1. Introduction

Endoscopy has revolutionized modern medicine by allowing clinicians to visualize internal organs to screen for diseases such as colorectal cancer [1], lung cancer [2] and Barrett’s esophagus [3]. While endoscopy procedures are much less invasive than exploratory surgeries, they are still typically performed under patient sedation due to the size and limited flexibility of current instruments. The use of optical fibers was proposed as an alternative approach to reduce the size and increase the flexibility of endoscopes, thus potentially allowing safer, faster and cheaper office-based procedures. Imaging with coherent fiber bundles was a first step towards micro-endoscopy [4]. However due to the finite size of individual fibers (core and cladding), images suffer from pixilation artifacts and yield low pixel densities (typically ~30,000 fibers in a 1-mm diameter bundle, each fiber in the bundle corresponding to a pixel) [5]. Single optical fiber endoscopy was recently demonstrated by many groups using scanning mechanisms based on microelectromechanical systems (MEMS) [6], resonant micro scanners [7], and spectral encoding [8]. Single fiber endoscopes allow video rate color imaging [9,10] at high pixel densities (e.g. up to 160 000 pixels in a 1-mm diameter probe) [11], can emulate stereoscopic vision through interferometry [12,13] and can be packaged into thin (e.g. 1.6-mm diameter) [9] and ultrathin (e.g. 350-microns-diameter) [14] instruments for access to an increased number of organs through the bore of a 25-gauge needle.

Reflectance single fiber endoscopy is however subject to speckle noise as the use of lasers and single-mode (SM) fibers results in imaging that is both temporally and spatially coherent. Yelin et al. have shown that the use of a double-clad fiber (DCF) dramatically reduces speckle contrast while improving signal collection [15]. In order to preserve lateral resolution, the SM core of the DCF is used for illumination, while the multi-mode (MM) inner cladding is used for collection of partially incoherent light reflected from the sample. DCF’s have been applied to endoscopy for nonlinear imaging [16], and for multimodal imaging where the core is used for optical coherence tomography and the inner cladding for fluorescence detection [17]. An additional advantage of DCF for endoscopy is the increased depth of field, resulting from the larger collection diameter of this confocal setup [18]. Coupling light in and out of a DCF is typically performed using a free space beam splitter [19] setup resulting in a >6dB loss of the weak SM signal and a >3dB loss in the MM signal, making the system quite vulnerable to misalignment due to mechanical motions.

We herein present a novel achromatic (1265 – 1325 nm) double-clad fiber coupler (DCFC) for single fiber endoscopy obtained by fusing and tapering two commercially available DCF segments. This DCFC was integrated into a spectrally encoded imaging setup based on a wavelength-swept source. Its ability to separate SM signal from MM signal was exploited to combine low-speckle reflectance maps (MM signal) with interferometric depth profiles (SM signal combined with a reference arm in a Mach-Zehnder configuration) to obtain 3D reconstructions, thus emulating stereoscopic vision. The DCFC allows for an all-fiber setup which is practically lossless for the SM interferometric measurements, achieves as high out-coupling ratio than a beam splitter setup for MM signal, requires minimal maintenance and will accelerate the translation towards clinical applications.

2. DCFC design, fabrication and characterization

2.1 Introduction

An ideal DCFC for endoscopy allows SM illumination of the sample from the DCF core mode and signal collection from the core and inner cladding modes independently. This requires a null coupler for the core mode and an achromatic coupler for the inner cladding modes collecting as much light as possible. A 100% transmission of the core mode in the main branch is theoretically feasible. However, assuming a symmetrical design, the maximum theoretical value for transmission by inner cladding modes is 50% in each branch of the coupler.

2.2 Material and methods

As shown in Fig. 1a , a DCFC was made by fusing and tapering two commercially available DCF segments (Nufern, East Granby, CT, SM-9/105/125-20A). This fiber has a 9 µm diameter core, a pure silica inner cladding of 105 µm in diameter and a fluorine-doped outer cladding of lower refractive index of 125 µm in diameter. The core NA (numerical aperture) of the DCF is 0.12 while the inner cladding NA is 0.20. We used a custom computer-controlled fusion and tapering setup consisting of a traveling oxygen-propane micro-torch on a three-axis motorized stage and of two linear stages for stretching.

 

Fig. 1 DCFC characterization. (a) DCFC schematics. Red arrow and illumination cone represent the SM signal. Gray arrow and illumination cones represent the MM signal. The SM and MM injections are represented by splices with SM or MM fibers, respectively. (b) Spectral response of the single-mode signal transmission. (c) Spectral response of the multi-mode signal transmission.

