1. Introduction

CARS microscopy has evolved over the past decade as a powerful label-free imaging modality based on intrinsic vibrational contrast [1]. A wide variety of important applications involving CARS microscopy have been demonstrated. These range from imaging lipid droplet biology [2], imaging axonal myelin in spinal cord injuries and demyelinating diseases [3], identifying obesity related risks in cancer [4] and atherosclerosis [5] and the rapid detection of pathogens [6]. Moreover, it has been clearly shown that there is a tremendous benefit in combining CARS with other imaging modalities such as TPEF and SHG in a multimodal imaging and spectroscopy platform to obtain a complete picture of the health of the biological tissue [3–5,7]. However, in order to truly extend the benefits of multimodal CARS microscopy to human health, development of a multimodal CARS probe in the form of an endoscope or miniaturized hand held microscope is essential. In fact, significant progress has been made in the development of TPEF and SHG imaging endoscopes and miniaturized microscopes for in vivo imaging applications [8–13]. For CARS microscopy which involves the pump and Stokes beams for excitation of the nonlinear optical signal, progress towards fiber based endoscopy has been quite slow [14–16]. The key challenges have been (i) the efficient delivery of the ultrafast pump and Stokes light using optical fibers (ii) efficient fiber based collection of the CARS signal (iii) miniaturization of laser scanning mechanisms and (iv) efficient design of chromatic aberration corrected miniature optics for achieving high resolution CARS images. Past research efforts [15,16] have focused mostly on overcoming the challenges of fiber based light delivery and collection. In particular, laser scanning and focusing at the sample to generate a CARS image was achieved using standard macro-optics, such as a galvanometric scanner and a microscope objective. A miniaturized objective lens [17] was demonstrated for deeper penetration in CARS imaging, but was used with a standard microscope objective and a galvanometric scanner. Only recently, progress was reported related to the design and modeling [18] and implementation [19] of a fiber scanning based CARS endoscope.

While miniature microscopes and hand held probes based on TPEF [8–11], fluorescence confocal microscopy [12] and SHG [13] have been fabricated earlier, there is no report of a miniaturized CARS microscope to the best of our knowledge. In this paper, we demonstrate for the first time, a miniaturized multimodal CARS microscope based on MEMS scanning mirrors and custom miniature optics. Moreover, a single femtosecond pulsed laser is used as the light source to produce the CARS, TPEF and SHG images. A scheme first demonstrated by our group [20], using the supercontinuum generated in a nonlinear photonic crystal fiber (PCF) as the Stokes beam and part of the femtosecond pulse laser as the pump beam, is employed for CARS imaging. The high resolution and distortion-free images obtained from various samples such as a USAF target, fluorescent and polystyrene microspheres and biological tissue successfully demonstrate proof of concept, and pave the path towards future integration of parts into a handheld multimodal CARS probe for non- or minimally-invasive in vivo imaging. The use of a single femtosecond laser as the light source for the miniature multimodal CARS microscope holds further promise for making the whole setup more compact for future clinical use. We describe below, details of the design as well as the experimental setup to test the performance of our MEMS-based miniature multimodal CARS microscope.

2. Materials and Methods

2.1 Experimental setup

The light source for multimodal nonlinear optical excitation is based on a single Ti:sapphire femtosecond laser (Tsunami, Spectra-Physics, Mountain View, CA) producing ~65 fs pulses at 80 MHz repetition rate and tunable between 720 nm - 1000 nm. This light is split into pump and Stokes arms as shown in Fig. 1 . As described in detail in our earlier work [20], about 300 mW of this light at ~800 nm is directed into a commercial photonic crystal fiber (PCF) module (NKT Photonics, FemtoWhite CARS) for creating the Stokes beam. The remainder (~400 mW) comprises the pump beam used for CARS, TPEF and SHG imaging. The supercontinuum output from the PCF is band pass filtered (Chroma Technology) so that it consists of wavelengths between 1014 nm - 1067 nm. The pump beam is sent through a computer controlled delay stage and then recombined with the Stokes beam at the dichroic mirror. The diameter of the pump and Stokes beams is reduced using a pair of plano-convex lenses so that they do not overfill the MEMS mirror of diameter 500 µm. Light reflected at 45 degrees by the MEMS mirror is incident on the miniature optics held inside a stainless steel barrel on a vertical rail. The sample is mounted on a three axis automated stage so that it can be placed in the focal plane of the incident pump and Stokes beams. Nonlinear optical signal from the sample is collected by a long working distance, air objective (Mitutoyo, 20x, 0.42NA) in the forward direction. This is filtered by a 680 nm short pass filter (Chroma Technology) to remove the excitation light. Appropriate band pass filters are used to separate the SHG, TPEF, and CARS signals as described in Section 3.2. This light is coupled into a large (1mm diameter) core multimode fiber (Thorlabs) and detected by a photomultiplier tube (PMT) (Hamamatsu, H7422-40) for image generation.

