1. Introduction

Nonlinear optical microendoscopes (NOMEs) have emerged as powerful devices in two-photon fluorescence (TPF) and second-harmonic-generation (SHG) imaging, enabling high resolution, inherent optical sectioning, and label-free imaging with low tissue damage [1–8], and have shown huge potential in neuroscience [2] and clinical pathology [4]. Microendoscopy typically requires a flexible probe with a millimeter-order outer diameter for forward imaging, which is used to the area in front of the probe during observation of internal organs or freely moving animals. In NOME design, there are two major technical challenges. First, a flexible fiber must deliver femtosecond pulses with minimal distortion and collect backwards-travelling fluorescence with high efficiency. Second, a miniature scanning mechanism is necessary to form images and to enable the probe to be compact [9].

A fiber bundle was previously developed for NOME [10]. However, a single fiber is preferable since it eliminates major defects that a fiber bundle exhibits, including pixilation artifacts and optical cross-talk [11]. Microelectromechanical system (MEMS) mirrors have been developed for as miniature scanning mechanisms in NOMEs [10, 12–14]. The MEMS mirror alignment typically entails complicated alignment procedures and their fabrication requires much expertise, so they have not become widely adopted. Thus, instead of MEMS, deflecting the fiber tip by a piezoelectric mechanism appears more attractive than those using MEMSs. Recently, by using different amplification approaches (i.e., resonance [1, 4, 5], non-resonance [15, 16], or both [3]), piezoelectric mechanisms have driven a variety of optical fibers for NOMEs, such as single-mode fibers [1, 15], photonic crystal fibers [5, 17], double-clad fibers (DCFs) [4, 18], and double-clad photonic crystal fibers (DCPCFs) [8].

Recent innovations in micro-structured fibers offer additional features beyond those of conventional step-index fibers. In particular, DCFs and DCPCFs with large diameters benefit NOMEs greatly by enabling the dual functions of single-mode illumination and efficient signal collection. To provide an appropriate field of view within a specimen, it is important to achieve an adequate amplitude of the vibrating fiber cantilever, which is easy to be implemented for fibers with an ordinary diameter of 125 μm. However, the large diameter of DCFs and DCPCFs brings a challenge to driving capability due to higher stiffness, which results in a poor amplitude. It is critical to investigate the essence of driving fiber cantilever for an adequate amplitude by a piezoelectric actuator. In recent works, it turned out that the amplitude is severely affected by damping, including solid damp, aero-damping of the fiber cantilever and damping caused by glue at the joint point between fiber cantilever and actuator [19]. Usually, a weakly damped system is preferred for a larger fiber tip amplitude. For instance,a custom-designed 266-μm-diameter DCPCF has been driven and a very high mechanical quality factor has been achieved by using a minimal amount of glue to reduce damping [8].

However, asymmetry and environment perturbation are two inevitable problems in fabrication and practical animal imaging, resulting in imaging distortion or even high risk of fiber cantilever fracture. Since the base excitation of the fiber cantilever is provided by the movement of the actuator’s free end [19], the piezoelectric actuator should be considered as an energy conversion system with energy dissipation, converting the input electric energy into mechanical energy for vibrating fiber cantilever. It is reasonable and necessary to improve the structure of the actuator for lower energy dissipation, which reserves more energy for the fiber cantilever.

Here we presented an optimized actuator for driving a particular fiber cantilever in the view point of energy, which contributes to driving large diameter specialty fibers due to low energy dissipation. A DCPCF (DC-165-16-Passive, NKT Photonics A/S) with a large diameter of 350 μm is used for effective illumination and detection. Since this fiber enables reduced self-phase modulation due to the large-mode-area core [20], there is no need to use a complicated pulse shaper. In section 2, we investigated the features required for a particular fiber to vibrate with a sufficient amplitude, demonstrating that a large-diameter fiber needs more energy to vibrate. In order to drive large-diameter fiber with limited input power, we investigate and improve power dissipation of the piezoelectric elements. Strain and stress analyses revealed that the four-plate actuator enabled lower energy dissipation and higher driving capability. In section 3, a NOME was implemented based on the optimized four-plate actuator and enabled the large-diameter DCPCF to achieve an adequate amplitude. When applied to in SHG imaging of a rat tail tendon and ex vivo TPF imaging of a mouse lung, a mouse colon, and green-fluorescent-protein (GFP)-tagged mouse brain slices, our NOME exhibited an excellent signal-to-noise ratio and satisfactory scanning properties, suggesting that it has potential applications in neuroscience and clinical diagnoses.

