The widespread use of endoscopic imaging fibers is now commonplace in biology and medicine. These fibers allow microscopy, rapid imaging of tissues, and surgical guidance, potentially reducing the need for invasive removal of tissue for diagnosis. Originally developed in the late 1950s [1], current clinical state-of-the-art endoscopic imaging fibers cost several thousands of dollars, need to be sterilized between uses, and have limited use life cycles. It would be highly desirable to develop low-cost imaging fibers that can be disposable after a single use, thereby eliminating the need and costs of repeated sterilization between procedures, reducing effects of degradation and contamination crossover between patients, and significantly improving clinical workflow.

As with most fiber optic applications, the obvious way to reduce costs is to use mass-produced materials shared by the telecommunications industry. The significant challenge is to overcome the performance limitations imposed by the choice of low-cost starting materials to produce a high-quality imaging fiber. The main barrier to using low-cost materials derives from how an image is transmitted along the length of a fiber. In conventional endoscopic imaging systems, an imaging fiber (often referred to as a coherent fiber bundle) is formed of many thousands of cores. Each of the cores in the imaging fiber acts as a pixel to transmit part of an image down the length of the fiber which can be analyzed in real time at the proximal end. If a high-resolution image is desired, it is therefore a requirement that the individual cores of the imaging fiber are as close together as possible. However, in practice the core-to-core spacing is limited by cross-coupling of light between the cores, which will degrade the transmitted image. Cross-coupling increases when the cores of a fiber are placed in close proximity, ultimately limiting the resolution of the fiber. The strength of cross-coupling between cores also depends on a number of other factors: it will become worse if the refractive index contrast (or numerical aperture, NA) of the cores is low, the size of the cores is small compared to the wavelength of the transmitted light, or if the cores are similar in size [2].

In order to suppress coupling in imaging fibers a variety of techniques have been employed, all of which are likely to significantly increase the cost of the fiber. Fibers available from Schott AG [3,4] consist of stacked arrays of uniform cores made from specialty glasses with high index contrasts compared with the cladding glasses. They also offer fibers with absorbing interstitial elements, or leached fiber bundles where the interstitial glass is etched away, leaving a bundle of isolated cores joined at either end of the fiber and separated by air along the fiber length. Fujikura Ltd. [5] produces imaging fibers based on doped silica glasses, where the crosstalk is suppressed by using high-NA ($\sim 0.4$) step index cores with a random variation in size and random spatial distribution [6]. Though this fiber shares the silica glass material system with telecommunications fibers, its NA is on the extreme end of what is possible and it is not straightforward to acquire the raw materials to fabricate such a fiber economically. This results in high fiber manufacturing costs.

In this Letter, we present methods of fabricating high-quality imaging fibers using a single multimode telecommunications preform available from Draka-Prysmian (OM1 PCVD rod). Our techniques involve multi-stacking arrays of different sized cores such that no two adjacent cores are the same size. In our first fiber, we jacket three different sizes of rods drawn from our preform and stack them in a hexagonal array similar to fibers previously reported to reduce cross-coupling in multicore telecommunications fibers [7]. In our second fiber, we demonstrate a simple new technique to achieve low cross-coupling over a broad wavelength range by drawing the preform down to rods of various sizes and stacking these rods in a square array. The distribution of the rods in the square array is such that when it is drawn down it forms a uniform square stack which can then be easily restacked multiple times in order to form an imaging fiber of many thousands of cores. This eliminates the need to jacket and re-draw rods to form different sized cores, making this technique economical and rapid in comparison to our hexagonal fabrication method.

The core material for our first fiber was derived from a commercial preform manufactured for telecoms applications. It had a graded index germanium-doped core surrounded by a thin pure silica jacket. The core-cladding diameter ratio of the preform was 0.74 with a peak refractive index contrast corresponding to an NA of 0.3. The preform was drawn down to rods of three different sizes. Two sizes of the rods were jacketed in pure silica tubes with two different inner-to-outer diameter ratios, and the rest remained unjacketed. The jacketed rods were then drawn down again to form rods, all of which had an outer diameter of 1 mm, but with three different core diameters.

The uniform rods could now be stacked in a hexagonal array of 331 where no two neighboring rods had the same core diameter [see Fig. 1(a)]. The top end of the stack was wrapped in PTFE tape to hold it in place and allow it to be gripped in the chuck of the fiber drawing tower. Smaller sections of PTFE tape were also wrapped around the stack at different points in order to hold it in place. The stack was drawn unjacketed down to canes with each section of PTFE tape removed before it reached the furnace.

 

Fig. 1. Fabrication stages of our hexagonal array imaging fiber. The diagram in (a) shows the layout of our initial stack of 331 rods. Germanium-doped regions are indicated in gray and black (largest cores, light gray; medium cores, dark gray; smallest cores, black) and pure silica regions are white. Note how no two neighboring cores are the same size. The diagram in (b) shows the layout of our second-stage stack where each hexagon represents the initial stack of 331 cores. An SEM image of the final fiber is shown in (c). Doped regions appear lighter in the SEM. The insert in (c) shows a magnified section of the core pattern.

