The mid-infrared spectral range, spanning approximately from 2 to 15 μm, is known as the molecular fingerprint region, since most molecules produce characteristic vibrational signatures. Absorption bands arising from vibrational excitations are used as a contrast mechanism in FTIR spectromicroscopy in order to identify small particles and analyze complex mixtures. Biological tissues are particularly challenging due to their intricate spatial organization at the microscopic level and their complex chemical composition of thousands of low concentration compounds. Hyperspectral data can be either acquired via point-by-point rastering or by projecting the light onto a two-dimensional array detector. A major drawback of confocal point-by-point spectromicroscopy is the long acquisition time necessary for creating hyperspectral maps of large areas. When using a synchrotron as a mid-infrared radiation source, the high brightness of the beam provides high flux at small apertures. Since the diffracted-limited spot size depends on the probing wavelength [1], a small aperture size of approximately 3 μm is often used while affording reasonably short acquisition times [2–4]. Synchrotron-radiation-based FTIR spectromicroscopy is nowadays an indispensable tool for microbiology [5,6] and materials sciences [7,8]. Furthermore, temporal incoherence of synchrotron radiation sources avoids speckle and allows reaching high spatial resolution [9]. In the past few years, quantum cascade lasers (QCL) in the mid-IR have made tremendous progress, with tunability over a useful range of 3–13 μm using a combination of individual laser chips. Discrete-wavelength imaging systems using QCL sources have become commercially available [10,11]. However, the multivariate nature of the information required, e.g., for reliable clinical diagnosis or for most research applications, favors the use of broadband sources. Furthermore, the coherence of QCLs is often detrimental to infrared image quality. Therefore, powerful broadband light sources with continuous spectral power density in the mid-infrared, spatial coherence, and temporal incoherence are needed to complement the synchrotron source in spectromicroscopy experiments.

Supercontinuum generation (SCG) occurs when a narrowband short laser pulse undergoes extreme spectral broadening by a combination of phenomena due to the exaltation of optical nonlinearities [12]. Since the first report [13] by Alfano and Shapiro in 1970, this effect was demonstrated in a number of experiments. Optical fibers are highly appropriate to SCG thanks to the tight confinement of light in a material exhibiting large third-order nonlinearity over meter-long distances and the dispersion management offered by the strong waveguiding effect. In optical fibers, SCG combines the main advantages of the synchrotron light source, i.e., large bandwidth and high spatial coherence, in a compact configuration. Further, chalcogenide glass optical fibers [14] provide high IR-transmission and strong third-order nonlinearity. The approach developed by Petersen et al., yielding the broadest reported mid-infrared SC to date [15], was to pump a multimode chalcogenide fiber with a femtosecond optical parametric oscillator tuned to the zero dispersion wavelength (ZDW) of the fiber’s fundamental mode. Remarkable spectral broadening from 2 to 13 μm was reported, but the low repetition rate of the optical parametric oscillator limited the spectral power density to $\sim {10}^{-2}\mathrm{\mu W}/\mathrm{nm}$. More recently, the same group demonstrated broadband light generation in tapered chalcogenide glass fibers [16] with an increased power density of 10 μW/nm. Optical fibers composed of fluoride glasses such as ZBLAN (${\mathrm{ZrF}}_4\text{-}{\mathrm{BaF}}_2\text{-}{\mathrm{LaF}}_3\text{-}{\mathrm{AlF}}_3\text{-}\mathrm{NaF}$) or ${\mathrm{InF}}_3$ exhibit low loss from the ultraviolet (UV) to the mid-infrared up to 5 μm and allow for SCG throughout their transparency range [17]. Recently, a cascade of fluoride and chalcogenide glass fibers pumped by a thulium-doped fiber laser was shown to be promising for SCG up to 6.5 μm [18]. A successful application of supercontinuum source in infrared chemical mapping has already been reported in the C-H stretching region around $2950{\mathrm{cm}}^{-1}$, but with a limited spatial resolution of 20 μm [19]. We also point out that some other studies have already produced broader spectra via difference frequency generation in gallium selenide crystal [20] or via a combination of quadratic and cubic phenomena in gallium arsenide crystal [21] with high power density across the fingerprint region. However, we are not aware of any successful experimental demonstration of the suitability of these sources for spectromicroscopy. Here, we demonstrate that fiber-based supercontinuum sources in the mid-infrared enable diffraction-limited FTIR spectromicroscopy.

