1. Introduction

The development of the confocal laser scanning fluorescence microscope [1] and scanning microscopes utilizing multi-photon processes such as two-photon fluorescence [2] represent significant breakthroughs in the microscopic imaging of biological specimens. Due to the optical sectioning capability of these techniques, background-free imaging can be performed in highly scattering media such as biological tissue, which are otherwise difficult to image using wide-field imaging techniques. However, even with these techniques, the useful high-resolution imaging depth in biological tissue is limited to a few hundred micrometers due to the low penetration depth of visible and near-infrared wavelengths and sample-induced aberrations. In vivo optical imaging of biological tissue is, therefore, mostly limited to superficial layers of skin and hollow organs, where endoscopic imaging can be performed. For the diagnosis of diseases that require microscopic examination of tissue located deep within the body, the routine procedure is consequently to perform ex vivo imaging of biopsy samples surgically removed from the region of interest. This procedure is time consuming and can potentially lead to complications, for example, lymphedema, which results from the removal of lymph nodes in the staging of breast cancer [3]. Similarly, in preclinical studies utilizing animal models, the current common practice is to perform ex vivo imaging of tissue samples collected from animals sacrificed at each stage of the study. In addition to the logistical difficulties this procedure poses, the variability between results obtained from different animals is a major limitation [4]. Confocal and multi-photon microendoscopy, which involves the miniaturization of microscope optics to very small aspect ratios (ratio of diameter to length), similar to that of hypodermic needles, is expected to enable in vivo deep-tissue imaging [5,6]. However, the miniaturization of optical systems to needle dimensions presents significant optical design and instrumentation challenges, because it requires the miniaturization of either the imaging optics or the scanning mechanism or both [5].

A widely used scheme for miniaturization of confocal microscopes is based on fiber bundles [7,8]. The key advantage of using a fiber bundle is that it provides for proximal scanning, in which the scanning mechanism is located close to the excitation source (away from the imaging optics and the sample). Thus, only the imaging optics needs to be miniaturized and the conventional galvanometric mirror scanning arrangement can be used. However, the image quality is limited by the pixelated structure of the fiber bundle, and post processing is required to remove artifacts. Other schemes which seek to avoid the image quality limitations inherent in fiber bundles necessitate scanning arrangements located distal to the excitation source (close to the imaging optics), and therefore miniaturization of the scanning mechanism is also required. MEMS-based scanning mirrors [9] and fiber scanning [5,10–12] are typically employed. However, miniaturization of the scanning mechanism to diameters of several hundreds of micrometers is still challenging. For the miniaturization of imaging optics, gradient-index (GRIN) optics are particularly attractive as a cost-effective alternative to the direct miniaturization of conventional microscope objectives [13,14]. Microendoscopes based on GRIN optics with diameters of 350 µm to 1 mm, utilizing different imaging modalities such as two-photon fluorescence [15–20], wide-field fluorescence [16,21] and second-harmonic generation [22], have been reported. A multimodal imaging system combining confocal fluorescence, two-photon fluorescence and second harmonic generation imaging using 1 mm GRIN optics has been reported recently [6]. Here, we demonstrate a GRIN-based microendoscope, with a diameter of 350 µm and a length of 27.2 mm, which is capable of performing high-resolution deep-tissue confocal fluorescence imaging through a 22-gauge hypodermic needle. The 700 nm lateral resolution and 15 mm imaging depth, to the best of our knowledge, are the highest reported for a confocal imaging probe of these dimensions. A novel scanning arrangement utilizing a lensed optical fiber, which is amenable to being integrated into a hand-held probe for needle-based confocal imaging systems, is also presented.

