Optical coherence tomography (OCT) is an emerging imaging technology providing high-resolution crosssectional images of tissue structure on the micron scale in situ and in real time [1]. OCT depicts a reflectivity map from changes in refractive index from different tissues, but it lacks in chemical selectivity. Differentiating between pathologic and normal tissues with similar optical or morphological properties can be challenging with OCT as a standalone imaging modality. Different approaches have been studied to overcome this limitation, e.g., engineered microspheres were used to enhance the contrast [2]. Adding inherent chemical selectivity (relying on molecular eigenmodes) to OCT without the use of exogenous contrast agents can noninvasively improve the discrimination between healthy and pathological tissue [3]. Hence, a device noninvasively obtaining simultaneous molecular and structural (“morphomolecular”) information can help to diagnose diseases long before structural changes occur and expand the range of noninvasive diagnostic techniques. Molecular sensitivity can either be enabled in a spontaneous manner such as Raman scattering or in a nonlinear coherent process such as coherent anti-Stokes Raman scattering (CARS) [4,5]. Several groups have already demonstrated a combined Raman-OCT system. However, two separate independent light sources were needed making the system complex and cost intensive. Since the Raman effect is very weak, detection needs a sensitive and highly optimized instrumentation.

Furthermore fluorescence can hide the Raman signal resulting in a challenging extraction of the Raman bands [6–9]. Molecular-sensitive OCT with CARS was first presented by Boppart and co-workers [10,11]. CARS signals were obtained by heterodyne detection. Two highly sophisticated laser sources had to be used to generate the pump and Stokes wavelength to target only one vibrational mode at a time making the setup rather complex. A major drawback of CARS is the presence of nonresonant background, which is not related to any vibrational resonance. The nonresonant background results from instantaneous electronic nonlinearities that also lead to formation of polarization at the anti-Stokes frequency. Several detection methods have been developed to overcome this limitation [12–15]. Numerical techniques have been used to extract the resonant signal out of the combined signal [16–22]. However, these techniques either alter the resonant signal, require an assumption about the nonresonant background, require highly sophisticated setups with integrated spatial light modulator (SLM) and nitrogen cooled spectrometer or necessitate the implementation of two synchronized light sources preventing widespread application of these techniques. Recently, Ideguchi et al. [23] demonstrated dual-comb CARS with fully suppressed nonresonant background while recording spectral information over the whole fingerprint region in less than one millisecond. However, two femtosecond lasers were needed. Stimulated Raman scattering (SRS), a background-free alternative to CARS, requires expensive and complex setups [24,25]. A simple straightforward and cost-effective approach to access background-free CARS spectra is Fourier transform (FT) CARS [26]. In FTCARS the initial pulse impulsively excites all the vibrational modes lying within the bandwidth of the laser source, and a second time-delayed pulse probes this field. By time resolving the CARS signal, the nonresonant signal can be completely eliminated due to the long coherence time of the molecular vibrations as compared to the instantaneous response of the nonresonant background.

In this work, we demonstrate for the first time, to our knowledge, a single-source spectral domain OCT (SD-OCT)/ online FTCARS-imaging modality applied in biological samples (human bone). In this approach, online FTCARS obtains time-resolved vibrational data and the background free CARS spectrum simultaneously without the need of averaging by use of an ultrahigh linear piezo stage, which also avoids the use of a second laser to monitor and correct for the scanning.