 

Fig. 1 Bench top miniaturized multimodal CARS microscope. Details of components are as follows: (2) Faraday isolator (3) Prism compressor (4) MEMS mirror (5) Miniature optics inside stainless steel tube (barrel) (6) Sample for imaging (7) microscope air objective (20x) for collecting nonlinear optical signal from sample (8) Lens for coupling light into collection fiber (9) Multimode fiber (10) Voltage Amplifier circuit to amplify electrical signal, for driving the MEMS mirror, provided by the Field Programmable Gate Array (FPGA).

Download Full Size | PDF

2.2 Miniature optics

The miniature optics were designed with the intention of integrating them at a later stage inside a portable miniature microscope for in vivo imaging of the rat spinal cord. This imposed the requirement that its distal end would have a tip whose outer diameter is no more than 3 mm. This meant that the optics had to be designed with a diameter of less than 2 mm such that it would still provide an NA of ~0.6, working distance of ~400 µm and enable sub-micron resolution imaging. Traditionally, gradient index (GRIN) lenses with relatively high numerical apertures (NA) of 0.5- 0.6 have been preferred in multi-photon microendoscopy applications, mainly because of their low cost and nearly diffraction-limited image quality [21]. However from our optical design analysis, it became clear that the two excitation wavelengths at 800 nm (pump beam) and 1040 nm (Stokes beam) separated by ~200 nm for CARS imaging of lipids, poses a significant challenge in terms of compensating the longitudinal chromatic aberration in the GRIN lenses. Hence we opted for designing a conventional miniaturized front end objective that would perform to the required specification. This is indeed very challenging since a large numerical aperture is required from small diameter optics. The final design consists of multiple lenses that are ~1.8 mm in diameter and have varying prescriptions. Two different glass types, SF4 and FK51 are chosen to compensate for the chromatic aberration at 800 nm and 1040 nm. The effective NA of the front end objective is 0.6 and the designed field of view is 100 x 100 µm with a working distance of 400 μm. Relay lenses are included in the optical design in order to image the MEMS mirror on the back aperture of the miniature objective with a magnification of 5x. Appropriate antireflection coatings are applied on all optics to maximize throughput of excitation and emission light. Figure 2(a) is a ray-trace diagram of the optical beam as it propagates through the miniature optics and focuses on the sample just past the coverslip, while Fig. 2(b) depicts the computer-aided design of the construction of the stainless steel tube (barrel) into which all of the miniature optics are assembled (BMV Optical, Ottawa, Canada). The fully packaged barrel is ~4.1 cm long and is shown in Fig. 2(c). A thin glass window seals off the distal end of the barrel thus permitting water immersion.

 

Fig. 2 (a) Ray-trace diagram of the optical beam as it enters from the right after the MEMS mirror and is relayed to the miniature focusing optics on the left, through a 5x beam expander assembly including the field lens. (b) Computer-aided-design of the barrel. (c) Photograph of the fully packaged barrel. A penny is included for size comparison.

Download Full Size | PDF

2.3 MEMS scanning and image reconstruction

Miniaturized laser scanning in fluorescence based endoscopes and miniaturized microscopes is achieved by means of cantilever fiber-scanners [8–10], as well as MEMS scanning mirrors [11,12,22–24]. MEMS scanners operating at resonant frequencies offer the advantages of adjustable and fast frame rates and allow batch fabrication.