2. Scanner design

2.1 Deformation energy of scanning fiber cantilever

To clarify the key factors necessary to achieve an adequate displacement, we firstly discuss the deformation energy of a fiber cantilever [21]. For a bending fiber cantilever with a fixed end, the deformation energy U is given by

The differential equation describing the deflection curve is $\frac{d^{2}v}{d{x}^2}=\frac{M(x)}{EI}$, where $v(x)$is the deflection. Thus,

At the nth resonance, the deformation profile of the cantilever is

For each resonant order n, the harmonic value ${\mu }_n$ is a constant. All of the harmonic values can be determined by the equation $\cos \mu \cosh \mu =1$. The first four ${\mu }_n$ values are 1.875, 4.694, 7.855, and 10.996.

We define a constant A as the amplitude and let $v(x)=A\Psi (x)$. For the nth resonance, $v(x)=A\Psi (x)$, where $t=x/L$. Specifically,

Thus, U can be written as

As shown in Eq. (5), U is proportional to E, the fourth power of d, the square of A, and L. For the certain resonant mode, ${\Psi }_n(t)$ is known. Therefore, the expression ${{\displaystyle {\int }_0^1\left[{\Psi }_n^{\text{'}\text{'}}\left(t\right)\right]}}^{2}dt$ is a definite value for each value of n. In fact, ${{\displaystyle {\int }_0^1\left[{\Psi }_n^{\text{'}\text{'}}\left(t\right)\right]}}^{2}dt$ identically equals 1 m⁻⁴, independent of n.

For a particular fiber, E and d can be determined. In nonlinear optical imaging, the scanning mechanism should offer an adequate field of view (FOV) of about 200 μm in diameter on the sample, which corresponds to a fiber cantilever tip displacement of about 1 mm considering the gradient index (GRIN) lens magnification of 4.8. In our previous work [22], we demonstrated that L is the primary factor influencing the resonant frequency in scanner design.

In addition, we address the deformation energy of two different scanning fiber cantilevers (a conventional SMF cantilever and a DCPCF cantilever). Since SMFs and DCPCFs are both made from SiO₂, we assume that they have the same E and consider the situation in which both fibers have L = 20 mm, to achieve appropriate resonance frequencies, and A = 1 mm, to achieve adequate FOV. According to (5), since E, A, and L are all constant, $U\propto d^4$. More specifically, a 20-mm-long conventional SMF (with a diameter of 125 μm) only needs 5.85 fJ of deformation energy to achieve A = 1 mm, while a DCPCF (with a diameter of 350 μm) needs 359 fJ, 61.5 times that of a conventional SMF, to achieve the same A.

To meet the increased requirement of deformation energy, the most obvious way is to increase the input power directly. For piezoelectric elements, elevating the driving voltage can increase the input power. However, the coercive fields of piezoelectric ceramics limit the maximum driving voltage, because such materials typically depolarize if the coercive field is above 1 kV/mm. Unfortunately, the driving voltages of current piezoelectric fiber scanners using conventional fibers have already essentially reached this limiting value.

Since the input power cannot be increased further, it is necessary to decrease the power dissipation. Some progress have been made by reducing the damping of the fiber cantilever and the attachment with the use of very little glue. Here, we present a different method, aiming at decreasing the power dissipation of the piezoelectric elements, which has no conflict with damping reduction of the fiber and the attachment. In the deformation process, because of the molecular motions during displacement, some regions of the piezoelectric elements become hotter, while other regions become cooler. These internal energy changes break the thermal equilibrium between the various parts of the vibrating structure, causing heat flow to occur and energy to be dissipated [23]. The conventional piezoelectric monolithic tubes used in two-dimensional scanning are integrated tubes with quartered outer electrodes [24]. Although these piezoelectric tubes are ultra-compact, their monolithic structures result in high mechanical damping. The high energy dissipation prevents the resonator from reaching a high elastic potential. Consequently, conventional piezoelectric actuators cannot enable DCPCF to achieve amplitudes adequate for tissue imaging.