2. Optical design of the microendoscope

The design approach is similar to that of reported GRIN-based microendoscopes utilizing one-photon fluorescence (wide-field) [16] and multi-photon processes [15,16] but achieves a significantly smaller aspect ratio than previously reported designs. The optical design and analysis were performed using Zemax optical design software. The optical layout of the GRIN microendoscope is shown in Fig. 1(a) . It is a doublet consisting of a long (26.55 mm, 1.75 pitch) low-NA GRIN relay lens (0.1 NA) and a short (0.65 mm) high-NA GRIN objective lens (0.5 NA). The diameter of the microendoscope is 350 µm, which allows it to be inserted through a 22-gauge hypodermic needle, thereby enabling deep-tissue imaging at depths up to ~25 mm. As shown in Fig. 1(a), a low-NA input beam is launched into the GRIN relay lens. In the optical design described above, we used an input NA of 0.06. The collimated beam at the distal end of the relay lens is the input for the short, high-NA GRIN objective lens that, in turn, produces a diffraction-limited spot at a working distance of ~60 µm in air (~85 µm in tissue). The object-space NA of the GRIN microendoscope, as determined from the Zemax simulation, is 0.29. By proximal scanning of the input beam over a distance of 294 µm, a 60 µm distal field of view (FOV) can be imaged without vignetting. In principle, the entire back aperture of the GRIN relay lens can be scanned to achieve a FOV of ~70 µm. However, beyond a circular region of 294 µm in diameter, the intensity reaching the image plane will gradually decrease due to vignetting. The rays originating from the on-axis field point are shown in blue and those corresponding to the two off-axis field points are shown in green and red. The spot diagrams corresponding to the on-axis field point (blue) and an off-axis field point (red), together with the Airy ring, are shown in Fig. 1(b). The close-to-diffraction-limited performance is apparent from the spot diagrams, which are confined within the Airy ring, which is the first minimum of the diffraction pattern [23]. As described in the introduction, distal scanning mechanisms are difficult to miniaturize to fit into the inner diameter of a needle. However, the use of a long GRIN relay lens to relay the scanning excitation beam to the objective and the fluorescence signal to the back end of the relay lens allows the scanning mechanism to be conveniently positioned outside the needle, avoiding the stringent requirements for ultra-miniaturization.

 

Fig. 1 (a) Optical design of the GRIN microendoscope. Rays with different colors correspond to different field points. The schematic is stretched in the vertical direction for clarity. The working distance is ~60 µm in air, which cannot be resolved on the scale shown. (b) Spot diagrams and the Airy rings corresponding to the on-axis and off-axis image points represented, respectively, by the blue and red rays in (a).

Download Full Size | PDF

The experimental implementation of the above design requires a low-NA scanning beam input to the GRIN relay lens. This is typically achieved using a combination of x-y galvanometric or MEMS mirror scanners and a well-corrected microscope objective with the appropriate NA. We use a novel lensed-fiber scanning arrangement to implement the required input NA. Simple and low-cost graded-index fiber lenses can be fabricated by splicing a short section of multimode GRIN fiber to a single-mode fiber (SMF) [24]. The length of the GRIN fiber required to ensure the proper input NA for the GRIN relay lens can be estimated using paraxial ray-matrix theory for Gaussian beams. Using the output beam from a visible-wavelength SMF (mode-field diameter 3.5 μm) and a gradient parameter, g = 5.7 mm⁻¹, the required length of the GRIN fiber was determined to be 375 µm. The proximal end of the GRIN microendoscope is located at the beam waist of this lensed fiber. The beam waist radius of 3.15 μm corresponds to a divergence angle θ₀ of 0.049 rad. In order to convert the divergence angle of the Gaussian beam into an NA value suitable for comparison with the input NA of the GRIN microendoscope, we use the 5% intensity divergence angle of the beam (NA_5%=1.22θ₀=1.22λ/πw₀). This convention is chosen for two reasons. Firstly, it is the commonly used definition for measuring the NA of fibers. Secondly, in Zemax, the definition of NA is based on the marginal ray angles, and the 5% intensity radius is a sensible measure for defining the marginal ray cone because it encircles 95% of the total beam power. Using this definition of NA, the lensed fiber described above provides the required input NA of 0.06.

A 2D image of a sample is generated by mechanically raster-scanning the lensed fiber across the back end of the GRIN microendoscope, as illustrated in Fig. 2(a) . The distinctive advantage of this approach is that the input beam to the GRIN microendoscope is the same over the entire field of view, thereby eliminating any off-axis aberrations typically introduced by relay lenses used in mirror scanning systems. In general, off-axis aberrations in a scanning optical microscope separately arise from the laser scanning optics and the imaging optics. In a system utilizing mirror scanning arrangements, the relay lenses introduce off-axis aberrations [25]. This is in addition to those introduced by the imaging optics. In the case of the lensed fiber scanning arrangement, off-axis aberrations in the scanning optics are eliminated, as the beam is on-axis until it reaches the imaging system. Thus, for the system described here, the off-axis aberrations arise from the imaging system alone and are small, as illustrated by the spot diagrams in Fig. 1(b). This scanning arrangement is especially suitable for integration into a handheld probe module.

