1.

Introduction

Optical coherence tomography (OCT) has been developed for over a decade. Currently, it is widely used for microscale diagnosis and therapy due to the method’s specific advantages such as high depth and transverse resolution and its especially nondestructive nature.¹ There are many types of OCT systems that differ according to the signal processing in terms of time and frequency, beam forming, guidance, and application fields. Endoscopic OCT is used for the high-resolution diagnosis and therapy of internal organ systems in fields such as cardiology, urology, dermatology, ophthalmology, gastroenterology, cochlear implants, and for obtaining biopsies of various types of cancers.²3.^–4 The noticeable disadvantages of endoscopic OCT are the limited scanning depth for dispersed media and a small lateral scanning range. To overcome these limitations, many bi-directional improvements have been proposed for OCT probe guidance; these include side imaging and a forward imaging scope. The side imaging technique expands the scanning range through circumferential motion in conjunction with longitudinal translation, which is effective for inspection around small cylindrical passages such as blood vessels.⁵ An optical fiber with a scope probe is mainly actuated through base rotation, and its rotational acquisition combined with its longitudinal motion allows for the collection of inter diagnostic information around the outer walls.⁶ In contrast to side imaging, the forward imaging method emits a laser beam in the forward direction and controls beam forming using a microelectromechanical systems (MEMS) mirror or some active material. This microendoscope is used for scanning forward areas when directed toward a target surface. Different types of forward imaging OCT systems are described in Ref. 5.

Even though a microendoscope can acquire images on flat surfaces which are not accessible by the side imaging scope, it has a narrow field of view. Specifically, because the MEMS mirror or the cantilever-type guide is in the endoscopic tube, the scanning range is limited by its encirculating tube. For resolving this challenging issue, single mode fiber has been introduced for pose control of OCT probes,⁷ and dexterous robotics manipulation has been used in OCT guidance for intraocular surgery.⁸ Interested readers are encouraged to refer to Refs. 5, 9, and 10 for well-summarized literatures about the various endoscopic OCTs, including different probe designs, typical parameters, and their limitations.

The functions and specifications of OCT systems depend on their applications. Surgery for installing cochlear implants is one of the most challenging procedures for which the OCT system is of significant importance.¹¹12.13.^–14 In this paper, a parallel guidance OCT system is proposed for inter diagnostics with active cannulas. It could act as a solution in situations where the entrance passage to a target is narrow and crooked, and the manipulation of a rigid instrument could damage the surrounding structures, i.e., cochlear implant intervention. These and similar situations are currently handled using flexible instruments having limited configurations. The OCT system is designed to be integrated into and controlled wby active cannulas for minimally invasive endoscopic diagnosis.

The remainder of this work is organized as follows. In Sec. 2, the OCT system and its integration with active cannulas are described. In Sec. 3, the parallel guide system is presented with some demonstrative explanation, and finally, the experimental results are presented with conclusions and an outline of future work.

2.

Endoscopic Optical Coherence Tomography System for Biological Diagnosis

OCT is a noninvasive optical imaging technique that is generally used for biological tissues.⁶ By using a low-coherence optical interferometric method, OCT images of micrometer scale resolution can be obtained. This resolution is considerably higher than those of magnetic resonance imaging and other noninvasive imaging methods. Thus, OCT has been widely used for obtaining two- or three-dimensional images of tissues.¹⁴ However, applications of OCT are limited because the maximum scan depth of conventional OCTs is usually less than a few millimeters. An enhanced type of OCT has overcome this limitation by using a MEMS-VCSEL¹⁵ types swept source, which can image a few tens of millimeters. For solving the problem of the small range of depth, an endoscopic OCT was developed and studied.¹⁶ It uses a flexible optical fiber with a small-diameter gradient-index (GRIN) lens. Thus, this system could be integrated with catheters or insertion needles for image acquisition from biological targets located deep within the body.

