1. Introduction

Coherent Raman spectroscopy (CRS) [1] utilizes the signature of intrinsic molecular vibrations to achieve chemical specificity in nonlinear imaging without the need for extrinsic fluorescence labels. Two methods of CRS have been implemented for biomedical imaging. Coherent anti-Stokes Raman scattering (CARS) generates a blue-shifted wave at a frequency ω_as = 2ω_p - ω_s, where ω_p, and ω_s, are the frequencies of the pump and the Stokes. The dependence of the anti-Stokes signal on the intensities of the pump and Stokes pulses is $I_{AS}\propto P_s{P}_p^2$, where P_s and P_p represent the powers of the Stokes and the pump pulses, respectively. An alternative method of CRS is stimulated Raman scattering (SRS), the stimulated optical process corresponding to ordinary Raman scattering, which involves the energy transfer between the pump and the Stokes beams through the well-known stimulated Raman process. The amplitude of the SRS signal is I_SRS$\propto$P_sP_p. Although the energy transfer is usually very weak, with a relative power change on the order of 10⁻⁶ or less, it can be detected by phase sensitive detection techniques such as lock-in amplification [2]. This involves the intensity modulation of the pump (or Stokes) beam at a radio frequency (RF) and measuring the amplitude of the crosstalk induced in the Stokes (or pump) beam at the same frequency of the modulating waveform.

The first implementations of CARS and SRS imaging used bulky (and expensive) Ti:sapphire laser systems [2–5], permitting optimization of CRS methods for imaging. For practical application in research or clinical settings, however, a more compact and robust system is desirable. Hence, a number of labs have turned to fiber-laser based systems, with a variety of schemes for generating pump and Stokes pulses [6–19]. In our implementation, we use soliton self-frequency shifting (SSFS) in a photonic crystal fiber for tunability [11,18], an approach which has also been implemented for CRS imaging by other workers [6, 20–22]. SSFS offers the advantage compared to some other fiber-based methods of rapid tunability [23] with an intrinsically transform-limited pulses.

Photothermal imaging is another, less well-known, label-free microscopy technique [24–26] based on the detection of optically induced local heating. Photothermal detection is a sensitive technique most suitable for detection of light absorption by molecules that do not fluoresce. In the photothermal process, an intensity-modulated pump beam is focused in the sample, and the absorbed energy is released in the sample as heat, resulting in modulation of the refractive index at the focal point. The probe beam interrogates this induced modulation and is detected by a photodetector. Photothermal imaging of blood cells [27] and simultaneous CRS and photothermal imaging of vascular structures [28] have been reported.

Although both CRS and photothermal microscopy are based on the same optical pump-probe configuration, their imaging mechanisms and functionalities are different and complementary. In this paper we demonstrate a multi-modal setup based on a single femtosecond fiber laser to perform CARS, SRS and photothermal microscopy. Applications to imaging of lipid distributions in mouse brain and hemoglobin in blood cells are presented.

