1. Introduction

Cancers arising from the surface lining of hollow organs account for the most of cancer deaths in the world. Endoscopic diagnosis, including staging, plays an important role in the management of these cancers, and forms the foundation of any minimal invasive intervention. Upper-tract urothelial carcinoma (UTUC), urothelial cancer of the ureter and renal pelvis, accounts for 5-10% of urothelial carcinomas. the annual incidence is about 2 new cases per 100,000 persons in the United States [1]. Radical nephroureterectomy (RNU) is currently the gold standard of treatment for most UTUCs [1]. Unfortunately, this may represent a significant over-treatment for about 40% of all UTUCs which are non-invasive and have very low progression risk [2]. Recent studies show that endoscopic treatment is an alternative method for low-grade and low-stage UTUCs, and this conservative treatment preserves renal function and is preferred because of the risk of cancer occurrence in the contralateral renal unit [1]. Accurate tumor staging and grading are important for effective endoscopic treatment of UTUCs. However, accurate staging and grading of the UTUC is difficult because of the ureter’s thin wall and narrow caliber which make endoscopic biopsy ineffective. High resolution imaging using endoscopic optical coherence tomography (OCT) may be an effective method to stage and grade UTUCs. OCT has previously been shown to be capable of grading bladder cancers [3]. It has recently been demonstrated that endoscopic OCT can identify the layered structures of the ureteral wall and can potentially be used to stage UTUCs [4, 5]in the ureter. But when the UTUC is localized in the renal pelvis, a side-viewing probe designed to image the side wall of tubular organs like ureter is not appropriate. The irregular shape of the renal pelvis would be better accessed by a forward-viewing probe.

To aid in staging of tumors in the renal pelvis by visualizing the layered structure of the urothelium, lamina propria and renal parenchyma, a miniature, flexible, scanning and forward-viewing endoscopic OCT probe is needed. The outer diameter of the probe should be less than 1 mm to fit through the working channel of a flexible ureteroscope. Although OCT probes smaller than 1 mm are challenging to fabricate with good optical properties, previous work has demonstrated that it is feasible (e.g [6–10].). Small (<1 mm diameter), side-viewing scanning endoscopic probes have been reported [7, 9], but not forward-viewing. Scanning forward-viewing probes have been demonstrated using various technologies such as PZT actuators [11], MEMS mirrors [12], counter-rotating wedges [6], magnetic coils [8] and rotating torque cables [13], however all of them are either too large [11–13], or not flexible [6, 8], and thus are not suitable for deployment through a flexible ureteroscope. Thus there is an unmet need for a small, scanning, forward-viewing flexible endoscopic OCT probe.

In this paper, we describe a flexible, 0.7 mm diameter (including protective sheath), cone-scanning, forward-viewing OCT probe which can fit through the working channel of a flexible ureteroscope and is feasible for renal pelvis imaging. The probe optics consist of a section of core-less fiber and a section of GRIN fiber spliced to the end of SM fiber. The probe is intended for use in a common-path OCT (CP-OCT) system, so the relatively high back-reflection from the GRIN/air interface is used as the reference signal [14]. Besides simplicity and compactness, advantages of a CP-OCT system include phase stability and not needing dispersion compensation [15]. A disadvantage of CP-OCT is that it is challenging to adjust the reference signal power in order to optimize the OCT system sensitivity. In the design presented here, the reference signal intensity is adjusted by the polish angle on the GRIN fiber end.

2. Methods

A photograph of the OCT probe optical elements is shown in Fig. 1(a). The distal focusing optics consist of a 200 µm section of core-less fiber (HPWR040, OFS) as a spacer to allow the light to diverge, and a 150 µm section of GRIN fiber with g = 6.4 mm-1 (GIF625, Thorlabs Inc.) to focus the beam. The core-less fiber and the GRIN fiber were fusion spliced to the end of a single-mode (SM) fiber (SMF-28e, Corning Inc.). The mode-field diameter of the SM fiber is 9.2 µm, and the core diameter of the GRIN fiber is 62.5 µm. An ABCD matrix simulation method [16] was used to calculate the expected beam spot size (16 µm FWHM) of the probe, while fabricated probes with spot sizes less than a specification of 18 μm yield high image quality and are acceptable.