Download Full Size | PDF

The fabrication process starts by stripping the DCF regions to be fused from its polymer coating and cleaning them with acetone. The fibers are pressed together by holding clamps containing V-shaped grooves and inspected with a microscope mounted over the setup. For on-line characterization of the core mode transmission, one DCF is spliced with SMF on both ends connected respectively to a broadband source and to an optical spectrum analyzer. The two DCFs are fused side-by-side with the micro-torch traveling over 4-8 mm along the fibers for approximately 2 minutes. The fused structure is then stretched, at a slightly lower flame temperature, at a rate of 0.1 mm/s, with the micro-torch traveling back-and-forth along a constant 8 mm length. This heating length ensures transition slopes small enough for an adiabatic (lossless) transition of the core mode [20,21]. The tapering process is stopped when our target of 15 dB coupling ratio is reached. The reduction factor 0.3 of the constant region is deduced from the total elongation of 20 mm. The device is packaged on a quartz substrate while still under tension on the setup and then inserted in a stainless steel tube. The inner cladding mode transmission is characterized post-fabrication.

Losses in the fundamental core mode are minimized through an adiabatic transition from the full-diameter fused section to the reduced constant section of the DCFC [21]. As the constant section has a moderate reduction factor of 0.3, coupling between the fundamental modes of each fiber is negligible.

Coupling of the inner cladding modes may be explained using geometrical optics. For reasons of etendue conservation in the down-taper section (i.e. the first transition section), as the cross-section of the coupler diminishes, the propagation angle of each ray increases. Most rays are guided by the outer cladding of the new waveguide consisting of the two fused DCFs. In the up-taper section, rays having a propagation angle smaller than the critical angle resume propagation in the inner cladding. Rays essentially distribute uniformly across the coupler cross-section, resulting in an approximate 50:50 transmission. The high number of modes propagating in the inner cladding favors achromaticity and power equipartition.

2.3 Results

Core signal transmission of the DCFC is monitored during fabrication with a conventional broadband source and an optical spectrum analyzer. Figure 1b shows core mode signal transmission in each branch of the coupler. The average isolation between the cores of the two branches is 15 dB and transmission in the main branch is >95% across the illumination spectrum (1265-1325nm).

Inner cladding signal characterization requires more attention. In order to have repeatable results, care must be taken to excite equally each mode guided by the inner cladding. This can be done by using a low spatial coherence source like a halogen lamp or a diffused wavelength swept laser source. Figure 1c shows the transmission of the inner cladding signal in each branch of the coupler. Inner cladding transmission in each branch of this DCFC ranges between 40 and 46% over the illumination spectrum.

2.4 Discussion

We fabricated an achromatic DCFC with a core transmission of >95% and an average transmission >42% for the inner cladding modes. These results compare favorably with previously reported data obtained from couplers made from two fused-silica SM fibers recoated with lower index polymer using either a side-polishing method (15% transmission through inner cladding modes at 750nm) [22] or a twisting method (30% transmission at 750nm) [17]. DCFC were also made for nonlinear microendoscopy using photonic crystal fibers with the advantage of supporting propagation of ultra short pulses, but with the inconvenient of lower collection of inner cladding light (<3%) [23].

In addition to the increased collection power through the inner cladding of this coupler, we note that the use of a DCFC allows a 6dB gain over the free space beam splitter approach for the core signal, in addition to being much more robust against misalignments and less sensitive to back reflections. However, compared to the free space approach, we hypothesize that losses from higher order modes will be higher since they correspond to more tilted rays and therefore are more prone to leakage into the outer cladding. Consequently, the effective NA of this DCFC for imaging is expected to be slightly lower than that of the inner cladding of the DCF.

3. Application of the DCFC to spectrally encoded endoscopy

3.1 Introduction

The DCFC presented and characterized in the previous section is compatible with many scanning techniques (MEMS, resonant microscanners and spectral encoding) and imaging modalities (reflectance, fluorescence). To demonstrate its potential for imaging, we integrated the DCFC into a spectrally encoded imaging system.