2. Experimental setup

The experimental setup is shown in Fig. 1. A femtosecond fiber laser (IMRA Femtolite-100) serves as the light source, providing optical pulses at 802 nm with approximately 120 fs temporal pulses width and 9.5 nm spectral linewidth. The repetition rate of the pulse train is 75 MHz and the average output optical power is about 100 mW. As shown in the inset of Fig. 1, the relative intensity noise (RIN) of this fiber laser is lower than −130 dB/Hz at 100 kHz, and has a typical 1/f noise characteristic below 2 MHz. However, no resonant peaks of RIN at high frequency were observed [10, 19]. The laser beam is split into two paths, providing the pump and the Stokes beams, respectively. The pump passes through a motorized delay stage (Opto-Sigma), and its intensity is modulated by a Pockels cell (Conoptics). In the Stokes path, a 2-m-long PCF (Crystal Fiber NL-PM-750) is used to introduce a SSFS so that the wavelength difference between the pump and the Stokes can be varied. Through the nonlinear SSFS process, part of the laser phase noise is converted into intensity noise, and therefore the RIN is increased by approximately 10 dB in the low frequency region as shown in Fig. 1. Nevertheless, the RIN level is still lower than −120 dB/Hz at 100 kHz. Computer controlled wavelength tuning of the Stokes beam is accomplished by a voltage-controlled variable optical attenuator (VOA), which controls the optical power that is coupled into the PCF. This allows the Stokes pulse to be continuously tuned from 850 nm to 1200 nm in the form of a fundamental soliton in the PCF [11]. A long-pass filter following the PCF blocks the remnant power at 802 nm and high-order solitons generated below 850 nm. A 5 cm SF6 glass rod is inserted in the Stokes path to introduce a linear frequency chirp, which matches the chirp introduced by the Pockels cell in the pump path, thus enhancing the spectral resolution through spectral focusing [29]. Additional linear chirping can also be applied for both the pump and the Stokes through multiple passes in SF6 glass rods if higher spectral resolution of CRS is required [18, 29]. The pump and the Stokes beams are recombined with a dichroic beam-combining filter and focused into the sample by an objective lens (Nikon 40x, 0.6NA). As this objective lens is not optimized for near IR wavelength, the loss for the 802 nm pump beam and the 1050 nm Stokes beam are approximately 1.5 dB and 7.3 dB, and the average power of the pump and Stokes on the sample are about 9 mW and 0.2 mW, respectively. A second, confocal, objective lens collects the signal. The sample is placed on a piezo stage (Nano-Drive H100-X), which can move in the horizontal x-y directions within 100 µm range. Each direction of the piezo stage is driven by a waveform generator (Agilent 33250A). The step size for scanning was 0.5 µm with an integration time of 30 ms per pixel. The acquisition is limited primarily by the speed of the xy scanning stage.

 

Fig. 1 Experimental configuration with controllable Stokes wavelength (λ_s) and relative pump delay (δt). BS: beam splitter, SM: silver mirror, PCF: photonic crystal fiber, LP: long-pass filter, BP: bandpass filter, SF6: SF-6 glasses rod, DC: dichroic beam combiner, OB: objective lenses, PC: pockels cell, Pol: polarizer, PD: photodiode. Inset on the right: measured relative intensity noise of the fiber laser (red) and of the soliton self-frequency shifted pulses (black).

Download Full Size | PDF

For CARS measurements, the intensity of anti-Stokes signal is detected by a spectrometer (Ocean Optics, Maya 2000 Pro), while both the Stokes and the pump wavelengths are rejected by a bandpass filter to avoid saturation of the photodetector. For SRS measurements, only the Stokes beam is allowed to pass through the long-pass filter and be collected by a large-area InGaAs photo-detector (TTI TIA-5251) with 125 MHz bandwidth. The output of the photo-detector is processed by a lock-in amplifier (Stanford Research Systems SR-850) to detect the stimulated Raman gain (SRG) on the Stokes beam introduced by the modulation of the pump, and to reject the large DC component of the Stokes beam. The relative pulse delay between the pump and the Stokes pulses is controlled through the motorized delay stage in the pump path. A laptop computer controls synchronization of the lock-in amplifier with the waveform driving the Pockels cell. Scanning of the piezo-electric scanning stage and data acquisition are also controlled by the laptop computer. The same setup as used for SRS is also used for photothermal microscopy, in which the Stokes beam acts as the probe. Due to the long time constant of the photothermal signal, typically on the microsecond level, the relative pulse delay of the probe relative to the pump is no longer important.

Mouse brain samples were prepared as follows. The mouse was transcardially perfused with ice-cold PBS under anesthesia and then with 4% paraformaldehyde. After decapitation, the brain was isolated and fixed overnight in 4% paraformaldehyde. It was then embedded in O.C.T. compound (Sakura Finetek USA) and sectioned coronally at 10 µm thickness using a vibrating microtome (Leica Microsystems). Slices were mounted on a slide with 10 microliter distilled, deionized H₂O and covered with a cover slip.

