1. Introduction

Cancer diagnosis in early stages is critical for determining an effective treatment plan. Conventionally, surgical biopsies are sectioned and stained to identify the abnormal region. However, a faster and less invasive scheme is desired to make the process more efficient. Multi-photon microscopy (MPM) has been recognized as a promising technique for label-free imaging of biological samples [1,2]. Various nonlinear imaging modalities such as third harmonic generation (THG), second harmonic generation (SHG), two and three photon excited auto-fluorescence (2PEF/3PEF) can provide valuable information about intrinsic properties of the biological samples which can be employed in diagnosis applications [3–6]. However, the key challenge to take advantage of these imaging modalities in gastrointestinal and colonic visualization is to miniaturize the MPM into a compact and lightweight distal probe while keeping the general performance. Interesting multiphoton endo-microscopy systems with a small diameter distal probe have been proposed by different group including Huland et al. [7,8]. However, since these are not flexible, they can not be used for gastrointestinal imaging. A flexible and lightweight endoscope can also be employed for studying neural activities of a freely behaving small animal [9].

To make a compact and flexible endoscope, we need to miniaturize the scanning mechanism, optical system, signal collection assembly and femtosecond laser. The scanning mechanism can be achieved in paraxial end (at the input) by using a fiber bundle [10]. However, this makes the signal collection assembly more complicated and also the image resolution would be limited by the gap between the individual fiber cores in the fiber bundle. Another approach is to use microelectromechanical systems (MEMS) scanning mirrors in the distal end (inside the probe) [11, 12]. However, the endoscopes based on MEMS are relatively bulky due to the restriction on the size of the actuator and its electronic circuit [12]. Resonant-based piezo electric tube (PZT) is an alternative scanning tool. PZTs are commercially available with outer diameter (OD) as small as 1 mm [13,14]. The spiral scanning of a fiber optic can be achieved by appropriately modulating the PZT driving voltage.

The next part that needs to be miniaturized is the focusing optical system. In order to generate nonlinear signal a high intensity excitation is required. This is achieved by focusing the femtosecond laser beam coming out of the optical fiber using a compact high numerical aperture (NA) optical system. In the absence of aberrations, the diffraction limited spot size is MFD/m where MFD stands for mode field diameter of the fiber and m=NA^out/ NAⁱⁿ is the angular magnification of the optical system which is equal to the ratio of the output and input numerical apertures. In most studies a fixed GRIN lens has been employed to obtain the desired spot size. However, the commercial GRIN lenses with NA^out>0.5 have a very short working distance (< 200 μm) and a small field of view (FOV) [15, 16]. They are also designed for a particular wavelength (typically λ < 900nm) and suffer from chromatic aberration. Therefore, a customized optical system is desired to eliminate these limitations.

The typical laser sources for MPM are ultrafast lasers based on Ti:sapphire crystals [13,17]. However, they are not yet suitable for clinical environment due to their bulky size, alignment requirement and high cost. Fiber lasers on the other hand are compact, alignment-free and much less expensive. In our previous study, we have shown that one can perform SHG/THG imaging using a compact ultrafast Er³⁺-doped fiber laser which is mode-locked by carbon nanotube saturable absorber (CNTSA) [3, 18]. Since the laser wavelength is 1.56 μm, the SHG/THG spectrum fall into the region where the optics and detectors perform the best. However, 1.56 μm wavelength is not ideal for deep undersurface imaging of biological tissues due to high water absorption [19]. In [20], Horton et al. employed 1700 nm wavelength in a benchtop multiphoton microscope and successfully tested for imaging subcortical structures within an intact mouse brain with a long penetration depth.

In this paper, we present a compact multi-photon endoscope design which employs a customized optics to increase the working distance (> 700 μm in water) as well as FOV (> 400 μm). The spiral scanning is achieved by resonant-based PZT. This type of scanning suffers from non-uniformity of imaging speed across the FOV. Here, a dynamic sampling is proposed to overcome this issue. The endoscope prototype has an outer diameter of 5 mm and a rigid length of 4 cm. However, the outer diameter can be reduced to 3.5 mm without losing its performance by choosing a thinner PZT which is commercially available. Then, the probe can be inserted into 3.7 mm endoscope accessory port.