 

Fig. 1 (a) Microscope image of a fabricated OCT probe consisting of SM fiber, core-less fiber, and GRIN fiber; (b), (c) Light back-reflected at the GRIN fiber/air interface and re-coupling to the SM fiber from a flat-polished probe tip; (d), (e) Light back-reflected and re-coupled from an angle polished probe tip.

Download Full Size | PDF

In the probe, a small amount of light is back-reflected at the fiber/air interface. The back-reflected light will propagate back through the GRIN fiber and the core-less fiber, and then part of it will re-couple into the SM fiber. The recoupled back-reflected light serves as the reference signal of the CP-OCT system. The size of the field of the back-reflected light at the SM fiber/core-less fiber interface is larger than the SM fiber mode-field diameter. To simplify the calculation to predict the amount of light recoupled into the SM fiber, we assumed that only the light incident within the SM fiber core area will couple into the fiber. As the light beam intensity has a Gaussian profile, with the majority of the light being in the center, when the GRIN fiber end is flat polished (0 degree angle), the center part of the back-reflected field will couple into the SM fiber and result in the highest coupling efficiency [Fig. 1(b) and 1(c)]. If the GRIN fiber end is angle polished, there will be an offset between the beam center and the SM axis, which will result in lower coupling efficiency [Fig. 1(d) and 1(e)]. So, the sensitivity of the OCT system can be optimized by adjusting the reference signal power [17], [18], and this can be done by altering the polish angle of the probe tip [18].

The fabricated probe optics were mounted in a metal housing with a 150 µm inner diameter and 8 degrees of curvature. The metal housing was then glued into a flexible torque cable[Fig. 2(a)]. As the metal housing is curved, the beam from the probe will be offset from the probe’s center axis. A motor-driven fiber-optic rotary joint was used to rotate the torque cable, generating a cone-scanning pattern. All the rotating parts were housed in a biocompatible plastic sheath to isolate the rotating parts from the sample. For our design, the tilt of the fiber optics, and therefore the scanning range of the probe is mainly limited by the size of torque cable, with bigger torque cables resulting in larger scan ranges but bigger probe diameters. In this demonstration, a thin torque cable with a 0.5 mm outer diameter (0.3 mm inner diameter) was used, producing a cone-scanning circumferential range of about 1.0 mm. The outer diameter of the plastic sheath was 0.7 mm. Figure 2(b) shows a photograph of a 0.5 mm forward-viewing probe without the plastic sheath with a ruler for scale.

 

Fig. 2 (a) Structure of the OCT probe: the probe end is bent to offset the focus point from the center axis and glued in a flexible torque cable. (b) Image of the flexible 0.5mm forward-viewing probe without the plastic sheath. (c) Schematic of the CP-OCT system, SLD: super luminescent diode, CIR: fiber circulator, RJ: rotary joint

Download Full Size | PDF

The probe was tested using a common-path spectral-domain OCT system [Fig. 2(c)]. A 1310 nm super-luminescent diode (SLD, IPSDM1312, Inphenix) source with a 75nm FWHM bandwidth and a 15mW output power was used as the light source, a linear-in-wavenumber spectrometer with line-scan camera (SUI-LDH, Goodrich, Corp.) [19] was used to detect the spectral interference signal, and a miniature fiber-optic rotary joint (FORJ RFCX-121-28, Princetel, Inc.) was used to couple the light from the SLD to the probe. The measured axial resolution of the system is 15 µm.

To choose the optimum polish angle to maximize the sensitivity of the CPOCT system, the recoupling efficiency of the back-reflected reference light as a function of polish angle was estimated, assuming that only the back-reflected reference light incident within the SM fiber core area will couple into the fiber. As the light beam intensity has a Gaussian profile, the recoupling efficiency as a function of polish angle is similar to a Gaussian distribution, as shown by the black solid line in Fig. 3(a). Using this estimate of the reference signal, the sensitivity S as a function of polish angle was estimated:

 

Fig. 3 (a) Normalized re-coupling efficiency (compared to the flat cleaved tip) and system sensitivity as a function of polish angle. The black solid line is the calculated normalized coupling efficiency of the reference signal; the red dots are the normalized measured coupling efficiency; the black dotted line is the calculated sensitivity; the blue diamonds represent the measured system sensitivity. (b) Different noise intensities as a function of polish angle. The black solid line is the shot noise, the red dotted line is the excess noise, and the blue dotted line is the receiver noise.