In our microscope, a two dimensional scanning MEMS mirror (Fraunhofer IPMS, Germany) with a diameter of 500 μm is employed for beam scanning and image generation. The device consists of a circular silicon plate in gimbal mounting suspended by a total of four torsional spring bars [25]. The reflectivity of the mirror plate is enhanced by a thin layer of aluminum and was measured to be ~80% at 800 nm. This is in excellent agreement with the reflectivity value reported for a bulk aluminum mirror. Based on this measurement of 80% reflectivity at 800 nm, we expect the reflectivity at 1040 nm to be 96% or better (as it would be for a bulk aluminum mirror). Independent resonant oscillation of the mirror plate (fast axis) and the frame (slow axis) itself, is set up by applying a high voltage to the comb electrodes adjacent to the mirror and frame. A Field Programmable Gate Array (FPGA) (Altera DE2) board running a 50 MHz system clock is programmed to sweep from higher frequencies to lower frequencies with a sweep time of 5 s, until it stops at the resonant frequency for each axis. A custom-built voltage amplifier circuit amplifies the rectangular waveform output from the FPGA to drive the MEMS oscillations along the fast and slow axes. When high voltage is applied to both axes, a Lissajous pattern is scanned, with a filling factor determined by the particular ratio of the slow and fast resonant frequencies. An optical scan angle of +/−17 degrees along both axes is obtained by applying 40 V and 70 V, at the resonant frequencies of 1.336 KHz and 16.99 KHz to the slow and fast axis, respectively. In order to avoid overfilling the back of the barrel in the optical geometry of the bench top setup the scan angle was reduced by setting the resonant frequencies to higher values of 1.429 KHz and 17.225 KHz for the slow and fast axis respectively.

The nonlinear optical signal from the sample is collected in the forward direction as shown in Fig. 1. The loss in efficiency in the signal collection when the MEMS scanner is tilted at its most extreme angles is measured to be ~8%. This was traced back to a slight clipping of the beam at the entrance of the field lens and barrel assembly when the MEMS mirror scans, causing subsequent loss in excitation power at the sample. The optical signal from the sample is sent to the PMT followed by an amplifier-discriminator unit (Ortec 9327). TTL pulses from the discriminator are sent to the FPGA where they are synchronized with respect to the FPGA clock. The time difference between subsequent events is encoded and sent to the PC via a custom-made board that uses the FT2232H chip. A program running on the PC receives these data and saves them to disk. An image reconstruction program simulates the trajectory of the laser and creates a mapping table. It uses the stored data files in the PC and transforms the scanned vector data in a frame period into a 512 × 512 pixel image. The phase delay between the driving electrical signal and the mechanical response of the MEMS mirror is adjusted in order to remove ghost images in the final 512 x 512 pixel image.

We would like to now comment on the frame rates that are achievable with our multimodal miniature microscope. The goal is to achieve a self repeating Lissajous scan pattern of the optical beam that fulfills the conditions that i) every pixel in the 512 x 512 pixel image is hit at least once and ii) the spatial coverage is very uniform across the FOV. We first experimentally measure the resonance curves for the slow and fast axes at 40 V and 70 V, respectively. From this data, we determine the resonant frequencies for the desired optical scan amplitude. These slow and fast axis resonant frequencies are then expressed in terms of the number of system clock cycles (ticks), n_s and n_f, respectively, where the system clock of the FPGA is at a much higher frequency of 50 MHz. The resulting Lissajous pattern will self repeat after n ticks where n is the least common multiple of n_s and n_f. Thus the frame repeat rate is given by (50 MHz) / n. It should be clear that different choices of n_s and n_f, or in other words, the slow and fast axis resonant frequencies, will give different frame repeat rate. Our choice of the slow and fast axis resonant frequencies of 1.429 KHz and 17.225 KHz, respectively provided a good spatial coverage over a 512 x 512 pixel image. The resulting frame rate of 4 Hz is sufficient for our current requirement of imaging stationary samples. This frame rate was identical for all imaging modalities in our microscope. We collected a fixed amount of data (10MB, roughly 9 million nonlinear optical events) per image file which included multiple frames. The brighter images had more nonlinear optical events per frame, and therefore the time required to acquire this data was less.

