1. Introduction

The promise of chemical contrast for biological imaging has inspired extensive work towards the development of coherent Raman imaging methodologies [1–7]. Recently, imaging based on stimulated Raman scattering (SRS) has gained popularity [8–15] and has been demonstrated at video-rate [16]. Compared to coherent anti-Stokes Raman scattering (CARS), SRS microscopy provides comparable signal levels [15] and does not suffer from spectral distortions due to the nonresonant background that arises from the electronic response of the medium [8]. The first reported SRS microscopy performed multiplex SRS imaging using a 1 kHz laser source [8], and the same group recently developed a higher repetition rate source for multiplex SRS [17]. Unlike CARS spectroscopy, in SRS the signal is generated at the frequency of the Stokes pulse and is detected as either gain or loss in the incident beam amplitudes. Such small changes in gain or loss can be sensitively recorded by lock-in detection, which can be readily performed for single-channel (single vibrational mode) imaging. Thus most of the SRS microscopy to date has imaged single vibrational modes [9–11]. Other SRS microscopy approaches have used pulse-shaping methods to access the characteristic modes of different molecular species [12], used spectral focusing [13] or a time delay between a pump and Stokes pulses to select particular vibrational modes [14].

The desire to simultaneously image multiple chemical species motivates the development of multiplex SRS methods where high-resolution SRS spectra are obtained at each image pixel. However, since SRS is not a background-free method, SRS microscopy that utilizes multiplex frequency-domain detection must be capable of monitoring small gain or loss signals. While lock-in cameras have recently become available [18], they are not yet widely used. Here we present spectrally-resolved imaging based on Raman-induced Kerr effect spectroscopy (RIKES) [19], a close cousin to SRS and CARS. Single mode optical heterodyne-detected (OHD-RIKE) microscopy has been recently demonstrated [20]. RIKES spectroscopy has also been recently discussed [21] as an alternative to femtosecond stimulated Raman spectroscopy (FSRS) for studying chemical reaction dynamics [22, 23]. The multiplex RIKES microscopy method we employ is based on a single laser source, providing broadband chemical imaging, which we demonstrate on polymer and biological samples. We compare multiplex RIKES and multiplex SRS microscopy and discuss their advantages and disadvantages for various chemical imaging applications.

Like CARS and SRS, RIKES is a four-wave mixing process as illustrated in Fig. 1 . In CARS, the signal is generated at a new optical frequency, Raman-shifted from the pump field. While in principle this should make CARS a background-free measurement, in practice CARS suffers from a nonresonant background that is electronic in nature and offers no chemical specificity. In SRS, the signal is generated at the same frequency as the incident light and is manifest as a loss or gain in the incident fields. Thus lock-in detection or other difference-detection methods must be used to extract the signal. As in SRS, the RIKES signal is generated at the frequency of the input fields. However, in RIKES, the signal arises from anisotropic changes in refractive index induced by the exciting fields. These changes in refractive index exhibit resonances when the difference in frequency between a pump field of frequency ω_p and Stokes field of ω_S matches a Raman active vibrational mode of the medium (ω_p - ω_S = ω_vib) [24, 25]. In multiplex RIKES, shown in Fig. 1(c) a broadband Stokes pulse is employed with a strong pump field. The pump-induced birefringence causes a polarization change in the Stokes beam, which can be detected as transmission through a polarizer aligned to be perpendicular to the Stokes beam. The spectrum of the detected RIKES signal has been shown to reflect the spontaneous Raman spectrum [24].

 

Fig. 1 Energy level diagrams for (a) CARS, (b) single mode SRS and RIKES, (c) multiplex SRS and RIKES.