To ensure continuous spectral power density and temporal incoherence in the mid-infrared, we designed the supercontinuum laser source schematically depicted in Fig. 1(a). The high-repetition-rate pulse train was generated in a passively mode-locked Tm-doped fiber laser oscillator featuring a high-modulation-depth saturable absorber mirror and emitting long transform-limited Gaussian pulses with 40 ps duration centered at 1952.7 nm. In order to reach the energy levels required for mid-infrared supercontinuum generation, the pulse energy was further amplified in a cladding-pumped amplifier based on a large-mode-area Tm-doped fiber. The pulse train was amplified to an average power of 3.7 W, corresponding to 0.9 μJ pulse energy. The 4.15 MHz repetition rate of the pulse train ensured operation with several Watts of average power without any degradation of the pulse quality. Subsequently, the pulses were launched into a 20 m long infrared transparent ZBLAN glass fiber (ZSF-9/125-N-0.2, FiberLabs, Japan) where they underwent extreme spectral broadening by a combination of the processes of modulation instability and soliton fission. Modulation instability, which is favored by pumping the nonlinear fiber in the anomalous dispersion regime (ZDW $\sim 1.7\mathrm{\mu m}$ for ZBLAN glass [22]) with long pump pulses, is at the origin of both spectral smoothness and temporal incoherence in this system. Furthermore, this pump configuration maximizes the amount of light generated in the infrared [23]. More specifically, the high-energy input pulses trigger the formation of high-order solitons in the nonlinear fluoride fiber. The pulses break into fundamental solitons due to the combined effects of high-order dispersion and stimulated Raman scattering. The Raman-induced soliton self-frequency shift [24] (SSFS) contributes to the spectral broadening up to approximately 4 μm, limited by the increase of the chromatic dispersion and the decrease of the nonlinear coefficient. Furthermore, the nonlinear process of modulation instability results in large amplitude and phase fluctuations of the fundamental solitons resulting from the fission process. In turn, these amplitude and phase fluctuations are converted into wavelength fluctuations and temporal jitter through SSFS and dispersion. As shown in Fig. 1(b), the output spectrum is continuous and relatively flat from 2 μm to approximately 3.75 μm. In this range, the spectral power density is on the order of 1 mW/nm.

 

Fig. 1. Characteristics of the supercontinuum laser source. (a) Block diagram of the supercontinuum source. (b) Spectral profiles measured (blue) and computed (red) at the output of the supercontinuum source. For the experiments described in the main text, the spectrum is filtered by a bandpass filter centered at 3500 nm, indicated by the gray box. (c) Spectrally resolved modulus of the complex degree of mutual coherence. Inset: Intensity distribution at the output of the supercontinuum source in the filter bandwidth. The scale bar indicates 20 μm length.

Download Full Size | PDF

We have modeled the experiment in order to evaluate the residual coherence at the output of the laser. The spectro-temporal properties of the supercontinuum were studied by solving the generalized nonlinear Schrödinger equation, including amplified spontaneous emission noise in the Tm-doped fiber amplifier. On Fig. 1(b) the red curve shows the computed spectrum. When comparing the simulation results with the measured spectrum [Fig. 1(b) in blue] it is clear that the simulation is in agreement with the measurement. We have calculated the spectrally resolved modulus of the complex degree of mutual coherence, which gives a measure of the phase stability and temporal coherence [12],

The angular brackets denote an ensemble average over independently generated pairs of SC spectra [$E_1(\lambda ,t)$, $E_2(\lambda ,t)$] obtained from 10 simulations with noise. The results plotted in Fig. 1(c) show that the output spectrum exhibits some degree of coherence around the pump wavelength only and is almost fully incoherent in the vicinity of the C-H stretching band. For the experiments described below we used a 500-nm-wide bandpass filter centered at 3500 nm (Thorlabs FB3500-500) to select the spectral region corresponding to the C-H stretching vibrational modes. With these experimental conditions, the coherence length can be approximated to $\mathrm{\Delta }\lambda /{\lambda }^2$ corresponding to $\sim 20\mathrm{\mu m}$. The short coherence length of our laser source, which is comparable to that of a thermal source, and its high brightness make our laser source excellent for FTIR spectromicroscopy with diffraction-limited resolution.