The endoscope works at excitation wavelength of 1700 nm where water absorption and tissue scattering drop down. In three photon excitation, this wavelength is equivalent to ≈ 565 nm which can excite a wide selection of fluorescent dyes. Furthermore, a simple scheme is employed to convert a 40 MHz repetition rate ultrafast Er³⁺-doped fiber laser (working at 1560 nm) with 5 nJ pulse energy into a 1700 nm laser with 2.5 nJ pulse energy by soliton self-frequency shift within 2.5 m of single mode fiber. The performance of the system including the resolution and uniformity across the FOV is presented. Finally, the proposed endoscope is tested by performing various ex-vivo biological tissue imaging. To the best of our knowledge, this is the first flexible endoscope which has been successfully employed for THG imaging. Recently, THG microscopy has been used by multiple groups for label-free pathology, optical biopsy and detecting cancerous cells [3–6]. Therefore, development of a flexible endoscope with THG imaging capability will pave the way for future research in this field.

2. Endoscope design and implementation

Figure 1 shows the schematic and final prototype of the proposed endoscope. A 40 MHz, 400 fs pulse width and 1560 nm wavelength Er³⁺-doped fiber laser mode-locked with CNTSA is used as the main source. The pulse dispersion can be controlled by a grating pair designed in the laser. Soliton self-frequency shift (SSFS) is a well-stablished scheme to shift an ultrafast laser source to longer wavelength [20–23]. Here, the wavelength conversion to 1700 nm is performed by propagating the source beam through 2.5 m of standard single mode fiber (SMF 28, Corning). This will convert about 50% of the power to 1700 nm wavelength by SSFS in which shorter wavelength energy will be transferred to longer wavelength through intrapulse stimulated Raman scattering [24,25].

 

Fig. 1 Schematic diagram and final prototype of the compact fiber-based multi-photon endoscope. The spectrum and autocorrelation trace of the shifted soliton are also shown. MLL: mode-locked fiber laser, DM: dichroic mirror, PZT: piezo electric tube, PMT: photomultiplier tube, DAQ: data acquisition board.

Download Full Size | PDF

We coupled 200 mW power of the source which corresponds to 5 nJ pulse energy into the SMF fiber. The SMF fiber output spectrum as well as the autocorrelation trace of the shifted soliton has been shown in Fig. 1. The spectral window at 1700 nm has a full width half maximum (FWHM) bandwidth of 42 nm which is nearly transform-limited and contains 100 mW of total power (2.5 nJ pulse energy). A long pass filter (edge at 1.6 μm) is used to filter out the residual power around 1560 nm. The 1700 nm laser beam is then sent through the core of a double-clad fiber to be focused onto the sample using a customized objective lens (discussed below). The double-clad fiber is 50 cm long and has a core with 20 μm in diameter and NA of 0.11 and a primary cladding with 200 μm in diameter and NA of 0.45. The autocorrelation trace of the endoscope output pulse has been shown in Fig. 1. The pulse broadening is mainly caused by the anomalous dispersion of the double-clad fiber. In principle, we can have shorter laser pulses if we generate the shifted soliton in the double-clad fiber. This will require us to increase the input pulse energy. Alterntatively, we could keep the iput energy the same and make a customized double-clad fiber with a smaller core [26]. Overall throughput of the endoscope is about 50% therefore the average power on the sample is 50 mW which corresponds to 1.25 nJ pulse energy. Since the repetition rate of the laser is high (40 MHz), this pulse energy is enough to obtain high SNR for most nonlinear imaging applications.

The emitted signal is collected through the inner cladding of the fiber. A home-made fused double-clad coupler is fabricated by CO2 laser fusion splicing system (LZM-100 LAZERMaster, AFL) to separate the multimode emission signal propagating backward in the inner cladding from the excitation beam propagating forward in the central core of the fiber [27]. Finally, the emission signal will be divided to SHG and THG/3PEF channels using a dichroic mirror (FF705, Semrock) where THG signal can be separated using a narrow band filter (FF01-580/23, Semrock). The cooled photomultiplier tubes (H7422-40 and H7422-50, Hamamatsu) will then convert the light to electric signals to be amplified using current preamplifiers (SR570, Stanford Research Systems). The amplified signal is sent to a data acquisition board (PCI 6110, National Instrument) for sampling. In the following, we will discuss the scanning mechanism and optical system design.