Spectral encoding is a scanning technique in which broadband laser light is dispersed by a grating to perform a rapid 1D-scan of the sample [24] by encoding spatial location with wavelength. Spectral detection allows rapid imaging of large field-of-views without the need for fast actuators within the probe and may include interferometric acquisition to extract sample depth [13]. The second (or slow) scan axis may be performed mechanically with a micromotor [25] or by rotating the probe [26]. This scanning technique can be applied to endoscopy [8], confocal microendoscopy [24] as well as coherent anti-Stokes Raman spectroscopy (CARS) imaging [27] in order to yield high resolution (>1000x1000 pixels) reflectance [26] or fluorescence [28] images. Alternatively, spectral encoding may be performed with a wavelength-swept laser and temporal detection to obtain high frame rate images [29] and simpler detection schemes for fluorescence imaging [30].

In this section, we present a novel DCF spectrally encoded endoscopy (SEE) setup which couples wavelength scanning for rapid imaging, interferometric detection for depth assessment and a DCFC for high throughput, low speckle imaging. Indeed, since the DCFC is a null coupler for the SM channel and allows collection of 42% of the inner-cladding partially coherent light, this coupler can be used to acquire simultaneously low-speckle images of the sample via the inner-cladding light and height profiles via SM interferometric detection. The entire setup is fiber-based which makes it insensitive to misalignments and backreflections.

3.2 Material and methods

Figure 2 shows the DCF SEE experimental setup allowing simultaneous low-speckle imaging and interferometric height reconstruction. The laser source is a polygon-based rapid wavelength-swept laser [31] centered at 1302 nm (−10dB wavelength range: 1257nm-1347nm, instantaneous line width of 0.1 nm) and providing >25mW of average output power at repetition rate of 9.8 kHz. Two achromatic fiber-based couplers (90%:10% and 99%:1%) are used to tap laser light for triggering purposes and interferometric measurement respectively. Triggering is performed with a grating-based filter (not shown) which selects a wavelength and generates an optical pulse at the beginning of the wavelength-swept spectrum. The 99:1 coupler separates light into a sample arm (99%) and a reference arm (1%) which is matched in length and in polarization state (through the drop-in polarization controller) with the sample arm. Light from the sample arm passes through a circulator and is coupled to the core of the DCFC through a simple fusion splice. As the SMF28 (Corning Inc., Corning, NY) fiber and the single mode core of the DCF fiber (Nufern, DCF SM-9/105/125-20A) have similar mode field diameters, single mode transmission is achieved with negligible loss at the splice. Since there is negligible cross-talk between the cores of the two DCF in the DCFC, >95% of the laser light is transmitted to the imaging arm consisting of a collimating lens (NIR coated aspheric lens, Thorlabs, Newtown, NJ C220THE-C, f = 11mm), a galvanometer mounted mirror (Cambridge Technology, Lexington, MA), a transmission grating (Wasatch Photonics, Logan, UT, 1110 ln/mm) and a focusing lens (NIR coated achromatic doublet, f = 50mm (Thorlabs, AC254-050-C) or f = 75mm (Thorlabs, AC254-075-C). A polarization controller was used to optimize transmission through the grating which was used in Littrow configuration for increased diffraction efficiency in the first order.

 

Fig. 2 DCF SEE Imaging Setup. The polygon-based wavelength-swept laser is an external fiber cavity semiconductor optical amplifier based laser which delivers near infrared (NIR) light to the DCFC through a fusion splice (10% of the light is used for triggering purposes and 1% for interferometric detection, tapped from achromatic fiber couplers). The DCFC transmits light in its core for illumination of the sample by the imaging arm which consists of a collimating lens, a galvanometer mounted mirror (slow scan axis – Galvo), a transmission grating (fast scan axis) and an objective lens. Coherently backscattered light is collected by the core of the DCFC and sent to a dual-balanced InGaAs photo-detector through a circulator (C) and a 50:50 achromatic (1250-1350nm) fiber based coupler. Diffuse backscattered light is collected by the inner cladding and sent via the second branch of the DCFC to an InGaAs photo-detector. A rapid digitizer simultaneously acquires single- and multi-mode signals for image processing. The galvanometer is controlled by a separate A/D board (not shown). PC – polarization controllers.