2.2 Improved piezoelectric actuator

DCPCF needs requires much more deformation energy than conventional piezoelectric monolithic tube can provide. On one hand, the input power is limited due to depolarization of the piezoelectric elements; on the other hand, conventional piezoelectric monolithic tube suffers from severe power dissipation. Since the input power cannot be increased further, it is necessary to decrease the power dissipation. Conventional piezoelectric monolithic tube with quartered outer electrodes dissipates energy in two ways: thermal diffusion between adjacent electrodes and molecular bonding between adjacent interfaces. To reduce these two types of diffusion, we investigated a four-plate actuator as an alternative to a four-quadrant monolithic tube. Four separate plates are used in this design so that the thermal diffusion between electrodes can be controlled, and the electrodes are placed on the individual piezoelectric plates so that no molecular bonds, which would require energy in order to be broken, occur between adjacent interfaces.

Finite-element simulations were performed to investigate the energy dissipation of both the conventional tube and four-plate actuators. According to the resulting strain profiles shown in Fig. 1, the strain distributions in the coterminous areas of the adjacent plates are not identical. This strain inequality indicates temperature fluctuation, which leads to thermal diffusion and increases the entropy of the system [23]. If the flanks of the electrodes are abutted and integrated into the adjacent electrodes as in the monolithic tube, the ceramic will act as a thermal conductor. In contrast, the four-plate design reduces the heat flow between adjacent electrodes since the heat can hardly be transmitted through the air, because the thermal conductivity of air (0.03) is negligible compared with that of the ceramic (1.5). By controlling the thermal diffusion in this manner, we effectively eliminated one of the energy dissipation mechanisms.

 

Fig. 1 Calculated strain profile of four-plate actuator. Strain distribution implies temperature fluxion. Inset showing Areas A and B depicts lengths and thicknesses on different scale for clarification.

Download Full Size | PDF

For a monolithic tube, adjacent electrodes influence each other via the adjacent interfaces. Moreover, according to the stress profile of a monolithic tube shown in Fig. 2, the interfaces between electrodes are sometimes the areas in which stress is concentrated; thus, material fatigue may arise in these areas and potentially lead to structural damage. By comparison, since the electrodes in the four-plate regime are separated, no molecular bonds between adjacent interfaces, which would require energy in order to be broken, can be formed.

 

Fig. 2 Calculated stress profile of monolithic tube. Stress distributions in interfaces between electrodes implies that electrodes hamper each other.

Download Full Size | PDF

In other words, according to the stress and strain analyses performed using the finite-element model, the four-plate actuator, upgraded from the monolithic tube, relieves internal friction in two ways. First, the separation of the plates prevents heat from flowing between adjacent electrodes, thus eliminating thermoelastic dissipation. Second, since the electrodes are separated, no molecular bonds can be formed between adjacent interfaces, so no energy is dissipated to break such bonds. Compared with the monolithic tube, the four-plate design significantly reduces both the internal friction and the dissipation of mechanical energy. Due to its low energy dissipation, this four-plate actuator has a high driving capability.

2.3 DCPCF driving using four-plate actuator

Both scanning fiber cantilever and piezoelectric actuators have been discussed above. As mentioned previously, DCPCFs require much more deformation energy than conventional fibers do, thus, we developed a four-plate actuator design that exhibits negligible thermoelastic dissipation and molecular bonding. Considering the combination of a scanning fiber cantilever and a piezoelectric actuator in an integrated fiber scanner, four arrangements were compared to investigate the feasibility of driving a DCPCF with a four-plate actuator. As shown in Fig. 3, a four-plate actuator or a monolithic tube was used to drive a DCPCF or a conventional SMF. In the finite-element simulation, the coupled field models had the actual complexities and properties of the components composed of various materials (i.e., the piezoelectric elements, silica fiber, and conducting resin). The diameters of the DCPCF and the conventional fiber were 350 μm and 125 μm, respectively. The fiber cantilevers were both 20 mm long. The lengths and inner and outer diameters of the tubes were 20 mm, 1.5 mm, and 2 mm, respectively, and the plates had lengths, widths, and thicknesses of 20 mm, 1 mm, and 0.25 mm, respectively. Thus, the tubes and plates used in the piezoelectric drivers had the same thicknesses and similar circumradii. A driving voltage of 100 V_pp was used for both types of drivers in the simulation.