2.1. Scanning mechanism

PZT actuators are constructed from a piezoelectric tube with four electrodes positioned in quadrants around it. These are commercially available with OD as small as 1 mm which makes them suitable for compact scanning system. For the current endoscope prototype, we use a PZT with outer diameter of 3.2 mm (PT230.94, Physical Instrument GmbH). By driving two orthogonal pairs of quadrants with two sinusoidal waveforms with 90° phase difference and peak to peak voltage of ±40 V, a small circular pattern (≈ 10 μm) can be achieved in non-resonant mode. We modulate the amplitude of the sinusoidal waveforms by a triangular wave to obtain spiral scanning. To increase the scanning diameter (≈ 900 μm), we have passed 10 mm fiber optics cantilever through the PZT and set the frequency of the sinusoidal waveforms to be equal to the resonant frequency of the cantilever (35 Hz). The FOV can be controlled by the triangular wave amplitude. The sinusoidal and triangular waveforms are generated by the data acquisition board. and then amplified using a pair of bipolar piezo drivers (E-413, Physical Instrument GmbH). The number of circles in each spiral scan can be varied between 128 and 1024 depending on the FOV, scanning speed and resolution requirements.

One issue with resonant-based spiral scanning is its non-uniformity. In other words, the scanning speed increases as the probe moves from the center of the spiral toward the edge (simply because the scanning circles in the center are smaller than those at the edge while the scanning time for all circles is the same). Consequently, if one employs a constant sampling rate during the entire scanning time, the density of samples would be much higher around the center of the spiral (Fig. 2(e)). To avoid this issue, a dynamic sampling approach is developed to ensure that the density of samples is uniform across the scanning region. Note that in the proposed scheme, the pixel dwell time remains the same however the number of samples increases as we move toward the edge of the FOV. In other words, a high fixed sampling rate is chosen and then we down-sample dynamically as we go toward the center of the spiral (Fig. 2(f)). Therefore, the effective sampling rate will increase as we scan from the center of FOV toward the edge.

 

Fig. 2 (a) Customized objective design with two aspheric lenses and a fused silica spacer. (b) 3D view of the objective lens attached to the fiber using optical epoxy. (c) On-axis spot diagram generated using Zemax which shows that the spherical aberration is within diffraction limit. (d) Spot diagram of a single ball lens with the same working distance which can be made at the tip of the coreless fiber (for comparison). (e) Image nonuniformity due to constant sampling across the field. (f) Uniform image is achieved by implementing a dynamic sampling.

Download Full Size | PDF

2.2. Optical system design

In all previous endoscope designs, a fixed objective lens is employed to focus the excitation beam and collect the emission signal. For instance, in fiber scanning designs, fiber optics is scanned above a GRIN lens or a multi-element lens system and the beam is focused onto the sample by the lens [28–30]. For multiphoton endoscopes in particular, to generate nonlinear signal, the large mode-field diameter of the fiber needs to be imaged (focused) onto the sample with a transverse magnification of M ≈ 0.2 which results in a spot size of M × MFD which reduces FOV significantly. On the other hand, the FOV of the system is limited by the distance swept by the scanning fiber tip multiplied by the transverse magnification (FOV = D_scan × M). In addition to this, the previous endoscopy systems with a fixed objective lens suffer from coma and astigmatism aberration as well as vignetting for large FOV; therefore, it is not possible to increase the FOV by increasing the scanning fiber deflection. Note that imaging a large field of view in one shot is particularly important in endoscopy because typically there is no precise moving mechanism to acquire multiple images with smaller FOV and stich them togather in contrast with benchtop microscopy systems.

Here, a novel technique is proposed to inherently increase the FOV of the endoscopic systems. We realized that if we design the lens attached to the fiber tip so that the lens moves along with the fiber during spiral scanning, we can increase the field of view to FOV = D_scan by a factor of 1/M ≈ 5. Note that since the fiber deflection angle remains within ±1° for FOV of 400 μm (or ±200), the defocussing due to the tilt remains within the diffraction limit. However, if we define the FOV with FWHM criteria where the signal intensity around the edges is allowed to be as half as the signal at the center, a FOV of about 800 μm can be achieved.