The experimental setup is shown in Fig. 1. A custom built 76 MHz ultrafast Ti:sapphire laser [27] was used as a single light source for OCT and online FTCARS. The mode-locked laser produced sub-12-fs pulse-duration with a bandwidth of 90 nm centered at 780 nm, and a maximum output power of 1 W. The output power of the laser was adjusted via polarization beam splitter (PBS) and half-wave plate (HWP) placed before the Michelson interferometer. The total group delay dispersion (GDD) of about $4000{\mathrm{fs}}^2$ was compensated by chirped mirrors (Thorlabs DCMP175) and fine-tuned with fused silica wedges (Femtolasers). Near transform-limited pulses were confirmed by optimizing the second order autocorrelation at the sample position. The FTCARS signal was recorded as a function of the time delay between two identical collinear pump and probe pulses from the first Michelson interferometer as shown in Fig. 1. A second Michelson interferometer having a $50:50$ beam splitter was introduced in the setup (Fig. 1) for the SD-OCT. For FTCARS, the first pulse impulsively excites all molecular vibrational levels allowed by the laser bandwidth, and the second pulse probes the Raman coherence created by the first pulse. The probe pulse was time delayed with a highly linear piezo stage (Physik Instrumente P-629.1CD). This translation stage operated at a frequency of 8 Hz with the possibility to go up to a maximum of 20 Hz and a step resolution of 1 nm. A travel range of 1.8 mm corresponded to a resolution of better than $4{\mathrm{cm}}^{-1}$ and enabled continuous probing of the Raman coherence induced by the excitation. To perform spectral filtering of the CARS signal, a Razor edge long pass filter (Semrock SEM-FF01-715/LP-25) was placed before the microscope objective to cut off the blue edge of the laser spectrum. A microscope objective lens ($20\times$, 0.40 NA, Newport, 8MX-20) was used in this experiment to focus the laser radiation onto the sample. The overall excitation and probe beam power was kept below 50 mW to avoid sample damage. The forward propagating CARS signal was collected by a similar microscope objective lens and short pass filtered (Semrock, SEM-FF01-720/SP-25) to isolate the CARS signal, which was then recorded using a photomultiplier detection unit (Hamamatsu H7422-40). A three-dimensional scanning stage (Physik Instrumente, M26821LNJ and PD73Z2CNW) was used for sample scanning to perform CARS imaging. The data acquisition was accomplished with a high-resolution DAQ card (ATS-660) at 5 MS/s. The online monitoring and data processing (low-pass filtering) was done via Labview 2012 (National Instruments).

 

Fig. 1. OCT-FTCARS setup: the output of a custom built Ti:sapphire laser is attenuated with a half-wave plate (HWP) and a polarization beam splitter (PBS). After long-pass filtering (LP) the laser light, the beam is guided into a Michelson interferometer to split the beam into the initial pump beam and a scanning beam (SCN) to generate the time-delayed probe beam. The beam is further coupled into a second Michelson interferometer for the OCT, consisting of a detection unit with a telescope (T), a grating (G), a lens (L), and a camera (C). Dispersion compensation is performed with chirped mirrors (CM) and wedges (W). The OCT scanning is performed with galvanometric scanners (GS). For the CARS, the signal was short-pass filtered (SP) after passing the sample (S) through the objective (O) and detected with a photomultiplier (PM).

Download Full Size | PDF

In the case of SD-OCT, the delay scanner (SCN) of the first interferometer was positioned at zero delay, and the second interferometer splits the incoming pulse with a $50:50$ beam splitter (Thorlabs, UFBS5050) into the reference and sample arms. The reference arm consisted of high-reflective Ag mirrors, a neutral density filter and a corner cube to match the path length of the sample arm. The sample arm used the same optical path as the online FTCARS. However, for OCT, the objective lens was replaced by a low-NA scanning lens (Thorlabs, LSM03-BB, 36 mm focal length) in combination with galvanometric scanners (Cambridge Technologies, 6220H) to acquire wide-field structural OCT imaging. The low-NA scan lens allowed for scanning a wide field of view without clipping the scanning beam at the back aperture of the microscope objective. The detection arm consisted of a grating (Wasatch photonics, 1200 l/mm), an objective lens (Zeiss, Planar T 1,4/85-ZF-IR-I), and a line scan camera (Atmel AViiVA EM4 2014). The OCT data was collected with a Camera Link Card (NI, PCIe-1427). The SD-OCT achieved a SNR of 95 dB, axial and transversal resolution of $\sim 4$ and 12 μm, respectively, with a roll-off $\sim 6\mathrm{dB}$ at 0.8 mm. We first performed a large structural scan $3\times 4\mathrm{mm}$ of the bone sample with a line rate of 37 kHz, which is limited by the maximum speed of the camera. We acquired 1024 A-scans along the fast axis with 512 B-scans along the slow axis, in less than 15 s.

The fast axis of the galvanometric scanner runs at 36 Hz, and the slow axis runs at 0.07 Hz and scans maximum angle of $\pm 7°$.

All the OCT processing is done prior preview in Labview. After the OCT data was collected, an en face image of the acquired data was generated to deliver a wide field structural map, which is depicted in Fig. 2.

 

Fig. 2. Structural en face scan of human bone obtained using the OCT setup. The indicated red box depicts the region of interest for the chemical mapping depicted in Fig. 4.

