchip lasers having similar specifications to the current one (several 100 mW power, ~1 ns pulse duration, and several 10 kHz repetition rate) can be applied to the present setup. Relatively longer pulse duration of a microchip laser compared with other standard lasers for CARS microscopy such as picosecond/femtosecond lasers was advantageous for keeping the temporal overlap of the pump and the Stokes light components, which enabled the present extremely simple configuration.

3. Experimental

Next we confirmed the utility of the proposed setup by using polystyrene bead samples. The experimental setup is the same as above. A sample was placed on a piezo stage (Mad City Labs Nano-LPS100) and the focused beam position was scanned by the stage. The sample was polystyrene beads dipped in water. An observed CARS spectrum from a polystyrene bead was normalized by that of water that corresponds to pure nonresonant background signal. After normalization, we applied the maximum entropy method to the spectrum to reconstruct a Raman spectrum [26]. When the method is applied, constant phase offset was added so that the base line becomes nearly zero. Fiber length, incident light power to the fiber, and exposure time were set as 33 cm, 280 mW, and ~10 ms, respectively. Observed CARS spectra of a 20 μm polystyrene bead and water are shown in Fig. 4(a). The CARS spectrum of water was magnified by 3.5 for comparison. The nonresonant CARS spectrum of water tailed up to ~1900 cm⁻¹ and thus resonant CARS signals may be observed within the spectral region. A normalized CARS spectrum generated by the spectra in Fig. 4(a) is shown in Fig. 4(b), which showed a typical shape of CARS spectrum. A reconstructed CARS spectrum from the normalized spectrum is shown in Fig. 4(c). It showed an excellent agreement with previous reports. A lot of resonances within the fingerprint region (500-1800 cm⁻¹) were observed such as C-C stretch (1002 cm⁻¹, 1449 cm⁻¹, 1584 cm⁻¹, 1603 cm⁻¹), C-H in-plane bending (1031 cm⁻¹, 1156 cm⁻¹, 1182 cm⁻¹), C-H out-of-plane bending (757 cm⁻¹), and C-C-C in-plane ring deformation (793 cm⁻¹) [27, 28]. The spectrum showed negative values around 900-990 cm⁻¹, which is an artifact. This can be straightforwardly corrected by applying a spline interpolation in place of the simple constant phase offset correction [26]. It was observed that noise levels gradually increased as wavenumber increased over 1500 cm⁻¹, which was caused by relatively lower efficiency of CARS generation at higher wavenumbers.

 

Fig. 4 (a) Observed CARS spectra of water and a polystyrene bead. Values were magnified by 3.5 for water. (b) Normalized CARS spectrum of polystyrene. (c) Reconstructed Raman spectrum of polystyrene.

Download Full Size | PDF

An observed CARS image of 2 μm polystyrene beads dipped in water is shown in Fig. 5. In this experiment, fiber length, incident light power to the fiber, and exposure time per a pixel were set as 50 cm, 250 mW, and 30 ms, respectively. The incident light power was attenuated to tens of milliwatts so that the light may not damage the sample. The field of view, pixel numbers, and pixel size were 15 × 15 μm², 100 × 100, and 150 × 150 nm², respectively. Again a CARS spectrum was normalized by that of water. Maximal CARS signal levels at 993 cm⁻¹, which corresponds to C-C stretch Raman resonance at 1002 cm⁻¹, was used for the mapping. The CARS image successfully visualized multiple condensed polystyrene beads.

 

Fig. 5 CARS image of 2 μm polystyrene beads. Scale bar: 5 μm.

Download Full Size | PDF

4. Discussion

The proposed light source has a great potential of an ultra-compact implementation. In fact, microchip laser has a compact scale (~10 cm or less for each dimension) compared with other light sources for CARS and fiber-attached module of the laser is commercially available, which enables a palm-top size implementation. The fiber-attached microchip laser enables perfectly alignment-free operation, which assures extreme stability of the proposed setup. Additionally, the proposed setup offers low-cost implementation thanks to a low-cost microchip laser and a short photonic crystal fiber.

A potential application of the present light source is endoscopic implementation. However, some technical challenges remain. First, backscattered CARS (epi-CARS) light should be efficiently lead to a spectrometer. Thus some optical components such as dichroic beam splitter and another optical fiber for collection of the CARS light need to be placed between the low pass filter and the focusing objective with compact size. Second, higher sensitivity sensor such as an EM-CCD camera may be needed to detect weak epi-CARS signals. Third, in order to obtain images from an in-vivo sample, scanning mechanism such as a MEMS mirror need to be equipped in a module rather than an external translation stage [29].

Although we have applied the maximum entropy method to reconstruct a Raman spectrum, there are several alternatives for analysis of an experimentally obtained CARS spectrum. For example, the time domain Kramers-Kronig transformation is also routinely applied for reconstruction of Raman spectrum [30]. If the sample has isolated peaks in its Raman spectrum, some quantitative indices can be used to quantify the molar concentration without full reconstruction of a Raman spectrum [31].

We have successfully found some conditions of a laser source and a PCF that enabled efficient excitation of CARS. However, we should note that, in general, predicting CARS efficiency with different laser source conditions such as pulse duration is nontrivial because the whole process involves multiple nonlinear processes stated above and thus may require detailed numerical or experimental analysis.

Because the Stokes light generated in the PCF is broadband, in principle it can contribute as pump and probe light in CARS process as well as single pulse CARS setups [11–14]. However, as well as previous works with similar setups, we did not observed such a contribution in experimental data [4, 19, 22, 31]. This may be because the supercontinuum light seeded by a nanosecond laser is incoherent and thus CARS processes where two or more Stokes light components are involved cancel due to uncorrelated phase of each process [22].

5. Conclusion

We proposed and demonstrated a new simple and compact light source for multiplex CARS microscopy. The setup has performance of sensitivity, bandwidth, and spectral resolution suitable for biomedical applications, which covers the entire fingerprint region (500-1800 cm⁻¹). The setup also enables ultra-compact, alignment-free, and low-cost implementations on the basis of current technologies of microchip laser. It was found that combination of shorter PCF length and higher incident power to the PCF was preferable for better performance of bandwidth and sensitivity. A Raman spectrum of a polystyrene bead in the entire fingerprint region and a CARS image of 2 μm polystyrene beads were successfully obtained. We believe that the present work greatly expands the utility of CARS microscopy to a wide variety of fields such as endoscopy [20, 21].