Figure 1 shows the implemented OCT system where 1(a) is a fundamental OCT system and 1(b) is the integrated endoscopic system with active cannulas. In Fig. 1(b), the reference arm is noted by $\mathbf{R}$ and the sample arm is noted by $\mathbf{S}$; these are implemented into the inner tube of the active cannulas for the probe to be feedback-controlled by a control computer. The system uses spectral-domain OCT (SD-OCT), which affords better detection sensitivity and faster image scanning rates than conventional time-domain OCT. It uses a broadband light as a light source and does not need a scanning reference mirror. In this study, a superluminescent diode (SLD) with a center wavelength of 780 nm and full width at half maximum (FWHM) bandwidth of 50 nm was used as the light source. The center wavelength was selected as 780 nm because the absorption coefficient of water increases as the wavelength increases in the near-infrared region, and those of hemoglobin and melanin are too high in the visible region.¹⁷ The light from the SLD is passed through a $2\times 2$ coupler and is divided as $50:50$. Half of the light is directed at a retroreflector, which is used as a reference mirror, and half passes through the GRIN lens on its way to the sample. The diameter of the GRIN lens was 1 mm, which is small enough to be inserted through an active cannula. The length and working distance of the GRIN lens were 2.68 and 5076 nm, respectively. The axial resolution of the system is $6.5\mu \mathrm{m}$ in air and $4.9\mu \mathrm{m}$ in water, and its scan depth is 6 mm in air and 4.5 mm in water. The system’s lateral resolution is $23.7\mu \mathrm{m}$ in air and $31.5\mu \mathrm{m}$ in water.

Fig. 1

Schematic diagram of OCT system:(a) SD-OCT, (b) integrated OCT system.

Light reflected from the sample and the retroreflector (reference mirror) interfere in the coupler and are detected by a line scan charge coupled device camera having a resolution of 2048 pixels after passing through a diffraction grating and a collimating lens. The transmission grating (WasatchPhotonics, Logan, Utah, $1200\mathrm{l}/\mathrm{mm}$ @840 nm) and triplet collimators (Thorlabs, Newton, New Jersey, F220APC-780) were used for the experiment.

The line scan camera acts as a spectrometer because the two light streams are spectrally decomposed during their passage through the grating. The data obtained by the camera were sampled with equal wavelength increases and interpolated into $k$-space before Fourier transformation ($k=2\pi /\lambda$). The Fourier-transformed data represent the axial depth image of one point. The theoretical sensitivity of the system is 115.2 dB and its imaging speed is 50 A-scan/sec, where the scanning rate is set up with the proposed parallel moving speed on the probe guidance at the computation speed for Fourier image processing, and the operational conditions determine the imaging quality with resolution. A few system specifications are summarized in Table 1.

Table 1

OCT system specifications.

3.

Guidance and Control System

3.1.

Active Cannulas System

The proposed endoscopic OCT system is implemented with active cannulas, which act as flexible steerable manipulators that can guide the endoscope toward the target through a small passage. Active cannulas have attracted attention in the field of surgical robotics and among device manufacturers because of their steerability and flexible structure. They can access a target through small, crooked entrances, or natural orifices. Active cannulas consist of consecutive tubes of reduced diameters. Each tube has straight and curved sections from its base.

Active cannulas have been developed and clinically validated with a composition of successive telescopic, concentric, precurved, and superelastic tubes for in-depth analysis from mechanical, geometric, and systemic viewpoints. They are intended for various types of small-scale laparoscopic surgeries such as endonasal skull base surgery, brain surgery, blood vessel surgery, and reaching lesions in the peripheral lung.¹⁸ Active cannulas exhibit a novel and delicate motion performance at the millimeter scale due to enhanced dexterity of the end-effector because of the telescopically interlinked structure. This structure is realized by following a methodology of efficient kinematics computations, affording individual tube rotation and translation for controlling the tip’s pose. For satisfying microsurgical requirements while extending the device’s applicability to cochlear implant operations, especially for scanning the inner surface to find an intrusion point into the cochlea, an OCT integrated with active cannulas is proposed and implemented in this study. The objective of the developed active cannulas-guided OCT system is to be as minimally invasive as possible when inside the body.