 

Fig. 2 (a) Experimental setup used for the confocal fluorescence imaging experiments using the GRIN microendoscope and the lensed fiber scanning arrangement. BS: dichroic beamsplitter; PMT: photomultiplier tube. (b) The GRIN microendoscope inside a 22-gauge flat-end needle. (c) The microendoscope in situ inside a thick tissue, placed through a guiding needle. The arrows in 2(b) and 2(c) point to the lensed fiber end used for scanning.

Download Full Size | PDF

3. Experimental setup and system characterization

The schematic of the experimental setup used for imaging experiments is shown in Fig. 2(a). The 488-nm excitation light from an Argon ion laser was coupled into a SMF (460 HP, Thorlabs Inc, United States) using a microscope objective. The input end of the fiber was angle-cleaved at approximately 8° to prevent the reflections from the air-fiber interface from reaching the detector. The distal end of the fiber was spliced to a GRIN fiber lens to image the output beam of the SMF onto the GRIN microendoscope with the required NA of 0.06. Scanning of the input beam was achieved by mounting the lensed fiber end on a two-axis translation stage consisting of a motorized stage (Newport Inc., United States) and a piezoelectric stage (Physik Instrumente, Germany), which acted as the slow and fast scanning axes, respectively. The fluorescent signal from the tissue retraces the excitation beam path and passes through the dichroic beam splitter, a filter assembly consisting of a notch filter (NF02-488S-25, Semrock, Inc., United States) and a long-pass filter (522 DF 32, Bio-Rad, United Kingdom) and is detected by a photo-multiplier tube (PMT). The single-mode fiber acts as the confocal pinhole by rejecting out-of-focus fluorescent light [26]. Data acquisition and motion control were implemented using a DAQ card (NI 6115, National Instruments, United States) and LabView software. The GRIN microendoscope design described above was manufactured by GRINTECH GmbH, Germany. To make a mechanically robust imaging probe, the microendoscope was mounted inside a 22-gauge flat-end needle and fixed using optical adhesive (Norland Inc., United States) as shown in Fig. 2(b). Guiding needles were used to guide the probe deep into the tissue and to deliver the fluorescent label as shown in Fig. 2(c).

The GRIN fiber lens was fabricated by splicing a short section of multimode GRIN fiber (GIF 625, Thorlabs Inc., United States) to the end of the SMF using a fusion splicer. Due to the uncertainties in the value of the g-parameter and the deviation of the refractive index from the ideal parabolic profile, the results of the Gaussian beam analysis described in the previous section were only used as a starting value for an experimental optimization of the GRIN fiber length. The optimization was performed by measuring the output of the lensed fiber using a near-field beam profiler (Ophir Optronics, United States). The beam profile data for the lensed fiber used in the experiments is given in Fig. 3 . The length of the GRIN fiber was 362 µm. Figure 3(a) shows the beam profile at the beam waist. The beam waist radius of 3.25 µm corresponds to an NA of 0.058. However, the plot of the beam radius as a function of distance from the GRIN fiber end (Fig. 3(b)) shows an asymmetric variation of the beam radius on either side of the beam waist and a higher NA than the value calculated from the beam waist radius alone. We also note that the beam profiles measured on either side of the beam waist (data not presented) showed a non-Gaussian behavior. Therefore, the beam radii given in Fig. 3(b) should be considered only as an approximate representation of the actual beam propagation. Depending on the process used in the commercial production of GRIN fibers, the deviation of refractive index from the ideal parabolic profile can range from small to significant [24]. We believe that the non-Gaussian propagation observed here is due to the aberrations arising from the deviation of refractive index profile from the ideal parabolic profile, as well as small imperfections arising from the manufacturing process (cleaving and splicing) of the lensed fiber.

 

Fig. 3 (a) Beam profile at the waist of the lensed fiber output beam. (b) Beam radius (1/e², from Gaussian fit) as a function of distance from the lensed fiber end.