3. Results

3.1 Microscope characterization using standard samples

The resolving power of the microscope was investigated by acquiring transmission images of a USAF resolution test target (Edmund Optics). The femtosecond pulsed laser tuned to 720 nm, followed the pump beam path as shown in Fig. 1. A water drop was placed on top of the tip of the barrel and the USAF target glass slide was placed facing down, touching this water drop such that the smallest features on the target were centered on the focused beam spot with a reduced average power of 1.5 mW. The reconstructed image is shown in Fig. 3(a) . The smallest element 6 in the 7th group is at the left side of the image. It has a line spacing of 228 line pairs/mm, corresponding to a line width of approximately 2.2 μm. There is no distortion in the shape of the individual lines in the image except for a conical distortion in which the right side of the image is slighty rotated counter clockwise. This is a known artifact due to the 45 degree angle between the slow axis of the MEMS mirror and the incident light beam [26] and will be corrected in future image post processing.

 

Fig. 3 (a) Transmission image in 720 nm light of the smallest bars in Group 7, element 6 on a USAF1951 target obtained with the miniature multimodal microscope. Part of the number ‘6’ can be seen in the top left side corner of the image. (b) TPEF image of a sample of 1 µm fluorescent microspheres on a thin #1 coverslip (c) CARS image of 20 µm polystyrene beads

Download Full Size | PDF

We next performed TPEF microscopy on a sample of diluted solution of 1 µm fluorescent microspheres (Polysciences Inc., PA, USA). Light from the femtosecond laser at 800nm was reflected off the MEMS with close to 80% reflectivity and ~64% of this light was transmitted through the optics inside the barrel, such that an average power of ~28 mW is focused at the sample. The reconstructed TPEF image is shown in Fig. 3(b). Most of the 1 µm spheres are seen to coalesce together, however a few individual spheres can be clearly resolved. The intensity profile is plotted for such individual 1 µm fluorescent spheres that are in focus across the FOV. The average value of the full width at half-maximum of the Gaussian curve fitted to the intensity profile gives the value of the lateral resolution and this is determined to be ~1.3 µm for our benchtop multimodal miniature microscope. This is in good agreement with the design value of 1 µm.

For CARS imaging experiments, a small drop of diluted solution of 20 µm and 4.5 µm polystyrene beads on a # 1 cover slip was used. The pump beam at 800 nm and the Stokes beam containing the near IR filtered output at 1057 nm are focused into the volume of beads, so that the aromatic CH vibration in polystyrene at 3045 cm⁻¹ Raman shift gets resonantly excited. Figure 3(c) is the CARS image of the 20 µm spheres obtained with the miniature microscope when the average power at the sample was ~28 mW in the pump beam and ~0.8 mW in the Stokes beam. From this image it is seen that the FOV is ~70 x 70 µm. The slightly decreasing intensity to the right of the FOV in Fig. 3(c) is because of slight beam clipping owing to the non-perfect alignment of the barrel with respect to the MEMS mirror on the vertical rail. The axial resolution of the miniature microscope was experimentally measured by CARS imaging of 4.5 µm polystyrene beads (Polysciences Inc.) in steps of 1 µm along the “z” optical axis. The maximum intensity values of the line profiles in the z stack were plotted as a function of z step. A full width at half-maximum value of the Gaussian curve fitted to this plot resulted in an axial resolution of 12.74 µm for the miniature CARS microscope. This is larger than the design value of 3 µm and is mainly attributed to residual chromatic aberration inside the miniature objective [17] as well as slight optical misalignment in the beam paths of the bench top microscope system.