 

Fig. 3 Calculated frequency responses of scanners under driving voltage of 100 Vpp near first-order resonant peaks. Insets: Finite-element models for piezoelectric scanners using tube or four plates to drive SMF or DCPCF.

Download Full Size | PDF

We compared the harmonic responses of the DCPCFs and the conventional fibers driven by the different piezoelectric elements (i.e., a single tube or four plates). The four different assembles are displayed in the insets in Fig. 3. Considering the first-order resonance, the conventional fiber driven by a single tube exhibits a resonant peak at 289.0 Hz, and the amplitude of the distal tip is 3 mm; by comparison, the conventional fiber driven by four plates achieves an amplitude of 12.4 mm at 288.8 Hz. More importantly, the DCPCF actuated by the tube to vibrate at the first resonant peak (i.e., 700 Hz) barely reaches an amplitude of 129 μm, which indicates its ineffectiveness for biological imaging. In sharp contrast, the DCPCF driven by four plates exhibits a peak amplitude of 565 μm at 691 Hz, which is sufficient to provide an adequate FOV.

3. Experimental results

3.1 Fabricated DCPCF scanner and prototype probe

The piezoelectric elements' geometries significantly influence the vibration amplitude. According to the above analysis and previous work [22], we deduced that longer, narrower, and thinner piezoelectric elements can enable larger-amplitude fiber vibrations. To investigate these effects further, we fabricated several piezoelectric scanners with various geometries in this study. Specifically, piezoelectric elements 0.3 mm thick, 1.2 mm or 2 mm wide, and between 5 mm and 20 mm long were tested, respectively, as shown in Fig. 4, and a 23-mm-long DCPCF cantilever was used since it exhibited sufficient vibrations at the desired resonance frequency. Compactness and robustness should also be considered in scanner design, which is a trade-off.

 

Fig. 4 Measured vibration amplitudes of DCPCF cantilevers with different sizes of piezoelectric elements. W: width, T: thickness, PZT: piezoelectric ceramic transducer.

Download Full Size | PDF

In this prototype scanner, each plate (PZT-5, Aerospace Piexoelec) had a length, width, and thickness of 16 mm, 1.2 mm, and 0.3 mm, respectively, and the DCPCF cantilever was 23 mm long. The resonance characteristics were measured by imaging the oscillation pattern of a fiber tip by using a position-sensitive detector (PSM 2-10, On-Trak). Scanning traces were also recorded for image reconstruction. At the driving voltage of 100 V_pp, the scanning-range peaks in the two orthogonal directions (Directions 1 and 2) occur at 518 Hz and 520 Hz, which correspond to scanning ranges of 1.20 mm and 1.09 mm, respectively, as shown in Fig. 5(a). A spiral scanning pattern could be achieved by adjusting two amplitude-modulation sine wave signals with a 90° phase difference. When the driving voltage gradually increases from 0 V_pp to 100 V_pp, the scanning ranges in both directions increase linearly, as shown in Fig. 5(b). According to the measured values, the scanner’s properties are sufficient for tissue imaging.

 

Fig. 5 Scanner properties: (a) frequency responses around resonant peaks in two orthogonal directions, and (b) scanning range versus driving voltage at resonant frequencies.

Download Full Size | PDF

We designed a miniature probe based on the scanner, for which a schematic and a photograph are shown in Figs. 6(a) and 6(b), respectively. The compact probe consists of a DCPCF piezoelectric scanner, a GRIN lens, and the housing. The scanner has a resonant peak near 520 Hz. This DCPCF delivers the femtosecond excitation pulses into the single-mode core and collects the nonlinear optical signals within the multimode inner cladding. The GRIN lens (GT-MO-080-018-810, GRIN Tech) is fixed at the end of the probe, focusing the laser into the sample and gathering the emitting signals. In general, GRIN lens benefit from their small diameter but suffer from off-axis aberrations and chromatic aberrations. In our experiments, it is found that the periphery part of an image was slightly ambiguous compared with the central part due to off-axis aberrations. The focal shift caused by chromatic aberrations usually results in severe reduction of collection efficiency due to small collection area in the fiber. However, taking advantage of our large-diameter fiber, a large collection area of 165 μm helps alleviate this problem. The housing is made of stainless steel tubing (Grade 304 Stainless Steel) and holds and protects the scanner and the GRIN lens. The rigid portion of the probe is 53 mm long and 3.5 mm in diameter.