For the spectromicroscopy experiments described in the following the fiber output laser beam was collimated to approximately 25 mm using a 90-deg off-axis parabolic mirror and routed into a Thermo Scientific 8700 FTIR spectrometer. A Thermo Scientific Continuum IR microscope operated in transflection mode, equipped with a $50\mathrm{\mu m}\times 50\mathrm{\mu m}$ MCT/A detector was used to record chemical images of a liver sample by measuring the reflectance of the sample in the C-H stretching region around $2900{\mathrm{cm}}^{-1}$ (3.45 μm). Liver specimens were obtained from the Centre de Ressources Biologiques (CRB) Université Paris-Sud, Orsay, France. Access to this material was in agreement with French regulations. The sample was a thin section of human liver from a fine-needle biopsy from a non-alcoholic fatty liver (NAFL) disease patient and was prepared by cryosectioning to 10-μm thickness without embedding in paraffin and mounted on a low-e IR reflective microscope slide (Kevley Technologies Chesterland, Ohio).

As shown in Fig. 1(b), the output spectrum of the laser extends from 2 to 4 μm with approximately 1 mW/nm power density. While this high power level is necessary for enabling and fully exploiting the nonlinear processes within the optical fiber, it could damage optical components, especially the aperture blades and the detector, as well as the sample when tightly focused. Besides, it is important to limit the power on the MCT detector in order for it to operate in its optimal linearity range, where the detector response is proportional to the actual signal strength. Therefore, we reduced the source intensity by two orders of magnitude using a neutral density filter (NDIR20B, Thorlabs Inc. Newton, New Jersey). We also reduced the operational bandwidth by means of the bandpass filter centered at 3500 nm (Thorlabs FB3500-500). Furthermore, the bandpass filter avoids spectral folding from above the Nyquist limit, which in the case of commercial FTIR systems is typically half the frequency of the He–Ne laser in the interferometer ($\sim 7900{\mathrm{cm}}^{-1}$).

In order to evaluate signal-to-noise ratios (SNR), spectra were measured at $4{\mathrm{cm}}^{-1}$ resolution with 1–256 co-added scans, with a mirror velocity of 1.89 cm/s. The root mean square (RMS) noise level was estimated in the $4000–4200{\mathrm{cm}}^{-1}$ range, where no specific absorption peaks are present, in absorbance units, and the SNR was calculated as the inverse of the RMS noise value.

We compared the performance achieved with the laser source to that achieved with the synchrotron source for two small aperture sizes of $5\mathrm{\mu m}\times 5\mathrm{\mu m}$ and $3\mathrm{\mu m}\times 3\mathrm{\mu m}$. At the $5\mathrm{\mu m}\times 5\mathrm{\mu m}$ aperture size, the laser source yields a SNR above 200 in 8 scans, while 32 scans of the synchrotron radiation source are needed to reach the same SNR level. The FTIR maps and selected spectra shown in Fig. 2 were obtained at the $3\mathrm{\mu m}\times 3\mathrm{\mu m}$ aperture size.

 

Fig. 2. Comparison of spectral maps recorded with the fiber-based supercontinuum source and with the synchrotron source. Chemical maps of lipid vesicles in a liver section generated on the ${\nu }_{\mathrm{asym}}{\mathrm{CH}}_2$ absorption of lipids at $2920{\mathrm{cm}}^{-1}$. (a) Optical micrograph of the mapped area. The scale bar indicates 20 μm length. (b) Map recorded with 128 scans at $3\mathrm{\mu m}\times 3\mathrm{\mu m}$ aperture with the synchrotron source. (c) Map recorded with 16 scans at $3\mathrm{\mu m}\times 3\mathrm{\mu m}$ aperture with the supercontinuum laser source. (d), (e) Spectra of the C-H stretching regions showing ${\mathrm{CH}}_3$ and ${\mathrm{CH}}_2$ peaks recorded with (d) the synchrotron radiation source and (e) the supercontinuum laser source. Blue and red colors on (d), (e) correspond to the positions of the same color markers on (a). Insets: SNR measured versus the number of scans at the highest resolution of $3\mathrm{\mu m}\times 3\mathrm{\mu m}$.