To design an objective lens attached to the fiber tip, we have explored the following schemes. 1) Making a ball lens on tip of the fiber by fusing the fiber tip using a fusion splicer [31, 32]. The spherical ball lenses suffer from a significant spherical aberration and therefore the resulting spot size is limited to around 10 μm which is too large for multiphoton systems. 2) Printing an aspheric lens on tip of the fiber using two-photon direct laser writing systems (e.g. Nanoscribe GmbH) [33]. The quality of the aspheric surfaces printed by this method are not yet comparable with polished glass lenses. 3) Attaching a custom designed objective lens to the fiber tip using optical epoxy. The objective lens can be a single GRIN lens or a multi-element molded glass aspheric lens. However, the GRIN lenses typically have a limited working distance which makes them inappropriate for deep tissue imaging and are usually designed for λ < 900 nm [15,16].

Figures 2(a, b) show the proposed objective lens which is constructed from a coreless fiber (FP200ERT, Thorlabs), a fused silica spacer and a pair of molded glass aspheric collimating and focusing lenses (354430-c and 354140-c, Thorlabs). The coreless fiber is spliced to the double-clad fiber while the fused silica spacer is glued to the coreless fiber from one end and to the collimating lens from the other end using optical epoxy. The fused silica spacer allows the beam to fill the back-aperture of the collimating lens. The collimated beam is then focused by the second lens. This design provides a working distance of 1 mm in water. A housing holds the focusing lens attached to the first lens with a 1 mm airgap. Both aspheric lenses have anti-reflection (AR) coating. The resulting on-axis diffraction limited spot diagram generated by OpticStudio (Zemax LLC) is shown in Fig. 2(c). Note that the solid circle shows the airy disk and blue dots show the real ray intersections with the image plane. For comparison, we have also shown the resulting spot diagram from a single ball lens which provides the same working distance in Fig. 2(d). One can see that by replacing two aspheric lenses with a single ball lens on tip of the coreless fiber the resulting spot size would become 5 times larger. The whole system is mounted in a stainless steel tube with OD of 4.5 mm. A glass cover-slip can be attached to the bottom of the housing to protect the lens when the probe is in-contact with the sample. This protective window will limit the effective working distance to 700 μm in water.

3. Performance characterization

The THG lateral and axial resolutions are obtained by imaging a silicon nano-waveguide and ring resonator chip as shown in Fig. 3. The waveguide width and depth are about 200 nm and 100 nm, respectively. Therefore, the point spread function or lateral resolution of the objective lens can be determined by measuring the width of the waveguide image. Figure 3(b) shows THG intensity across a section of the ring resonator. The full width half maximum of the Gaussian fit is 2.2 μm which corresponds to the lateral resolution of the system. The axial resolution is determined to be 12.7 μm (Fig. 3(c) by measuring the signal FWHM as we move the endoscope in Z direction with 2 μm steps.

 

Fig. 3 (a) THG image of a silicon nano-waveguide and ring resonator with FOV = 200 μm. Since the waveguide width and depth are about 200 nm and 100 nm respectively, it can be used to determine the endoscope resolution. (b) THG signal across a section of the ring resonator. The corresponding Gaussian fit shows a FWHM of 2.2 μm which is considered as the lateral resolution. (c) THG signal from the ring resonator as we move the endoscope in Z direction. The corresponding Gaussian fit shows an axial resolution of 12.7 μm. The scale bar is 25 μm.

Download Full Size | PDF

To measure signal uniformity across the field of view, we acquired THG images (Fig. 4(a, c)) from the surface of a GaAs chip for two different FOV. Figures 4(b, d) show the relative THG intensity across the horizontal (from left to right) and vertical (from bottom to top) lines. The results indicate that the signal remains within 65% on the edges for FOV = 400 μm. However, considering FWHM criteria, FOV is about 800 μm. Note that in theory the signal intensity is expected to remain within 90% in the entire 400 μm field (simply because the spot size remains almost the same). However, the misalignment in assembling the optical and scanning systems results in signal drop down‘ around the top left of the FOV. The stability of the SSFS-based laser during the acquisition time can also be confirmed by the smoothness of the images in Fig. 4(a, c). Since the THG signal has a cubic dependence on the excitation peak power, slight fluctuation of the peak power will result in huge ring-shaped contrast fluctuations in the image as we scan the laser across the FOV. The THG spectrum from GaAs chip was measured (Fig. 4 (e)) by replacing the PMT1 with a spectrometer (Ocean Optics). The spectrum peak is around 567 nm which is correspond to the third harmonic of 1700 nm laser central wavelength.