 

Fig. 6 Probe based on piezoelectric fiber scanner: (a) schematic of probe, showing scanner and GRIN lens fixed in steel housing, and (b) photograph of probe.

Download Full Size | PDF

3.2 NOME setup and characteristics

We also developed a nonlinear optical imaging system based on the probe shown in Fig. 7. In this system, femtosecond pulses are generated by a Ti:Sapphire laser (MaiTai eHP DeepSee, Spectra-Physics), and the optical density is modified by a neutral density filter (ND filter, 50Q04AV.2, Newport). The laser has a built-in prechirp unit providing group velocity dispersion ranging from −8900 fs² to −24500 fs² at 800 nm, which can compensate for the positive dispersions from the DCPCF and the other optical components together with a pair of gold coated reflective diffraction gratings (GR25-1208, 1200 lines/mm, Thorlabs). After passing through the dichroic mirror (DM, FF665-Di02-25 × 36, Semrock), the pre-chirped laser pulse is coupled into the DCPCF by an objective (20 × / 0.65 NA, Olympus) and is then guided to the probe end for illumination. The emitting signal is collected by the GRIN lens and is then focused back into the inner cladding of the DCPCF. Different bandpass filters (BP filter, Semrock FF01-485/70-25 for intrinsic TPF, Semrock FF01-395/10-25 for SHG, Semrock FF01-525/40-25 for GFP) are used before the photomultiplier tube (PMT, Photosensor Modules H7422-40, Hamamatsu) to detect various signals.

 

Fig. 7 Layout of fiber-optic NOME.

Download Full Size | PDF

The illumination light maintains a short pulse duration after passing through the optical system, due to the reduced nonlinear effects. Figure 8 shows the pulse widths measured by a frequency-resolved optical grating (FROG, Swamp Optics). By compensating for the positive group velocity dispersion of the DCPCF with the negative dispersion provided by the laser’s built-in prechirp unit and pair of gratings, a pulse width of 75 fs could be achieved with a power of 30 mW in the DCPCF core at a wavelength of 800 nm. Femtosecond pulses delivered through the fiber core suffered less from SPM because of the large, 16-μm-diameter, single-mode core. When the power in the fiber core is 60 mW, the pulse widths of 110 fs and 85 fs at wavelengths of 800 nm and 900 nm, respectively, are still sufficient for endoscopic imaging.

 

Fig. 8 Second-order intensity autocorrelation curves of laser pulses in DCPCF core with different powers and different wavelengths.

Download Full Size | PDF

3.3 Optical resolution and imaging results

We experimentally investigated the imaging performance of the nonlinear micro-endoscope. At a driving voltage of 100 V_pp and a circular scanning frequency of 520 Hz, each spiral frame consisted of 500 ms of imaging scanning and 300 ms of retracing and settling, which indicating a frame rate of 1.25 fps. Higher frame rate can be achieved by using motion braking method to reduce non-imaging period. The excitation wavelength was 800 nm. To evaluate the spatial resolution, the intensity profiles of the 200-nm-diameter fluorescence beads (F-8811, FluoSpheres), whose two-photon images are presented in Fig. 9(a), were measured in the lateral and axial directions. Figure 9(b) shows the two-photon images of 10-μm-diameter fluorescence beads (FPS-100M4, Spherotech). The optical resolution and FOV are sufficient for biomedical imaging. The full-widths at half-maximum (FWHM) of the Gaussian fits to the fluorescence intensities show that our NOME has a lateral resolution of 2.2 μm and an axial resolution of 9.1 μm (Figs. 9(c) and 9(d), respectively). The FOV was determined to be 200 μm at a driving voltage of 100 V_pp.

 

Fig. 9 Two-photon images of (a) 200-nm-diameter and (b) 10-μm-diameter fluorescent beads. Fluorescence intensity profiles (dots) of 200-nm-diameter fluorescent bead in (c) lateral and (d) axial dimensions. Blue traces are Gaussian curves fitted to data points.