Download Full Size | PDF

where S_peak is the OCT signal resulting from an ideal reflector as the sample, and ${\text{δ}}_{noise}^2$ is total noise of the system (variance). The model includes three sources of noise, namely receiver noise, shot noise, and excess noise, and different noise sources dominate at different levels of reference power [17, 20]. The model was modified to vary the integration time of the line-scan camera from 7 μs at flat polish to 17.3 μs at 2.1° polish angle and beyond, to represent the experimental constraint of avoiding saturation of the camera. Beyond 2.1° polish, the integration time was fixed at 17.3 μs (the maximum enabled by the camera). The reason both parameters were varied (reference arm power and integration time) was because the objective was to optimize the sentitivity of the whole instrument under test. In general, with increasing polish angle (decreasing reference signal intensity), the sensitivity will increase at first, but then decrease. This is because when the reference signal intensity is too high, excess noise (intensity noise) dominates over shot noise, compromising sensitivity, and when the reference signal is too low, receiver noise dominates, again compromising sensitivity. Figure 3(b) illustrates the three noise sources as a function of polish angle. Shot noise and excess noise are modeled, while receiver noise is measured. When the polish angle is smaller than 2.0°, excess noise (the red dotted line) is dominant. When then polish angle higher than 3.6°, receiver noise is dominant (blue dotted line). And between 2.0° and 3.6°, shot noise is dominant (black line). From the simulation results, it was found that a 2.3° polish angle would result in the highest sensitivity. The black dotted line in Fig. 3(a) illustrates the calculated probe sensitivity as a function of polish angle.

To verify the simulation results, a flat polished probe was fabricated, and the recoupling efficiency and sensitivity were measured, then the same probe was polished at 2° and tested, then polished at 4° and tested again. During the angle polishing process, the fiber probe is fixed on a rotary mount (PR01, Thorlabs Inc.), and the polishing angle is controlled by adjusting the rotary stage. The measured back-reflected reference light recoupling efficiencies were 0.24%, 0.14%, and 0.03% for the 0°, 2°, and 4° polish angles, respectively. The red dots in Fig. 3(a) show the measured normalized recoupling efficiency (relative to the flat polished GRIN fiber end) as a function of polish angle. The experimental result is very similar to the simulation. The difference between the simulation and the experimental results are likely caused by fabrication variability, such as the polishing angles [18] and the core-less fiber and GRIN fiber lengths.

The sensitivity of the probe at the three polish angles was also measured. For each measurement, a silver coated mirror was aligned to maximize recoupling into the probe (6.4 mW, at 1.4 mm from the probe tip). Then the mirror was tilted to attenuate the reflectivity by −36.3 dB to 1.5 μW. The camera integration time for 0° and 2° polish angles was adjusted to 7 μs and 15.4 μs, respectively, such that the reference arm intensity did not exceed 85% of the saturation level. The integration time for the 4° polished probe was set to 17.3 μs, which is the longest integration time possible for the camera at 47,000 A-scans per second. The measured sensitivities for the 0°, 2°, and 4° polish angles were 92.4.0 dB, 96.1 dB, and 90.7 dB, respectively (blue diamonds in Fig. 3(a)), and the 2° polish angle resulted in the highest sensitivity, about 3.7 dB higher than the flat-polished probe, in agreement with the simulation. The measured sensitivity was about 13 dB lower than the simulation results at all 3 polish angles, likely due to signal fall-off, loss in the spectrometer, and other sources of noise and loss not accounted for in the model.