Download Full Size | PDF

An input NA higher than the design value of 0.06 can influence the performance of the confocal probe in a number of ways. In the absence of aberrations, it will lead to a narrower excitation point spread function. However, our ray tracing simulations indicate (data not presented) that the increased aberrations resulting from a higher NA input beam are not negligible and will counteract the improvements in the resolution. As described in the following, we experimentally verified that the output beam from the lensed fiber characterized above nonetheless enables sub-micron lateral resolution when launched into the GRIN microendoscope.

The excitation point spread function (PSF) of the GRIN microendoscope in the lateral dimension was obtained by measuring the beam profile at the focal plane. The beam profiles along the x and y directions are shown in Fig. 4(a) and (b) for on-axis and off-axis image points, respectively. A small amount of off-axis aberration is evident from the asymmetric beam profiles in Fig. 4(b), which correspond to the edge of the FOV (~30 µm from the center in the y direction). The full width at half maximum (FWHM) of the on-axis beam profile is ~1 µm. In principle, the PSF of the resulting confocal system is expected to be narrower than the excitation PSF, as it is given by the product of the excitation and detection PSFs. This is consistent with the lateral resolution measurements described below.

 

Fig. 4 Excitation beam profiles (λ=488 nm) measured at the focal plane of the GRIN microendoscope along the x and y directions (as per Fig. 3 (a)) for: (a) on-axis image point; and (b) off-axis image point (~30 μm from center).

Download Full Size | PDF

In order to quantify the lateral resolution of the probe in the confocal fluorescence mode, images of 500 nm fluorescent beads (Invitrogen, United States) were acquired, as shown in Fig. 5(a) . The intensity measured across a single bead is shown in Fig. 5(b). The measured FWHM of 870 nm corresponds to a lateral resolution of approximately 700 nm if we approximate the measured intensity as the convolution of two Gaussian functions of width 500 nm (bead) and 700 nm (confocal PSF). The axial resolution of the imaging probe in the confocal fluorescence mode was measured by scanning a thick fluorescent slide through the focal volume along the optical axis [27]. When the focal volume is completely within the fluorescent slide, the intensity will be a maximum, and by moving the focal volume away from the fluorescent slide, the intensity gradually decreases, as shown in Fig. 5(c). The FWHM of the derivative is used as a measure of the axial resolution of the confocal system. In this case, the measured FWHM was ~11.5 µm.

 

Fig. 5 (a) Image of 500 nm fluorescent beads obtained using the confocal probe. (b) Intensity distribution measured across a single fluorescent bead with a FWHM of 0.87 µm. (c) Axial resolution.

Download Full Size | PDF

4. Demonstration of deep-tissue imaging

The deep-tissue imaging capability of the confocal probe was demonstrated by imaging fluorescently labeled bovine muscle tissue at a depth of approximately 15 mm. The mechanical holder of the needle probe limited the usable length of the needle to ~18 mm, but an optimized mount would readily permit a usable length of about 25 mm for our GRIN microendoscope. 18-gauge guiding needles were inserted to a depth of approximately 15mm inside thick (>3cm) tissue samples. They were used to deliver the fluorescent label (FITC, 250 µg/ml, Sigma Aldrich, United States) to the region of interest and to guide the probe deep into the tissue, as demonstrated in Fig. 2(c). Approximately two hours after the application of fluorescent labels, the labeled region was rinsed with saline solution injected through the same guiding needle. No attempt was made to optimize the labeling procedure. Approximately 150 µW of excitation power at the sample plane was used for the imaging experiments. The image of muscle fibers obtained at a depth of ~15 mm inside the tissue is shown in Fig. 6 (a) . The periodic structures running perpendicular to the individual muscle fiber orientation are sarcomeres, the basic functional unit of contraction in muscle fibers. These fine structures are clearly visible even at the circumference of the image, indicating the high optical performance of the imaging probe throughout the entire field of view. The periodic intensity variation measured along the dashed line in Fig. 6(a) is shown in Fig. 6 (b). The periodicity of the intensity variation is ~2 µm, which is consistent with the earlier reports of sarcomere length in bovine muscle fibers [28].

 

Fig. 6 (a) Confocal fluorescent image obtained at a depth of ~15 mm inside a bovine muscle tissue sample. (b) The intensity variation along the line shown in (a).