Download Full Size | PDF

Figure 10 shows an ex vivo image of a mouse colon. Topical application of acriflavine (0.1%wt/vol) strongly stained only the superficial epithelial cells. The imaging process was performed 10–20 min after topical application. Due to the high NA of the GRIN lens, individual enterocytes are visible, and several tubular glands (crypts) surrounded by enterocytes are presented in the field of view of 200 μm.

 

Fig. 10 TPF image of mouse colon after topical application of acriflavine. Crypts and enterocytes are visible.

Download Full Size | PDF

Staining is not ideal for in vivo clinical imaging diagnoses, but the intrinsic TPF/SHG signals of biological tissues are weak, making them difficult to detect. In our NOME, a DCPCF is used to achieve effective pulse delivery and fluorescence collection. The DCPCF’s high-NA, large-collection-area, multimode inner cladding enables high-efficiency fluorescence collection (the diameter and NA of the inner cladding are 163 μm and 0.65, respectively). We also performed intrinsic TPF/SHG imaging of a mouse lung and a rat tail tendon. The excitation wavelength was 790 nm, and the pulse width was about 75 fs. Semrock FF01-485/70-25 and Semrock FF01-395/10-25 BP filters were used for intrinsic TPF and SHG imaging, respectively. Figure 11(a) shows an unaveraged intrinsic ex vivo TPF image of the mouse lung. Anatomical details such as the lumens and aveolar walls are clearly identifiable. The laser power from fiber core on the sample was ~30 mW. During the entire imaging process, no tissue damage was observed. Figure 11(b) shows an unaveraged SHG image of the collagen fibers taken from the rat tail, in which individual collagen fiber bundles in the tendon structure are clearly visible. There is some slight distortion in the center of the image due to piezotube hysteresis. Since the morphology of a collagen fiber network is closely related to its physiological state, the SHG microendoscope may provide useful information for clinical diagnosis. Due to the advantages of the DCPCF, both the TPF and SHG imaging results depict the internal structures clearly, demonstrating that DCPCFs are promising in future real-time in vivo imaging.

 

Fig. 11 (a) Unaveraged intrinsic TPF image of unstained mouse lung. Lumens and aveolar walls are clearly visible. (b) Representative unaveraged intrinsic SHG image of rat tail tendon.

Download Full Size | PDF

Since our NOME exhibited a sufficient optical resolution, FOV, and great signal lever, we also demonstrated the potential for use in neuroscience. Specifically, we imaged GFP-tagged mouse brain slices using an excitation wavelength of 900 nm, a pulse width of about 85 fs, and an excitation power through fiber core on the sample of about 10 mW. A BP filter (Semrock FF01-525/40-25) was used for GFP collection. The results are shown in Fig. 12, in which individual neuron cell bodies as well as axons and dendrites in the hippocampus are distinguishable. Due to the benefits of the DCPCF, these images exhibit excellent signal-to-noise ratios, which has potential in functional calcium imaging.

 

Fig. 12 TPF images of GFP-tagged neurons in mouse brain slices. Excitation power through fiber core was about 10 mW at 900 nm. Cell bodies, axons, and dendrites are visible.

Download Full Size | PDF

4. Conclusion

In this study, we investigated the energy requirement of a bending fiber cantilever at resonant mode, indicating large-diameter fiber needs more deformation energy to vibrate. Then, we developed an optimized four-plate piezoelectric actuator with low energy dissipation, which contributes to energy accumulation of the fiber cantilever. A forward-imaging NOME based on a 350-μm-diameter DCPCF driven by four-plate piezoelectric actuator has been implemented. The performance characteristics of the NOME system were tested, confirming that it delivered of sub-100-fs excitation pulses without a complicated pulse shaper, and acquired images of a stained mouse colon, an unstained mouse lung, a rat tail tendon, and GFP-tagged mouse brain slices without frame averaging. Especially, images of brain slices show great S/R ratio at the excitation power of 10 mW. Distortion-reduced pulse delivery and high efficiency collection in this NOME are promising to be applied to neuroscience research and medical diagnoses. In addition, this four-plate actuator, in principle, is compatible with all kinds of special fibers, making it helpful and promising not only for NOMEs, but also for other imaging modalities in microendoscopes and microsurgery.