Download Full Size | PDF

This overview is used as a prescreening to zoom into certain regions of interest for obtaining a detailed chemical spectrum. The identification of CARS signals was simplified due to online tracking, resulting in a straightforward sample and the optics alignment. All the spectra were optimized for maximum peak amplitude and collected in less than 125 ms.

Figure 3 shows a background free CARS spectrum of $\sim 100\mathrm{\mu m}$ thick human bone sample obtained from an archaeological excavation. To our knowledge, it is the first time that this technique is used to extract CARS signal from a biological sample online. Figure 3. (left) shows the time domain CARS signal of bone. The strong signal at the zero delay is due to the nonresonant background. After online low-pass filtering of the signal at longer delays, far away from the zero delay, the beating pattern due to the two dominant vibrational modes can be seen in the online low pass filtered data shown in the inset. FT of the indicated region in Fig. 3. (inset, left) yields the background-free CARS spectrum as shown in Fig. 3 (right).

 

Fig. 3. Time domain data of 100 μm thick human bone with a scan delay of 1.2 mm (left), online low pass filtered signal of the indicated region of the time domain data (inset), resolved CARS spectrum after FT of the time domain data without the contribution of the nonresonant background (right). The Raman peak located at $\sim 960{\mathrm{cm}}^{-1}$ corresponds to the phosphate content and the peak at $\sim 1070{\mathrm{cm}}^{-1}$ corresponds to the carbonate.

Download Full Size | PDF

A conventional bright-field microscopy (Axio Imager 2, Carl Zeiss) image of the human bone sample is shown in Fig. 4 (left). The microscope image of the bone sample was in good agreement with the CARS image of the phosphate band with the Raman peak located at $\sim 960{\mathrm{cm}}^{-1}$ representing the most prominent mineral peak of bone. Figure 4 (top, right) shows an enlarged image of the marked area in the bright field image (orange box). The CARS image of the region of interest (ROI) ($\sim 30\mathrm{\mu m}\times 30\mathrm{\mu m}$) was acquired with a step size of 1 μm in Fig. 4 (bottom, right). The CARS signals generated correspond to an imaging depth of $\sim 20\mathrm{\mu m}$. Both images depict concentric rings called concentric lamellae surrounding an opening called the central canal (Harvesian canal) indicating an osteon of compact bone, which is composed of bone matrix [28].

 

Fig. 4. Brightfield image of the indicated region in Fig. 2 (red box). The orange box shows the region, where the chemical mapping was performed (left). The zoom into the indicated region (right, top), chemical phosphate map of the indicated region (right, bottom).

Download Full Size | PDF

Osteons are characteristic for mature bone which change shape during bone growth, modeling and remodeling. Therefore, CARS can be used to differentiate between mature and immature bone [28]. CARS spectrum of bone shows vibrational bands according to the tissue composition with the main Raman line located at $\sim 960{\mathrm{cm}}^{-1}$ and a small peak at $\sim 1070{\mathrm{cm}}^{-1}$ correlated to the inorganic content consisting of carbonate. The carbonate-to-phosphate ratio is a measure of bone quality and could be used to determine bone aging, bone type, and its biomechanical properties [29].

In conclusion, we have demonstrated a simple multimodal imaging platform employing a single light source providing simultaneous structural and chemical selectivity in the sample under investigation. OCT was capable of delivering a large structural scan, which was used to choose a ROI to perform chemical mapping using CARS. Background-free CARS spectra were obtained with rapid online FTCARS by implementing a fast highly linear piezo nanopositioning stage to extract CARS spectroscopic parameters in a single scan without the need of averaging in the 800 to $1200{\mathrm{cm}}^{-1}$ spectral range with a spectral resolution of better than $4{\mathrm{cm}}^{-1}$. The potential of this technique was demonstrated in human bone samples, which revealed the phosphate and carbonate bands of the bone matrix. This technique could be further extended using a similar setup with a sub-6-fs laser and software upgrade for the piezo stage to extract CARS spectroscopic parameters covering the spectral range from fingerprint to lipid within 50 ms. Future implementation of field-programmable gate arrays (FPGA) for high-speed data processing will ensure real-time preview. The presented online FTCARS approach does not require integration of a calibration or correction laser making the system simpler for applications. Furthermore, this system allows for fast label-free chemical information in contrast to traditionally histological methods. CARS spectra of various locations within large samples can be extracted sequentially by arbitrary point scanning within 125 ms to investigate different components. Therefore, this system might offer great potential for fast histo- and cytopathology of specimens with unknown composition.