 

Fig. 4 (a, c) THG image of a GaAs chip surface for two different FOV. (b, d) Relative intensity across the field of view. Considering FWHM criteria, the FOV about 800 μm. (e) GaAs spectrum show a peak at around 567 nm wavelength which corresponds to third harmonic of 1700 nm excitation wavelength. The scale bar is 100 μm.

Download Full Size | PDF

We further explore the endoscope performance by imaging some fixed biological samples (from Amscope). Figures 5(a, b) demonstrate the 3PEF (orange color) image from a thin section of mouse kidney and rabbit lung respectively which are fixed and H&E stained. In Fig. 5(c), human chronic inflammation of colon (#4915, Konus microscope slide set) is imaged which is also fixed and H&E stained. All images in Fig. 5(a, b, c) have a FOV = 600 μm and the images were taken by a frame rate of 15 s/frame at 512 × 512 pixels.

 

Fig. 5 3PEF (orange) from a thin section of (a) stained mouse kidney and (b) rabbit lung. (c) 3PEF from a stained human inflammation of colon. (d) SHG (red) and THG (green) from an indoor plant leaf (unstained). (e) A strong THG signal from a lens cleaning tissue (unstained). (f) 3D view of THG image acquired from a male mosquito eye with 200 μm depth (unstained). The scale bar is 80 μm.

Download Full Size | PDF

The limitation in imaging speed comes from the low resonant frequency (35 Hz) of the scanning system due to heavy objective lens. We were able to decrease the image acquisition time to 7 s/frame by shortening the fiber cantilever (which resulted in a resonant frequency of 75 Hz). In this case the PZT driving voltage needs to be increased to ±75 V to achieve the same FOV (≈ 900 μm). Although we agree that the current acquisition time is not ideal for imaging dynamic activities within biological samples, we believe it can be used for some diagnosis and morphology applications. For instance, in vivo imaging of mouse brain vasculature has been tested by acquisition time from 8 to 20 s/frame in [20]. Note that a lighter and smaller optical system (e.g. with OD of 1mm) can be designed to increase the resonant frequency but it also results in shorter working distance. Therefore, we are trying to use two-photon direct laser writing technology to print the lens on the tip of the fiber to make the objective lens lighter while keeping the working distance about 500 μm.

Label-free imaging capability of the endoscope has been demonstrated in Fig. 5(d, e, f) with a frame rate of 15 s/frame. In Fig. 5(d), we show the THG (green) and SHG (red) signal from an indoor plant leaf. Figure 5(e) shows a strong THG signal from a lens cleaning tissue. The imaging depth is expected to increase because longer excitation wavelength experiences less scattering in biological tissue. To investigate that, a THG image stack (Fig. 5(f)) is acquired from a mosquito eye fixed on a microscope slide underneath a cover slip (Amscope). The image FOV is 900 μm and depth is 200 μm. Although we only demonstrated the imaging depth up to 200 μm, there is no physical limitations in the proposed endoscope working distance or the employed excitation wavelength (in terms of water absorption or tissue scattering) not to be able to increase the imaging depth. Therefore, by optimizing the laser pulse delivery, we believe we can increase the penetration depth to 700μ m in the future prototype.

4. Conclusion

The possibility of nonlinear imaging such as THG/SHG and 3PEF at 1700 nm wavelength with a compact and lightweight endoscope has been demonstrated. The low water absorption and low scattering in this wavelength should make it possible to increase the imaging depth. A customized lens was designed and attached to a double-clad fiber to increase the endoscope working distance (>700 μm) and FOV (>400 μm). The nonuniformity due to piezo tube resonant-based scanning was corrected with dynamic sampling across the field. Finally, the endoscope was used to successfully image biological samples and its performance was characterized. The lateral resolution was determined to be 2.2 μm. In the future prototype, the endoscope OD and rigid length will be reduced to 3.5 mm and 2.5 cm, respectively.