However, the OCT system with active cannulas has some limitations in terms of parallel motion of its end-effector, i.e., the scope cannot have a sufficient parallel scanning motion without a large motion of the outer tubes. In this case, the entrance or passage could be damaged or the scope could vibrate with the active cannula’s bodies, thus introducing undesirable noise into the scanning result. Furthermore, the existence of nonlinearity and critical limits known as bifurcation¹⁹ could lead to unexpected pose errors, which is not desirable for microscale measurements in narrow spaces. The use of a cantilever-type probe could increase the scanning range on the frontal surface. However, probe orientation, which determines the beam direction, is varied per scan. Furthermore, it is difficult to determine its orientation of the tip because of the complex nonlinear characteristics of the interaction between the actuation force and the deflection. If accurate probe orientation during a scan is not known, the result does not reflect the actual state. This problem is significant when measuring flat surfaces. When the target surface is flat, the reflective signals weaken as the deflection angle increases. In other words, a parallel guide for the OCT probe is required for obtaining clear images through expansion of the lateral movement range. To this end, we propose an endoscopic OCT system which is guided in parallel for solving the abovementioned difficulties.

The active cannulas are made of a shape memory alloy (Nitinol)tubes manufactured by Nitinol Devices & Components Inc., Fremont, California, and they are controlled by a sophisticated control system. The active cannulas or steerable manipulators have multiple curved sections in the longitudinal direction; the shapes of these sections can be restored up to deformation. Each constructive tube has straight or precurved regions which are manufactured by heating at over 500°C for 30 min in a brass mold, followed by tempering in cold water. The brass mold for the thermal process is elaborately designed to ensure that the tubes have moderate stiffness and flexibility. In the process, there is much stress on maintaining the increasing and cooling speeds of the temperature for the object; hence, the mold has a specific design shape. It could be important in clinical aspects to specify these parameters on the tubes in the active cannulas as shown in Table 2.

Table 2

Active cannulas tubes specifications. [mm]

Tube	Do	Di	t	ls	lc	Iz	k
Outer	3.000	2.820	0.180	100.0	0	0.872	5.5
Inner	2.363	2.130	0.233	80.0	20	0.520	8.4

In Table 2, two tubes have differences in their outer/inner diameters ($D_{\mathrm{o}}/D_{\mathrm{i}}$) with a thickness ($t$), a straight length (${\ell }_{\mathrm{s}}$), curved lengths (${\ell }_{\mathrm{c}}$), area moments of inertia ($I_z$), and curvatures ($\kappa$) where all values are denoted based on [mm]-dimensions. The inner diameter $D_{\mathrm{i}}$ for the inner tube is important so as to allow the OCT probe to pass through the inside of the tube; the curvature with a curved length of the tube is a clinical design parameter to inspect or access the target for the OCT guidance. This parameter setup was designed with its major clinical target a cochlear implant, but it could be differentiated depending on its clinical purposes.

The integrated robot hardware system has dimensions of $80\times 70\times 400\mathrm{nm}$ in, $\text{width}\times \text{height}\times \text{length}$, as shown in Fig. 2.

Fig. 2

Active cannulas system.

The system has been specifically designed for control using active cannulas. It weighs less than 1.5 kg, has a portable size, and offers other facilities such as a power transducer and accessory cables. The control system is composed of four DC motors with two cannula tubes, two motors of which take part in the translational motion and the others in the rotational motion of each cannula tube. The inner tube fixes the probe inside and the outer tube changes the curvature of the inner tube by curvature overlapping by relative translation. The rotation of the probe pose is also performed by rotational motions of both tubes. With two-tube motions, the resultant workspace on the probe position presents a conic volume¹⁸ for covering a wider diagnostic space than a straight probe. All motors are controlled using a Maxon (Postfach, Switzerland) EPOS2 driver connected to a PC with a 1 Mbps CAN communication channel. The control module is written in C++.