Download Full Size | PDF

To form the final fiber, 37 of the canes were stacked and jacketed in a pure silica tube. This stack was then drawn to fiber with an outer diameter of 525 μm using a vacuum to collapse the interstitial spaces. A vacuum cannot be applied during our earlier drawing stages as our stacks are unjacketed. The early-stage stacks are drawn cold to keep the interstitial holes open until the final vacuum is applied. Too much vacuum processing can also lead to an excessive amount of bubbles in the fiber. This should be avoided as the resulting fiber may be weaker. A scanning electron micrograph of the final fiber can be seen in Fig. 1(c), where the insert shows a higher magnification in order to see the core pattern. The final core diameters in the fiber were 2.78 μm, 2.45 μm, and 2.12 μm with a center to center separation of 3.71 μm. This range of core diameters was chosen as a compromise between pixel density and crosstalk. There were 12,247 cores in total in the final fiber.

Our second imaging fiber was formed from a graded index preform nominally identical to that used to make the first fiber. However, our key innovation was to form a stable stack from rods drawn to different outer diameters, so that there was no need for a jacketing stage. The fabrication method uses $N$ different sized elements stacked in an $N\times N$ array to form a uniform square element. Once the uniform square element has been formed, it is easily stacked multiple times to build up a large array of cores. The stacked squares can easily be drawn down again and restacked in order to easily build up very large arrays.

The example shown in Fig. 2 is a three-stage process. First, a $5\times 5$ array of five different rod sizes (2.23 mm, 2.52 mm, 2.74 mm, 2.95 mm, and 3.17 mm) was stacked in a square stacking jig such that each size only appears in each row or column once. Figure 2(a) shows an end view of this stack. The ends of the stack were fused together using a hydrogen torch and PTFE tape was wrapped around the length at several points in order to hold the stack in place. The stack was fed into the furnace and drawn down to 2.5 mm sided squares, the PTFE tape being unwound before it reached the furnace. This simple process generated a set of square unit elements with a similar cross section to that shown in Fig. 2(a).

 

Fig. 2. Fabrication stages of our square array imaging fiber. A representation of our initial stack is shown in (a), where gray indicates germanium-doped core regions and white indicates the pure silica cladding regions. The second stacking stage of our process is shown in (b). An SEM image of the fiber can be seen in (c), where doped cores appear lighter.

Download Full Size | PDF

The unit squares were restacked with the same orientation in a $6\times 6$ array [Fig. 2(b)] and the ends were again fused and the central region held in place with PTFE tape. (We ensured the correct orientation by cleaving the square elements from the same side as they are drawn. The direction of the cleave edge is then obvious.) This stack was drawn to 4.5 mm sided squares and restacked in a $3\times 3$ array. The final stack was placed into a jacket tube with pure silica packing rods around the outside and drawn to canes under a vacuum to remove the interstitial gaps. Finally, the canes were drawn to fiber. An SEM image of the final fiber can be seen in Fig. 2(c). The core diameters in the fiber were between 2–3 μm with 3–4 μm center-to-center separations depending on the particular pairs of core sizes. There were 8100 cores in total in the final fiber. The outer diameter outer diameter the fiber was 550 μm with an imaging square size of 450 μm along the diagonal. All of the cores in our fibers are few-moded. Although it may seem that a fiber with only single-mode cores would perform better, we have found that the compromise of lowering the NA to achieve single-mode guidance is far more detrimental to the core-to-core coupling than any gain from fundamental-mode-only operation.

To compare the performance of our fibers we performed two tests. The first was to acquire fluorescence images of 1951 USAF test targets, and the second was to transmit a fringe pattern and measure the degradation of the fringe visibility with wavelength. All of our tested fibers were $\sim 90\mathrm{cm}$ in length.

An endoscopic fluorescence imaging system was built in order to obtain test target images which could be used to determine the resolution of our fibers at different wavelengths. A supercontinuum source filtered to two excitation bands (420–510 nm and 600–650 nm) was used as a light source for our experiment. The filtered excitation light passed through a dichroic beam splitter and was coupled into our fibers through an aspheric lens with an NA of 0.5. The USAF 1951 targets were imaged at zero working distance from the distal end of the fiber with either a green fluorescent or red fluorescent slide placed behind them. Light emerging back out of the proximal end of the fiber was imaged onto a CCD camera after passing through the dichroic beam splitter and a second collection filter with two wavelength bands, 520–600 nm (green band) and 650–750 nm (red band). These wavelength ranges for collection were chosen to be in the range of several reported chemical imaging probes, which have the potential to detect the presence of bacterial or fungal pathogens [8–13]. Using a two-color (wavelength band) system allows for the detection of multiple biological targets through the same imaging system [14]. Images of the test targets taken at our two wavelength bands can be seen in Fig. 3. Images from a commercial fiber can also be seen for comparison.

 

Fig. 3. USAF 1951 test target fluorescent images taken through our two fibers and a commercial fiber from Fujikura Ltd. (FGIH-30-650s). (a) Hexagonal array 520–600 nm, (b) hexagonal array 650–750 nm, (c) square array 520–600 nm, (d) square array 650–750 nm, (e) FGIH-30-650s 520–600 nm, (f) FGIH-30-650s 650–750 nm.