3.2 Multimodal imaging of biological tissue

We next demonstrate multimodal ex vivo imaging of biological tissue samples with the miniature microscope. A 0.5 mm thin section of a fixed dorsal root from a YFP mouse (Jackson Laboratory, Bar Harbor, Maine) was mounted on a slide, covered with a thin cover slip and imaged facing down for TPEF and CARS, as described above. In this sample, the axons selectively express yellow fluorescent protein (peak emission ≈530 nm) whereas the lipid rich myelin surrounding the axons is label-free. The frequency difference between the pump and Stokes light sets up a coherent vibration of the CH bonds at 2845 cm⁻¹ Raman shift in the lipid molecules of myelin. A 65 nm bandpass filter centered at 645 nm (Chroma Technology) is used in the collection beam path to selectively pass only the CARS signal. Proof of principle forward CARS image of unlabeled myelin surrounding the axons is seen in Fig. 4(a) . The bright strands of lipid-rich myelin, marked by an arrow, can be clearly distinguished in Fig. 4(a). When the Stokes beam was blocked, no appreciable signal was detected. This confirms that the contribution due to YFP emission at 645 nm is negligible and that the image in Fig. 4(a) is primarily due to CARS emission.

 

Fig. 4 Multimodal imaging using miniature CARS microscope (a) forward - CARS image at a Raman shift of 2845 cm⁻¹ from label free myelin surrounding YFP labeled axons in fixed dorsal roots of mouse. (b) TPEF image of YFP labeled axons at excitation wavelength of 870 nm. (c) SHG image from unlabeled rat tail collagen.

Download Full Size | PDF

The excitation wavelength for optimal two photon absorption in YFP is known to be ~970 nm [28]. However, our laser could only be tuned to ~870 nm where mode locking was still possible. The TPEF image of the same sample of fixed YFP mouse dorsal root as in Fig. 4(a) was obtained at 870 nm and is shown in Fig. 4(b). Although the signal to noise ratio is poor, the YFP labeled axons in Fig. 4(b) can be identified. The laser was tuned back to 800 nm for SHG imaging and a short pass filter (Chroma Technology, VT, USA) was used in the collection beam path to selectively pass only wavelengths below 450 nm. Figure 4(c) illustrates the SHG image obtained from a 0.5 mm thin section of fixed rat tail collagen. The wavy type-I collagen fibers are well resolved in the image.

4. Discussion and conclusion

We have demonstrated a novel miniature multimodal microscope capable of performing CARS, TPEF and SHG imaging. The excitation light is scanned in a Lissajous pattern by means of a two dimensional scanning MEMS mirror that is 500 µm in diameter and is focused on the sample by a miniaturized probe containing miniature relay optics and a multiple lens objective that is 1.8 mm in diameter. The miniature objective corrected for chromatic aberration is able to generate a strong CARS signal corresponding to the vibrations of the CH bonds at 2845 cm⁻¹ and 3045 cm⁻¹ Raman shifts. Proof of principle images of fluorescent and polystyrene beads as well as biological tissue obtained with our setup demonstrate very high resolution and the shapes of features remain consistent throughout the FOV.

Our current setup serves as a test bench for a future portable, handheld prototype of a miniature multimodal CARS microscope for in vivo imaging. In fact, integration of the components is underway in which the excitation light is delivered by a large mode area PCF [16] and the nonlinear optical signal from the sample is collected by the same miniature optics in the epi-direction, separated from the excitation light by a miniature dichroic mirror and collected by means of a large core multimode fiber. A frame rate of 4 Hz is obtained for the particular values of resonant frequencies chosen for the slow and fast axes in this paper. For imaging dynamic phenomena, or for positioning the sample in the FOV, we plan to generate video frames. The video frame rate will be achieved by using the “sliding” Lissajous pattern technique as described in Ref. 12. This technique uses the fact that unlike raster scan, the Lissajous pattern scans the whole FOV several times in each frame, and by marking the start of a frame at a different “sliding” spot, faster video rate can be achieved without compromising the spatial coverage. For situations where not enough nonlinear optical events are present, the image will be enhanced by brightening the image and/or reducing the image size (256x256 instead of 512x512).

The axial resolution for CARS imaging is expected to improve and get closer to the design specification of 3 µm due to better optical alignment and collinearity of the fiber delivered pump and Stokes beams in this prototype. It is also notable that the light source for our miniature multimodal CARS microscope is based upon a single femtosecond Ti:sapphire laser and the use of a PCF for Stokes generation. A femtosecond fiber laser such as in Ref. [29] could be used to replace the tabletop Ti:sapphire laser making the entire setup compact and amenable to translation to the bedside.