Download Full Size | PDF

Using the laser source [Figs. 2(c) and 2(e)], we obtained $\sim 1\%$ RMS noise level with 16 co-added scans. The map was comparable to that obtained with the synchrotron source with 128 co-added scans [Figs. 2(b) and 2(d)], which is an eight-times slower acquisition. To compare the performance of the two sources, the signal-to-noise ratio for the supercontinuum source and the synchrotron source versus the number of scans at the aperture of $3\mathrm{\mu m}\times 3\mathrm{\mu m}$ are reported in inserts of Figs. 2(d) and 2(e). Figure 2 shows that the supercontinuum source allows measuring chemical images matching those obtained with the synchrotron source at diffraction-limited aperture sizes. At this resolution, the stability of the laser system is sufficient to obtain SNRs suitable for spectromicroscopy studies.

However, we have observed slight discrepancies between the spectra recorded with the laser and the synchrotron source. First, it is important to point out that we did not perform any baseline correction on the spectra in order to show the real quality of raw data and, most importantly, that this quality was reached with only 16 scans using the laser source, which is eight times faster than the measurement performed with the synchrotron radiation. The spectra recorded with the SC source appear noisier below the $2850{\mathrm{cm}}^{-1}$ peak. On one hand, the supercontinuum emission in this region is already decreased as shown on Fig. 1(b). On the other hand, due to the combination of the small aperture size and the cutoff frequency of the applied bandpass filter, the effect might seem more pronounced. Further, we also observe a slight baseline shift and an apparent decrease of the $2950{\mathrm{cm}}^{-1}$ peak when using the laser source. In both the synchrotron radiation case and the laser case, the red spectra peak heights at $\sim 2925{\mathrm{cm}}^{-1}$, which are proportional to concentration, are consistently $\sim 0.6\text{a.u.}$, while the blue spectra are $\sim 0.1\text{a.u.}$ with the synchrotron source and $\sim 0.25\text{a.u.}$ in case of using the laser source. We attribute these discrepancies to the combination of alignment and source instability of the laser as well as the small aperture and raster step size. The existence of the above-mentioned instabilities is further supported by the non-theoretical increase of the SNR versus the number of scans [inset, Fig. 2(e)]. These discrepancies can be amplified, as collecting eight times fewer averages in case of the laser measurement compared with the synchrotron source makes them appear, somewhat misleadingly, more significant.

As shown in the insets of Figs. 2(d) and 2(e), the SNR is improved by increasing the number of scans. In the case of the synchrotron source, the improvement is expected to be limited by the room temperature blackbody. However, for the laser source, the SNR is expected to keep on improving, since the room temperature blackbody has a much lower flux than the laser. Furthermore, we have observed that the SNR for the laser source remains somewhat constant when decreasing the aperture size. This is in contrast with the case of the synchrotron source where the signal-to-noise ratio obtained for the full beam intensity decreases by a factor of 100 if the aperture size decreases from $20\mathrm{\mu m}\times 20\mathrm{\mu m}$ to $3\mathrm{\mu m}\times 3\mathrm{\mu m}$. We attribute this behavior to the higher photon flux passing through this small aperture in the case of the laser source.

In conclusion, we have demonstrated, in focusing on the spectral region centered on 3.5 μm, the possibility to perform infrared spectromicroscopy measurements using a custom supercontinuum laser source. Our study shows that the SNR obtainable with the laser source is comparable to or surpasses those available with a synchrotron source in shorter acquisition times even at very small aperture sizes. Such performance at such small aperture size cannot be achieved with the thermal source due to its low brightness. For a comparison of the supercontinuum source with the thermal emitter, see Supplement 1. However, the input power of the source can be further optimized to achieve ideal time/SNR balance. Furthermore, the implementation of focal plane array hyperspectral FTIR imaging is underway with the same source to enable two-dimensional FTIR imaging microscopy.

Although the spectral bandwidth of the presented supercontinuum laser is highly limited in comparison to the thermal and especially the synchrotron source, it is clear that specific applications will highly benefit from the possibility of diffraction-limited infrared spectromicroscopy using such bench-top sources. These bench-top sources will enable faster turnaround and open the way for the application of FTIR spectromicroscopy in time-sensitive, on-demand applications such as medical diagnosis.