Download Full Size | PDF

Light backscattered from the samples consists of a coherent (or singly backscattered) component and a diffuse (or multiply backscattered) component. The coherent portion is collected by the core of the DCF while a portion of the diffuse component is collected by the inner cladding. The DCFC transmits most of the coherent light back in the original branch while the diffuse component is propagated in the inner cladding and is equally split between both branches. Branch 2 of the coupler contains only the diffuse component and is sent to an InGaAs photodetector (New Focus, 2117-FC) and digitized using a high acquisition rate analog-to-digital converter (A/D) board (Alazar Tech, Pointe-Claire, Quebec, ATS9462 16 bit acquisition, 180MS/sec x 2 channels) to produce a low-speckle reflectance map. Branch 1 of the coupler contains both the coherent signal in its core and half the diffuse signal in its inner cladding. At the splice between the DCFC and the SM fiber, the diffuse component diffracts in the cladding and is ultimately absorbed by the polymer coating. Through a fiber circulator, SM light from the sample is recombined with the reference arm using an achromatic 50:50 coupler and is detected with a dual balanced InGaAs photo-detector (New Focus 1817-FC) to extract fringe data without the DC component. The imaging arm includes a variable delay line (not shown in Fig. 2) to accommodate different imaging arm lengths. Signal from the dual-balanced detector is digitized using the second input of the Alazar board at 180 MHz.

Figure 3 shows typical signals measured from the MM and SM fibers using a wavelength-swept SEE approach. MM data provides low-speckle, high intensity reflectance profiles while SM data from the Mach-Zehnder interferometer provide height information. This depth information is extracted from the frequency of the interferogram which results from the slight wavelength sweep (Δλ) occurring during the pixel dwell time (Δt). Neglecting dispersion over Δλ, and considering that the sample’s height is constant over a pixel, the frequency of the interferogram is proportional to the optical path difference, itself proportional to the height of the sample. The frequency of the interferogram is retrieved by Fourier transforming SM data over (Δt), which is digitized at an acquisition rate 36 times higher than the reflectivity MM data. This technique is a low bandwidth equivalent of optical frequency domain imaging [32] and can only detect the height of one interface, as opposed to other interferometric techniques such as optical coherence tomography detecting multiple interfaces inside a sample. To compensate for the low intensity regions (for example destructive speckle interference present in the single-mode data) the height reconstruction is filtered using a median filter. Final images emulating stereoscopic vision are obtained by mapping the low-speckle MM reflectance data on the 3D profile of the sample.

 

Fig. 3 3D image processing from superposition of multi-mode (MM) and single-mode (SM) data. MM data directly provide the intensity profile of the sample while the SM data is Fourier transformed to extract the sample’s height profile.

Download Full Size | PDF

Image acquisition, processing and display is performed through a custom C ++ platform which acquires 384 lines of 384 pixels per image (at a line rate of 9.85 kHz, corresponding to the laser sweep rate) resulting in an image acquisition rate of 25.5 frames per second. Height reconstruction was post-processed using an algorithm written in Matlab (The Mathwork, Natick, MA). It takes ~45 seconds with our post-processing algorithm in Matlab to reconstruct each height profile with a standard computer. While our current algorithm did not allow the required 384x384 36-points Fourier transforms to be performed at 25 frames per second, a fast FFT algorithm in C ++ , with a standard computer (capable of performing more than 1 million 64-point FFT per second) could retrieve a few 3D profiles each second, while displaying processing-free reflectance images at video rate.

3.3. Results

Figure 4 shows a spectrally encoded image of a rough plastic figurine acquired with our DCFC and the table top imaging system described in Fig. 2. The speckle contrast diminution using a DCF fiber can be clearly seen by comparing Fig. 4a – obtained with the SM channel for illumination and detection at circulator output – and Fig. 4b – obtained with the MM detection channel. Speckle contrast was characterized using a highly diffusive abrasive paper (1 micron grit, images not shown) and calculated as the standard deviation of a region of interest over its mean. Speckle contrasts for SM and MM collection paths were 0.8 ± 0.1 and 0.16 ± 0.03 respectively, resulting in a diminution of speckle contrast of a factor 5. Theoretically a fully developed speckle pattern yields a speckle contrast of 1 and drops as $C=1/\sqrt{N}$ where N is the number of independent speckle patterns or propagation modes [33]. Assuming that the reflected light on the sample conserves its linear polarization, the number of modes propagating through the inner cladding is $N\approx 8{\left(aNA/\lambda \right)}^2$ where a is the radius of the inner cladding, λ is the instantaneous wavelength and $NA$ is the numerical aperture of the fiber, which corresponds to ~520 modes. Assuming a fully developed speckle contrast for the perfectly coherent image, the multimode speckle contrast should drop to 0.04. The discrepancy between theoretical and experimental values may arise from the sample (e.g. not allowing fully developed speckle patterns), the coupler (e.g. the energy of each mode not being equally distributed over each branch) and/or the limited collection aperture (e.g. providing a limited sampling a non-uniform modal density). The hypothesis that higher order modes are not excited or are attenuated is also supported by the fact that, while we observe a 3 to 9 fold increase (depending on sample type) in signal from the inner cladding with respect to the core, this increase is not commensurate to the increase in number of modes. Finally, images obtained from the MM signal have a large depth of field and conserve the same lateral resolution than the SM ones. Indeed, measured values for depth of field, obtained by translating a mirror were 2.7 mm (theoretical value: 1.7 mm) and 4.9 mm (theoretical value: 5.6 mm) for SM and MM signals respectively. The 10%-90% edge response lateral resolution, measured using a US air force resolution grid, was 0.03 mm (theoretical value: 0.02 mm) and 0.03 mm (theoretical value: 0.027 mm) for SM and MM signals respectively. Clinically, this translates into a longer depth of field which allows imaging of samples with a more pronounced height profile, as exemplified in Fig. 4.