3. Results and discussion

To demonstrate the capability for multi-modal microscopy based on the setup described above with only a single fiber laser, we imaged slices of mouse brain and blood cells with the three techniques and compared the images. Figure 2 shows examples of images of a mouse brain slice. To obtain CARS and SRS images, the frequency difference between the pump and the Stokes beams was tuned to 2920 cm⁻¹, corresponding to a C-H stretching Raman resonance of lipids and protein [1,2,18]. Figure 2(a) and 2(b) show the SRS images of sliced mouse brain at different locations. The dark regions may represent the cross section of blood vessels in the mouse brain [30]. The intensity of the pump power was modulated at 100 kHz for this measurement, the highest frequency that can be handled by our lock-in amplifier. In addition to the lipid and protein structure, several bright spots also appear in Fig. 2(a) and 2(b). Figure 2(c) shows the CARS imaging of the same area as Fig. 2(b), which is identical to Fig. 2(b) except without the bright spots. Additional measurements indicated that when a relative time delay between the pump and the Stokes pulses was introduced, the lipid and protein signal in the SRS images of Fig. 2(a) and 2(b) disappeared while only the bright spots remained, with the signal amplitude relatively independent of the pump-Stokes time delay. We conclude that the signal in the bright spots results from a photothermal effect caused by light absorption at the pump wavelength.

 

Fig. 2 Images of mouse brain tissue slices. (a) SRS image (100 µm x 100 µm scan). (b) SRS image in a different region (50 µm x 50 µm scan). (c) CARS imaging in the same region as in (b). The pump-Stokes frequency detuning was set at 2920 cm⁻¹ corresponding to a Raman resonance of lipids and protein.

Download Full Size | PDF

To further illustrate the photothermal effect, the two insets in Fig. 3(a) show images obtained at 100 kHz pump modulation frequency, without (bottom left) and with (top right) ~10 ps of time delay between the pump and Stokes pulses. With the introduction of this time delay there is no temporal overlap between the pump and the Stokes pulses, and the signals due to SRG disappear. However the photothermal signals remain. The amplitude of the photothermal signal increases with decreasing modulation frequency as shown in Fig. 3(a). We interpret this to result from decreased heat transfer from the excitation volume between excitation pulses as the modulation frequency is increased. Meanwhile, since the RIN is higher at lower frequencies as indicated in Fig. 1, the signal-to-noise ratio (SNR) of the photothermal signal remains almost constant in the 10 kHz – 100 kHz frequency window. The SNR was on the order of 13 dB for SRS, and 20 dB for photothermal signals. Figure 3(b) shows the normalized signal amplitudes as a function of the pump powers for CARS, SRS and photothermal images. As expected, the CARS signal amplitude is proportional to the square of the pump power (shown with a slope of 2 in dB/dBm plot, where dBm is the power in decibel millwatts), and the SRS signal is linearly proportional to the pump power (shown with a slope of 1 in Fig. 3(b)). Figure 3(b) also shows that the amplitude of photothermal signal is linearly proportional to the pump power [31], indicating that the thermal modulation is induced by one-photon absorption instead of a two-photon effect [27]. Heme proteins are well known to absorb light in the near-infrared region around 800 nm [32]. Based on the known one-photon [32] and two-photon [33] cross sections for hemoglobin at 800 nm, we estimate that for our setup the one-photon contribution to the photothermal signal is roughly an order of magnitude stronger than the two-photon contribution, consistent with the observed linear dependence on pump power. The higher two-photon cross section at 830 nm (while the one-photon cross section remains approximately unchanged), and tighter focusing and/or higher laser power could account for the quadratic dependence observed in [27]. The measured photothermal signal in the mouse brain slices may arise from residual hemoglobin which was not completely washed out during sample preparation. Hence, hemoglobin (or other heme proteins) may give rise to the observed photothermal signal.