Download Full Size | PDF

Several experimental configurations for RIKES exist [25]. Here we use a standard approach that minimizes nonresonant background contributions. In this approach the pump field E_P is circularly polarized, while the broadband Stokes field E_S is linearly polarized. The RIKES signal is detected through a polarizer with its alignment perpendicular to the Stokes field. In this configuration the signal field E_RIKES is proportional to a combination of third order susceptibility tensor components:

Each complex susceptibility component contains a nonresonant background contribution:

Since both${\chi }_{1122}^{\left(3\right)}$ and ${\chi }_{1212}^{\left(3\right)}$will have similar nonresonant background contributions, these contributions largely cancel, allowing the resonant signal to be detected free from the distorting background signals seen in CARS. In addition, Kleinman symmetry dictates that ${\chi }_{1122}^{\left(3\right)}={\chi }_{1212}^{\left(3\right)}$off vibrational resonance [26], which minimizes cross-phase modulation signals arising from $\mathrm{Re}\left[{\chi }_{1122}^{\left(3\right)}-{\chi }_{1212}^{\left(3\right)}\right]$that appear in SRS [20]. Thus the primary source of background in a RIKES experiment is a birefringent background caused by depolarization of the probe field between the input polarizer and the analyzing polarizer [25]. Some degree of depolarization occurs at every optical element, and precise alignment is required to minimize the effect. In general, a Stokes field rejection of ~10⁻⁶ is sought. When sample birefringence is low, the birefringent background in RIKES can be considered to add incoherently to the RIKES signal, avoiding distortions to the RIKES spectrum [27]. In samples with higher birefringence, deviations of the pump field from purely circular polarization or the Stokes beam from linear polarization can lead to spectral distortions [28].

With slight modifications the RIKES setup can also be used to perform optically heterodyne-detected RIKES (OHD-RIKES) [25]. In CARS microscopy, heterodyne detection is complicated by the fact that a local oscillator must be found that has the same frequency content as the signal field, which is frequency shifted from the incident fields. In SRS the Stokes field acts as the local oscillator since it is detected along with the signal field. This can also be done in RIKES by introducing a controlled amount of Stokes field into the detector, for example by a slight rotation of the cross-polarizer [25]. However, to detect $\mathrm{Im}\left[{\chi }_{1122}^{\left(3\right)}-{\chi }_{1212}^{\left(3\right)}\right]$when a circular pump beam is used, a 90° phase shift must be introduced between the local oscillator and RIKES signal. This can be achieved with a λ/4 plate inserted before the sample in the Stokes optical path; a slight misalignment of the principal axis of the waveplate with respect to the Stokes polarization axis produces a slightly elliptical probe, where the degree of ellipticity controls the strength of the local oscillator that will leak through the cross-polarizer [25]. Alternatively, in RIKES with a linear rather than circularly-polarized pump, the RIKES signal is proportional to ${\chi }_{1122}^{\left(3\right)}+{\chi }_{1212}^{\left(3\right)}$, and the RIKES and Stokes fields are in phase [20, 25]. This configuration does not provide the nonresonant background suppression of the circular pump approach and was not pursued here because of detector saturation by the nonresonant background.

2. Experimental implementation

Our experimental setup is shown in Fig. 2 . A Ytterbium-doped femtosecond fiber laser (IMRA, FCPA mJewel D-400) provides 1042 nm, <350 fs, 2 μJ pulses at 200 kHz. A beamsplitter takes 10% of the output and focuses it into a nonlinear photonic crystal fiber [29] to generate the broadband Stokes pulse, an approach used in multiplex CARS [6]. The remaining light is spectrally narrowed using a combination of two bandpass filters (CVI XLL-1064 nm) that can be angle-tuned to provide ~10 cm⁻¹ pulses centered at 1042 nm to act as the pump. The pump pulses are then sent through a Glan polarizer and quarter wave (λ/4) plate to circularly polarize the light. The pump and Stokes pulses are focused at the sample by an air objective. As high-NA objectives cause significant reduction in polarization purity, we employ a polarization-maintaining objective (NA 0.5, Olympus UPLFLN20XP), providing a linear Stokes polarization purity of better than 1:10⁻⁴. A scanning piezo stage (PI P 612.2SL) raster-scans the sample to obtain 2D images. The RIKES signal is collected by a condenser (NA 0.8) and directed into a spectrometer (Jobin Yvon IHR320) where spectra are acquired with a Pixis 100F camera. We collect the spectra on the anti-Stokes side where the detector quantum efficiency is favorable. For multiplex SRS measurements the λ/4 plate in the pump was rotated to produce linear polarization parallel to the Stokes field. The analyzer (P3) was removed and a neutral density filter was inserted to reduce the transmitted light to avoid detector saturation. This allowed consecutive imaging of the same sample area with RIKES and SRS for direct comparison. Chopper 2 blocks the spectrometer camera during readout, running at a rate of 200 Hz. Spectra are acquired with an integration time of 2.5 ms, followed by 2.5 ms of readout time. With this scheme we achieve a spectral acquisition rate of 200 Hz.