 

Fig. 4 Spectrally encoded image of a plastic figurine. (a) SM path reflectance image. (b) MM path reflectance image. (c) Color encoded 3D reconstruction obtained from the SM path interferogram scaled with the MM reflectance map. (d) Picture of the figurine.

Download Full Size | PDF

To show the full potential of combining reflectance map and interferometric height determination with SEE, we obtained a sequence of 3D profiles of a biological sample. Figure 5 shows a movie (Media 1) of a rotating expired wasp obtained from a sequence of 99 simultaneously acquired MM reflectance maps and SM interferometrically determined height profiles. No special sample preparation was required. A closer examination of Fig. 5b reveals two types of height indetermination. In these acquisitions, the zero difference plane of the interferometer was located at the back of the wasp’s head. Because our interferometric setup solely yields the absolute difference between the reference and sample paths, we observe (in the antennas, for example) folding of the depth profile around the zero-delay point. Additionally, when the path difference is longer than the coherence length of the source (~4 mm) the fringe contrast falls below the system’s sensitivity and result in improper frequency measurement.

 

Fig. 5 Still frames extracted from a video sequence (shown online) of a wasp’s head obtained from a sequence of 99 detected MM reflectance signals and SM interferogram signals. (a) MM reflectance signal. (b) MM reflectance signal color encoded with the height obtained from the single mode interferograms (Media 1).

Download Full Size | PDF

3.4 Discussion

We showed that the DCFC applied to spectrally encoded imaging allows simultaneous acquisition of a MM signal having a higher intensity and a lower speckle contrast than the traditional SM collection scheme. Additionally, the filtered SM signal was used to interferometrically retrieve the sample’s height profile, emulating the stereoscopic vision. This quasi lossless collection and combination scheme combines signal-to-noise ratio (SNR) advantages of DCF imaging with robustness of an all-fiber component. The actual bench-size setup could be easily miniaturized, the only modification needed to the SEE probe designed by Yelin et al. [14] being to change the optical fiber to a DCF. When an endoscopic version of this setup is available, the SNR advantage may allow clinicians to rapidly screen for surface irregularities (such as colonic polyps, esophageal varices and ovarian tumors) via miniature endoscopes. This setup could hardly be reproduced without the DCFC as the 6 dB losses in the SM channel (incurred by the free space beam splitting approach) in addition to spurious back reflection from all air-fiber interfaces negatively affect SNR thus preventing rapid and accurate interferometric measurements. Moreover, as the DCFC is an all-fiber solution, its robustness and compactness facilitates migration towards clinical practice.

4. Conclusion

The growing use of DCF for imaging justifies the need for all-fiber couplers that are efficient, compact and robust. We presented here a DCFC that could be used in many single-fiber endoscopes. This all-fiber technology allows excitation and collection via the SM core of the fiber with negligible loss and collection of >42% of the multiply scattered light collected by the inner-cladding to the detection branch of the coupler. Moreover, our DCFC is produced by fusing and tapering commercially available DCF in a manner that is compatible with industry standards to fabricate this device reproducibly and at low cost.

This DCFC combined with a wavelength-swept SEE setup allows video-rate acquisition of speckle free images co-registered with 3D height profiles in order to emulate stereoscopic vision. This technology facilitates the migration of miniature endoscopy to the clinical world by improving image quality and depth perception, thus ultimately allowing visualization of pathologies less invasively and at earlier stages.