Download Full Size | PDF

In the green band images for both our fibers, several of the larger elements of group 7 are discernable down to element 4 in our hexagonal array fiber and element 3 in our square array fiber. (The elements have line widths of 2.76 μm and 3.10 μm, respectively.) These are comparable sizes to the core-to-core separations, indicating that very little light is coupling from an illuminated core into its neighbor in both fibers. In the red band the image contrasts are both degraded due to core-to-core coupling, principally between higher-order modes, which are visible in the dark regions. However, in our square array imaging fiber, the larger elements of group 7 are still discernable down to element 2, albeit with reduced clarity compared to the images taken in the green band.

Our second characterization method is a quantitative method developed to measure the global effects of core-to-core coupling and accurately characterize the performance of imaging fibers. This method has been developed to quantify the global coupling behavior of many cores in fibers where the crosstalk of individual core parings cannot be easily or quickly measured. In our fibers, no one core (or even a small set of cores) is representative of the whole array because of their different sizes and local environments (inside or at the edges of the stacked elements), and measurements of coupling from one core to its neighbors could be positively misleading. We have recently prepared an in-depth article detailing the method [15], so we will only briefly describe it here. The technique is similar to the measurement of the modulation transfer function in imaging systems [16], relying on quantifying the degradation in visibility of a transmitted fringe pattern as it passes through an imaging fiber. Two arms of an interferometer were interfered at a known angle to produce a vertical fringe pattern of known separation. The fringe pattern was transmitted down the test fiber and the visibility measured at the output on a CCD camera. Using a supercontinuum and a monochromator as an illumination source for our interferometer, we were able to measure the transmitted fringe visibility with wavelength. Our interferometer was set up with an angle of 2.29° to give 15 μm fringes at 600 nm wavelength. The fringe separation varies with wavelength due to diffraction and the corresponding variation can be seen on the top axis of Fig. 4. The fringe pattern was formed on the end of $\sim 90\mathrm{cm}$ sections of our fibers and the emerging pattern was imaged onto a CCD camera mounted on a goniometer in order to align the fringe pattern and CCD array. The output coupling was via a 0.5 NA aspheric lens.

 

Fig. 4. Fringe visibility measurements varying with wavelength for our hexagonal array fiber (red triangles), square array fiber (blue circles), and a commercial imaging fiber (Fujikura Ltd., black crosses). The visibility is measured after transmission through approximately 90 cm of fiber in each case. Error bars represent the standard deviation of a series of data points taken at a single wavelength, adjusting and then realigning the system between each measurement.

Download Full Size | PDF

In an ideal fiber, a fringe pattern would not lose any visibility as it passes down the length. However, core-to-core coupling reduces the contrast in the image and hence the measured visibility. The fringe visibility for our hexagonal array fiber, square array fiber, and a section of FGIH-30-650s [5] fiber from Fujikura Ltd. is shown in Fig. 4. Transmission is shown for wavelengths between 500 nm and 700 nm. The FGIH-30-650s fiber is common in commercial systems specified over our wavelength band. The center-to-center core separation of this fiber was measured to be around 3.5 μm, with core diameters of between 1.7 μm and 2.1 μm and NA reported to be 0.4 [6]. The cores in the fiber appeared to have a random arrangement. The fringe visibility of our hexagonal array fiber at short wavelengths is highest up to a wavelength of 550 nm, but its performance begins to degrade as the wavelength increases. As there are only three core sizes in this fiber, we attribute this degradation in fringe visibility to light coupling to the identical next nearest neighbor cores. This is reinforced in the test target images where it can be seen in Fig. 3(b) that one particular size of core is coupling strongly into dark regions in a higher-order mode. In our square array fiber, the fringe visibility at short wavelengths is lower than the hexagonal array but generally more consistent and higher over the whole wavelength range, only starting to perform below the FGIH-30-650s commercial fiber above 680 nm.

We did not take care to keep our fibers straight during any of our experiments, and their performance did not depend on bending. The fringe visibility of our square array fiber at 650 nm was tested with a bend radius of 2 cm, and no degradation was observed.

We have presented two new methods of fabricating endoscopic imaging fibers using graded index preforms designed for telecommunications. Our fiber, based on a square array with several different core sizes positioned such that identical cores are not in close proximity, gives significantly improved imaging performance over a broad wavelength range compared to a widely used commercial fiber with higher numerical aperture cores. Fluorescence images of a USAF test target with a 3.1 μm line width are discernable in our square array fiber in the wavelength band 520–600 nm and 3.48 μm in the wavelength range 650–750 nm. This fiber’s fabrication technique was based on a simplified stacking procedure using rods derived from telecoms preforms of lower index contrast than commercial imaging fibers (NA of 0.3 compared to 0.4). This procedure therefore allows imaging fibers to be produced from relatively low-cost starting materials, and potentially paves the way for cost-effective disposable imaging fibers for use in clinical procedures.

The data underlying the results presented in this Letter are available at [17].