 

Fig. 3 (a) Amplitude (left y-axis) and SNR (right y-axis) of the photothermal signal as a function of modulation frequency. Bottom-left inset: SRS image when pump-Stokes pulses are temporally overlapped, and top-right inset: photothermal image with 10 ps time delay between pump and probe pulses. (b) Normalized amplitudes of CARS (circles), SRS (diamonds) and photothermal (stars) signals as the function of the pump power in dBm.

Download Full Size | PDF

Figure 4 shows CARS and photothermal images of red blood cells on a slide. The human blood sample was collected by drawing ~200 µl of whole blood from a finger tip. Ethylenediaminetetraacetic acid (EDTA) was added to the whole blood at a concentration of 2.0 mg/ml to act as an anticoagulant. A thin layer of blood cells was produced by a blood smear. A second coverslip was placed over the sample and sealed with tape in a sandwich configuration. The CARS images with a 2920 cm⁻¹ pump-Stokes frequency difference shown in Fig. 4(a) and 4(b) clearly show the outline of blood cells in different regions on the slide. For comparison, Fig. 4(c) shows the photothermal image in the same region as Fig. 4(b). Figure 4(c) was obtained with a 20 kHz modulation frequency of the pump so that the photothermal signal is much higher than that of SRS, and therefore Fig. 4(c) is overwhelmingly a photothermal image. An identical image was obtained when the modulation frequency was increased to 100 kHz (the highest frequency of the lock-in amplifier used in our setup), where the photothermal signal remained higher than the SRS, and the signal was independent of the pump/probe delay. A much higher modulation frequency, in the MHz range, would be required to isolate SRS from photothermal for red blood cells. The strong photothermal signal is attributed to hemoglobin content in the blood cell, which absorbs pump photons at 800 nm and contributes to a modulation in the probe beam. In general, pump-induced local heating can create a number of photothermal effects, such as lensing, deflection, refraction and diffraction through thermal-induced refractive index change as well as volume expansion [31]. Photothermal imaging of blood cells has been previously reported [27]. In our measurements, the photothermal signal was linearly proportional to the pump power for both mouse brain tissue and red blood cells. The CARS images in Fig. 4(b) and the photothermal image in Fig. 4(c) were acquired in the same region. Both clearly show the circular shape of blood cells with a diameter of approximately 6 µm. However, they simultaneously reveal different physical and biochemical properties of the sample. On the other hand, SRS and photothermal images may be generated simultaneously, and the best way to separate them would be by changing the pump-probe time delay or the modulation frequency.

 

Fig. 4 Images of blood cells on glass slides. (a) CARS image with 2920 cm⁻¹ pump-Stokes frequency detuning, (b) CARS image in a different region on the slide, and (c) photothermal image at the same region as in (b) with 20 kHz pump modulation frequency.

Download Full Size | PDF

4. Conclusion

Based on a single femtosecond fiber laser excitation source, we demonstrated a multi-mode and label-free technique capable of performing CARS, SRS and photothermal microscopy simultaneously. Each of these three microscopy modalities has its own advantages and limitations, whereas the combination of them can help better understand the physical and biochemical characteristics of the sample. Sliced mouse brain samples as well as red blood cells were imaged to demonstrate the capability of the proposed technique. The distributions of lipid and hemoglobin in these sample slides were characterized. The excitation system is based on a simple fiber-laser source coupled with a photonic crystal fiber for SSFS. Although each of the components has been demonstrated previously, the combination represents an advance in the direction of a simple, flexible, and robust instrument for imaging in practical settings not tied to a massive optical table. Stokes pulse generation by SSFS opens the further possibility of tuning across or switching between different vibrational resonances on a very rapid time scale [23]. The speed of imaging, especially for SRS imaging, can be greatly increased by using higher modulation frequency and by using balanced photodetection to minimize the impact of high frequency RIN of fiber laser [19]. Faster sample scanning stage, and high power objective lenses optimized for near IR wavelengths will also help improve the system.