Download Full Size | PDF

In order to demonstrate the optical sectioning capability of the probe, a number of images were obtained at different axial locations deep inside the tissue (~15 mm), as shown in Fig. 7 . Images (a) – (f) were obtained by translating the sample in steps of 10 µm along the axial direction. The background-free images of individual muscle fibers clearly demonstrate the axial sectioning capability of the probe.

 

Fig. 7 Images of bovine muscle tissue obtained at a depth of approximately 15 mm. The images correspond to planes located 10 µm apart.

Download Full Size | PDF

5. Discussion

The end face of the GRIN microendoscope was not in contact with the tissue during the acquisition of the images shown in Figs. 6 and 7, but there was always a thin layer of saline between the end face and the imaged tissue. Attempts to image in direct contact with the tissue produced low-intensity images with no visible structure. We believe that this is due to the limited penetration of the fluorescent label below the tissue surface in our staining procedure, but the effect of direct contact with tissue requires further investigation.

Notwithstanding the use of miniaturized optical components that will tend to have more aberrations than conventional microscope objectives, the imaging performance of our system is quite close to the theoretical specifications expected from an ideal confocal fluorescence microscope with the same object-space NA as our GRIN microendoscope. The theoretical FWHM of the transverse and axial PSFs of a confocal fluorescence microscope with a given NA are $FWH{M}_{trans}=0.61\lambda /\sqrt{2}NA$ and $FWH{M}_{axial}=2n\lambda /\sqrt{2}N{A}^2$, where λ is the excitation wavelength and n is the refractive index [29]. For an object-space NA of 0.29 and a wavelength of 488 nm, as used here, we obtain a transverse resolution of 730 nm and an axial resolution of 8.2 μm in air. A rigorous comparison of these theoretical values with the measured data is difficult because of two competing effects that are non-trivial to quantify. The characteristics of the lensed fiber output beam shown in Fig. 3 suggest that the NA in these experiments was higher than the design value, which will tend to improve the resolution. However, the detrimental effect of aberrations will be more pronounced for higher NA values, which will offset the improvement in resolution. Whilst the measured transverse resolution of ~700 nm is consistent with the theoretical estimate, the measured axial resolution of 11.5 μm is clearly degraded. The reason for this could be that, in addition to the monochromatic aberrations, the axial resolution is also affected by chromatic aberration caused by the long section of GRIN relay lens. The resulting chromatic focal shift will cause degradation of the axial resolution and also of the sensitivity of the system due to the difference between the excitation and detection wavelengths in confocal fluorescence mode.

The imaging depth of the microendoscope is primarily determined by the length of the GRIN relay lens. In the current design, we have used a 1.75-pitch GRIN relay lens, which allows penetration depths of about 25 mm into tissue when mounted optimally. In principle, similar probes of even greater length can be designed using an N/4-pitch GRIN relay lens, where N is an odd integer. However, the severity of aberrations will also increase with the length. A ray tracing analysis (data not presented) showed that by using a low-NA GRIN relay lens, such as the 0.1-NA lens used here, aberrations can be kept to a minimum and that close to diffraction-limited transverse resolution can be achieved even with a microendoscope of ~50 mm length (3.25-pitch GRIN relay lens). However, as mentioned above, the increasing chromatic aberrations will degrade the axial resolution and the sensitivity. This is an optical design issue inherent in systems using the same fiber as the excitation and detection pinhole. It can be overcome by using separate excitation and detection fibers or pinholes together with a mirror-scanning mechanism, but the simplicity of the fiber-scanning scheme shown here will be lost.

6. Conclusion

We have demonstrated a hypodermic needle-compatible confocal probe using a 350-μm diameter GRIN microendoscope and a novel, compact and cost-effective scanning arrangement using a lensed optical fiber. A prototype system using the GRIN microendoscope and the lensed fiber scanning principle was used to perform high-resolution confocal imaging of tissue samples at a depth of 15 mm from the surface of the tissue. To the best of our knowledge, the lateral resolution of ~700 nm and the imaging depth of 15 mm is the highest reported so far for a confocal imaging probe of these dimensions. The novel scanning arrangement eliminates off-axis aberrations typically present in relay systems and is suitable for integration into hand-held probe modules for needle-based confocal imaging.