 

Fig. 2 Experimental setup for multiplex RIKES microscopy. BS: 90/10 beamsplitter; PCF: photonic crystal fiber; P1: Glan polarizer 1; DBS: dichroic beamsplitter; P2: Glan polarizer 2; P3: Glan polarizer 3.

Download Full Size | PDF

3. Experimental results

To demonstrate multiplex RIKES microscopy, we used a sample of 10 µm diameter polystyrene beads embedded in a toluene-based sealant (Cytoseal-60, Richard-Allan Scientific). Because the dense polystyrene beads caused significant birefringence and scatter as a function of sample scanning position, an additional chopper was added in the setup (chopper 1 in Fig. 2) to modulate the pump beam at 100 Hz. This allowed the acquisition of difference spectra on a timescale fast enough to perform imaging and remove the birefringent background.

Figure 3 shows images of polystyrene beads embedded in toluene sealant taken using multiplex SRS and multiplex RIKES. RIKES images taken with equal excitation conditions show better contrast and higher signal to noise than those taken with SRS. Because toluene and polystyrene both have similar Raman modes near 3000 cm⁻¹ [30], the ratio of peak strengths at 3050 cm⁻¹ and 2920 cm⁻¹ is used to enhance imaging contrast. SRS images show little to no contrast with high noise at each of the individual Raman modes and low contrast with high noise in peak ratio images. RIKES provides fair contrast with lower noise at 2920 cm⁻¹ with much higher contrast at 3050 cm⁻¹. The RIKES peak ratio images feature good contrast with low noise and compare very favorably to the respective SRS images. Polystyrene and toluene spectra for each method shown in Fig. 3(g) and Fig. 3(h) reflect the higher SNR seen in the RIKES images.

 

Fig. 3 Multiplex SRS images (left column) and multiplex RIKES images (right column) of polystyrene in toluene sealant. (a,b) Images at 2920 cm⁻¹. (c,d) Images at 3050 cm⁻¹. (e,f) Images of 3050 cm⁻¹/2920 cm⁻¹ peak ratio. (g,h) Average of 42 spectra taken in polystyrene beads and toluene with SRS and RIKES, respectively. Excitation conditions were 12.2 mW pump, 0.05 mW Stokes, 2.5 ms acquisition, 60µm x 60µm scan.

Download Full Size | PDF

Figure 4 illustrates the effect of the pump-chopping scheme for improving image contrast. RIKES images of polystyrene beads embedded in toluene sealant are again shown, plotting the ratio of 3050 cm⁻¹/2920 cm⁻¹ peaks without (Fig. 4(a)) and with (Fig. 4(b)) pump-chopping. Figure 4(c) shows spectra taken inside polystyrene and toluene sealant, averaged over 42 pixels. The upper two spectra show the averaged polystyrene spectra with the pump on (green) and off (yellow) to indicate the effectiveness of the chopping scheme, which largely removes the birefringent background. We note that the signal-to-noise ratio (SNR) of these images could be readily improved with higher Stokes power. Here we were limited by low power available from the PCF.

 

Fig. 4 RIKES images polystyrene beads embedded in toluene sealant (12 mW pump and 0.05 mW Stokes power, 2.5 ms/pixel). (a) ratio of 3050 cm⁻¹/2920 cm⁻¹ amplitudes without chopping to remove birefringent background, (b) ratio of 3050 cm⁻¹/2920 cm⁻¹ amplitudes with chopping, (c) average of 42 spectra taken inside polystyrene (ps) (brown) and toluene (blue) sealant. Also shown is the averaged polystyrene spectrum without the chopping scheme (green), along with the averaged birefringent background recorded when the pump beam is blocked (black).