3. Results

The CP-OCT system and probe was first tested by imaging human skin in vivo. Images of thick skin at the proximal interphalangeal joint of the human little finger were acquired using a 2° polished 0.5 mm flexible forward-viewing probe. As the 1.2 m long, 0.5 mm torque cable has limited torsional stiffness, the rotation speed was affected by the bending of the plastic sheath because of non-uniform friction. While the rotation frequency was maintained at 600 rotations per minute, noticable non-uniform rotation distortion (NURD) was observed (Fig. 4(a), where N denotes normal rotation and S indicates slow rotation). A simple correlation algorithm was used to reduce the NURD. The correlation coefficients $\text{C}(\text{n}, \text{n}+2)$, $\text{C}(\text{n}, \text{n}+4)$, $\text{C}(\text{n}, \text{n}+6)$ were calculated, where n represents the A-scan number. If $\text{C}(\text{n}, \text{n}+2)$,$\text{C}(\text{n}, \text{n}+4)$, $\text{C}(\text{n}, \text{n}+6)$ were all higher than 0.8, then A-scan n was deleted. Figure 4(b) is the corrected image, displaying significant improvement in uniformity. However, NURD is not completely suppressed in Fig. 4(b), so future work will include reducing non-uniform rotation and improving distortion correction. Furthermore, for the cone-scanning images generated by the forward-viewing probe, the scanning range is smaller at the top than at the bottom. Therefore, a rectangle-shaped display will distort the tissue structure. A fan-shaped display can represent the image undistorted, with a uniform aspect ratio, by “unwrapping” the 3-D cone into a 2-D fan [21]. Figure 4(c) is the fan-shaped display of Fig. 4(b), which shows the image with a uniform aspect ratio.

 

Fig. 4 OCT images of the proximal interphalangeal joint of the human little finger from a 2° angle polished 0.5 mm flexible forward-viewing probe. (a) Original image with non-uniform rotating torque cable; S represents the slow rotating area, and N represents the normal rotating area. (b) Correlation treated image from (a) producing a significantly more uniform image. (c) fan-shaped image from (b)

Download Full Size | PDF

Finally, the system was used to image the renal pelvis ex vivo. A porcine ureter with a kidney attached was obtained from the Research and Skills Laboratories at the University Hospitals Case Medical Center. The tissue was transported to the imaging laboratory in cold saline immediately after the pig was euthanized. During the imaging process, the miniature probe with 0.7mm plastic sheath was introduced into the distal end of the porcine ureter and advanced to the renal pelvis. The probe tip was placed in contact with the wall of the renal pelvis near the ureteropelvic junction (UPJ) during OCT image acquisition. Figure 5 shows the image of the renal pelvis at the UPJ. Three tissue layers are readily differentiated in the OCT image, which we identify as urothelium, lamina propria, and renal parenchyma based on known renal tissue structure and previous experience with OCT imaging of the ureteral urothelium [4]. In the OCT images, urothelium appears as a thin layer with low signal intensity, the lamina propria appears as a higher back scattering layer, and the renal parenchyma appears as lower back scattering. Differentiating these tissue layers is critical for tumor staging in order to determine the level of invasion. Future work will include optimization and standardization of the probe fabrication protocol to improve optical performance and reproducibility.

 

Fig. 5 OCT ex vivo image of a procine renal pelvis at the ureteropelvic junction shows different structure layers. U: urothelium, LP: lamina propria, RP, renal parenchyma.

Download Full Size | PDF

4. Conclusion

In conclusion, we designed and fabricated a 0.5 mm (0.7 mm OD with sheath) flexible cone-scanning forward-viewing OCT probe which can be deployed through the working channel of a flexible ureteroscope for renal pelvis imaging. This was designed for a CP-OCT system, where the reference signal comes from the back-reflection of the GRIN fiber/air interface. A 2° polish angle at this interface resulted in optimum system sensitivity. A correlation method was used to correct the NURD of the thin probe. Imaging of human skin in vivo and porcine renal pelvis ex vivo demonstrates the feasibility of this thin forward-viewing probe.

In general, for in vivo endoscopic OCT imaging, high frame rates (e.g. video rate) are desirable to mitigate motion artifacts. The system presented here imaged at 10 frames per second. Because the kidney is not actively moving, and the probe is in contact with the tissue surface, 10 fps is likely to be sufficient to prevent motion artifact. If faster imaging is needed, the rotary mechanism and image acquisition chain are capable of much higher rates. The limitation is that as a proximally-actuated rotary probe is made thinner and more flexible, it becomes more susceptible to NURD. This is due to friction between the torsion cable and the sheath, and decreasing stiffness. NURD can be minimized physically by judicious choose of sheath and torque cable materials and design, and by optimizing numerical correction by post-processing.

In the future, the objective of the system is for clinical imaging of the renal pelvis of patients through the working channel of a flexible ureteroscope. This miniature design may also enable other clinical applications which require small forward-viewing endoscopic OCT probes, such as real time monitoring of radio-frequency ablation (RFA) therapy for cardiac arrhythmias [13].