Download Full Size | PDF

We also performed RIKES imaging in onion cells, shown in Fig. 5 . As illustrated in Fig. 5(a) and (b), we see good chemical contrast: off-resonance, at 2750 cm⁻¹ we obtain low background signals while when resonant with the C-H stretch at 2940 cm⁻¹ we see cellular structure. Figure 5(c) shows a different onion region near the cell membrane, where birefringence is higher. At different spatial locations throughout the scan we observe that the RIKES signal changes magnitude and even flips sign, as shown in the spectra in Fig. 5(d). The sign flip that occurs over a single scan can be attributed to spatially-varying total birefringence across the sample. Shin et al. showed that the birefringence in onion cells can vary significantly between the cytoplasm and cell membrane [31]. The effect is strongest immediately after the sample is prepared and declines over the timescale of ~30 hours. As our images were taken within approximately one hour of sample preparation, the birefringence effects are extremely prominent around the cell membrane in the top right corner of Fig. 5(c). As a result, our exciting pulses have varying degrees of birefringence-induced polarization ellipticity across the scanned area. The effect of an arbitrary elliptical pump polarization on RIKES signal magnitude and lineshape has been described by Levenson and Song [28]. Perfectly circular polarization gives rise to a standard Lorentzian lineshape expected in a Raman spectrum. Deviation from circular polarization introduces dispersive elements into the lineshape due to constructive and destructive interference between Raman and frequency-independent contributions to the signal. Figure 5(c) and (d) illustrate the problem of lineshape variations across a non-homogenous sample due to changes in strain birefringence near the onion cell membrane.

 

Fig. 5 Multiplex RIKES images of onion cells with excitation conditions of 13 mW pump, 0.7 mW Stokes, 2.5 ms per pixel acquisition at (a) 2750 cm⁻¹ (off-resonance) and (b) 2940 cm⁻¹ (resonant with the C-H stretch). (c) Multiplex RIKES image at 2940 cm⁻¹ at a different sample location. (d) Spectra showing negative signal and positive signals taken at locations indicated in (c).

Download Full Size | PDF

4. Discussion

For microscopy applications, CARS has the benefit of generating a signal field that is higher frequency than either the incident pump or Stokes fields, in principle allowing background-free detection of the signal. In practice, however, a nonresonant background interferes with the signal, distorting lineshapes. Considerable effort has been made to suppress the nonresonant background in different CARS microscopy experiments using polarization [32], time-domain [7, 33, 34], pulse-shaping approaches [35, 36]. Alternatively, algorithms to remove the nonresonant background from multiplex CARS images have been developed [37, 38]. In many of the above cases, the nonresonant background has been used to amplify the weak CARS signal by acting as a local oscillator for heterodyne detection. Achieving an optimized SNR requires control of the relative strength of signal and local oscillator field [39], which is difficult to achieve with sample-dependent nonresonant background signal strengths.

In SRS, the signal is manifest as a gain or loss in the excitation fields. Thus the excitation field itself acts as the local oscillator for heterodyne detection, with the appropriate relative phase to probe the desired imaginary component of the third order susceptibility. However, in SRS, the relative strength of signal to local oscillator is not controlled, and if laser noise is significant, the use of excessive local oscillator degrades the SNR [25]. High-frequency modulation methods, combined with lock-in detection enable SRS detection in a frequency regime where typical laser sources have very low noise [9]. This methodology has been successfully demonstrated by a number of groups for single-mode SRS imaging [9–11]. For multiplex detection with a typical CCD array, readout times are slow, requiring that the detection be performed at low frequencies where laser noise is significant. For multiplex implementations using PCF-generated continua as the Stokes field this is particularly important since these sources can exhibit significantly higher low frequency noise than that of the pump pulse [40]. Thus the use of excessive local oscillator field can degrade the SNR, making control over the relative strength of signal and local oscillator particularly important for low frequency detection applications. Another practical concern is that the large Stokes field readily saturates the CCD, requiring filtering that reduces the amount of signal reaching the detector.

RIKES offers an attractive alternative to CARS and SRS for multiplex imaging because of its ability to suppress distorting nonresonant background and cross-phase modulation signals. It also probes different components of the third order susceptibility tensor, providing complementary information to other measurements. Heterodyne-detected RIKES (OHD-RIKES) is appealing because of its ability to control the relative amplitude of signal to local oscillator for optimizing the SNR, as clearly demonstrated in the spectroscopy literature [25]. In microscopy, the birefringent properties of the sample can complicate the ability to carefully minimize the birefringent background and maintain optimum polarization properties of the input fields. This places practical limits on the benefits that can be derived from RIKES [20]. In all of the polymer and onion samples we studied, we observed a large birefringent background caused by depolarization of the Stokes beam. Consistent with recent single-mode RIKES microscopy work, we note that this background is small relative to the nonresonant background present in CARS microscopy [20]. To remove the birefringent background we implemented a chopping scheme that significantly improved the imaging results. In the polystyrene/toluene samples, the birefringent background was readily removed by chopping. In this case we found that multiplex RIKES provided an enhanced SNR compared to multiplex SRS microscopy. Extensive analysis of signal-to-noise issues in RIKES, SRS and other coherent Raman methods has been performed by Eesley [27]. For our measurements, the main reasons for the poorer performance of multiplex SRS compared to multiplex RIKES are the loss of signal due to filtering to avoid detector saturation in SRS, and the larger relative amount of noisy Stokes field incident on the detector in SRS measurements. We note that this difference in performance is greater in multiplex compared to single-mode microscopy where lock-in detection can be used to suppress laser noise [20]. In some samples, such as regions of the onion tissue, we observed very large birefringent background signals, at times approaching the saturation level of the detector. In these samples we also observed distortions to the RIKES spectra induced by ellipticity of the exciting fields. These backgrounds distort the spectral lineshapes of Raman peaks and reduce chemically specific image contrast.

In general, sample-dependent birefringence will also reduce the ability to use OHD-RIKES to provide SNR enhancement. In CARS microscopy, it has been shown that OHD detection has its greatest advantage when the addition of a local oscillator field brings the overall signal level to within the shot-noise-limited regime [41, 42]. Signal levels necessary to reach this limit are detector dependent. For the CCD camera we used here, the birefringent background was often large enough to reach this regime [41], limiting the possible SNR improvement achievable with OHD-RIKES [27]. For optimum enhancement in OHD-RIKES, the strength of the local oscillator should be significantly larger than that of the background [27]. An additional complication for employing multiplex OHD-RIKES with circular pump excitation is the necessity of adjusting the relative phase of the signal and local oscillator to obtain the desired imaginary components of the third order susceptibility. This can be achieved by adding an achromatic λ/4 plate into the broadband Stokes beam, but may be difficult to implement while maintaining high polarization purity over the full bandwidth.

5. Conclusions

In summary we have demonstrated a single-laser approach to multiplex RIKES and multiplex SRS microscopy. With minor changes to the experimental setup, one can switch easily between multiplex RIKES and multiplex SRS imaging to probe different elements of the third order susceptibility. A simple modulation scheme can reduce the birefringent background in multiplex RIKES and generate images with good chemical contrast. In samples with low birefringence, such as the polymer samples we studied, we have shown multiplex RIKES to have a higher SNR than multiplex SRS for equivalent experimental conditions. In samples with large birefringent backgrounds, such as onion tissue, the ability to optimize the SNR and obtain undistorted spectra with multiplex RIKES microscopy is compromised, reducing the advantages of the method. Improvements in the SNR of both multiplex RIKES and multiplex SRS would be enabled by the use of a lock-in camera [18] that would permit higher frequency modulation as is used in single-mode SRS imaging [9]. In the current implementation, the ability to optimize the SNR via heterodyne detection is limited due to detector saturation from background signals. A lock-in camera would also reduce this problem, enabling OHD-RIKES, which has been shown to enhance the SNR under many conditions [25]. In addition, the use of balanced detection schemes for multiplex OHD-RIKES [43] could provide further optimization.