Given that one of the possible applications is cochlear implant installation, where the surface of the inner organ has a circular patch with a diameter less than 0.5 cm, two active cannulas are used for achieving efficient scanning performance in the range. In the two-tube system, the parallel guidance OCT scope is implemented on the tip-end of the inner tube, and the pose of the guidance system is controlled by translation and rotation, or any combination thereof, of each tube. Even though there are some useful scope pose compensation algorithms^17,20 for measurement images, it is important to achieve fine control of the probe’s pose for acquiring an analyzable and diagnostically significant image quality. The end-effector pose is determined through kinematic calculations and is achieved by controlling the translation and rotation of the telescopic tubes, as described in Refs. 18, 19, 21, 22, and 23. Although a positional error of less than 1 mm has been achieved in real experiments, it is difficult to improve the system’s accuracy. In addition, it is difficult to control using real-time pose sensing feedback because acquisition of accurate pose information with the existing sensors is difficult. Furthermore, during the translational motion of each tube, there is internal sliding contact between the tubes’ walls, which leads to nonlinear pose errors of the end-effector. Therefore, a diagnostic system based only on active cannulas is insufficient for microscale measurement.

Orientation control is a major issue in measurement. Probe orientation is more important because it is directly related to the direction of the OCT probe, where its reflective ratio is also affected by the target surface texture. Because the precurved region of each tube could be changed under some experimental conditions and the analytical model is characterized by fundamental uncertainty errors of the real model, as well as those related to repetitive control accuracy with the OCT scanning sampling, the calibration of tip orientation is indispensable. For that purpose, visible rays are transmitted through the optical fiber installed for OCT, and their arrival point on the focusing plane is marked on a graph paper slide and measured with geometrical considerations. After calibration, a pose accuracy of under 0.5 mm and 1° was achieved. Note that this procedure is performed on the system operation level, not on the imaging level to enhance the pose control accuracy of active cannulas, such that the probe accesses the target sample.

3.2.

Parallel Guidance System

The active cannulas are used for expanding the scanning range on a target surface that offers a limited motion configuration because of its hidden location, which may require passage through long, crooked openings. Thus, the end-effector, which is the end of the manipulator, could be used for obtaining OCT images of complex electronic devices or diagnosis of an inner organ in a noninvasive manner.

Even if the positions of the active cannulas are controlled with three consecutive tubes, the end-effector orientation is not easily or accurately manipulated because the system has its own critical zone and has nonlinearities such as bifurcation. With respect to the OCT performance, the parallelism of the scope with the target surface is the most important function. Therefore, the active cannulas are required to use an additional system for ensuring parallel probe scanning motion.

The OCT probe is equipped with the parallel guidance system as shown in the following figure.

In Fig. 3, the proposed parallel guidance system is shown, where (a) shows a three-dimensional (3-D) view with some notations, and (b) shows the geometric schematic. In Fig. 3(a), there are four joints from $C_1$ to $C_4$ connecting neighboring linkages. The probe is loaded at holders $P_a$ and $P_b$, and each holder is allowed to rotate to the linkage bars. The control link $L_c$ plays the important role of pulling or drawing the joint $C_3$ through $C_1$, and it is connected to the actuator in the active cannulas system. Here, the control link is superelastic, and it can exert forces in both directions along the flexible cannulas’ structure. The left side of the joint $C_1$ is located inside the tube of the active cannulas. In Fig. 3(b), the OCT probe is represented by positions $P_a$, $P_b$, and $P_t$, where the tip of the probe is denoted by the subscript $t$. There are some geometrical relations according to the length of the control link $L_c$.

Fig. 3

Geometric representation of parallel OCT guide: (a) parallel guidance, (b) geometric schematic.

When the length of a linkage bar is $l_0$ and the length from $C_1$ to $P_a$ is $l_1$, it is assumed that by varying $dl$ of the control link, the angle of the linkage bar on the horizontal plane, $\theta$, varies from ${\theta }_1$ to ${\theta }_2$. Then, the following relation can be presented.

(1)

$$l_0·(\cos {\theta }_2-\cos {\theta }_1)=dl/2.$$

From Eq. (1), it is derived that

(2)

$${\theta }_2={\cos }^{-1}(\frac{dl}{2l_0}+\cos {\theta }_1).$$

The differential expression of the vertical and horizontal motions is as follows:

(3)

$$dv=l_1·(\sin {\theta }_1-\sin {\theta }_2),$$

(4)

$$dv=l_1·\{\sin {\theta }_1-\sin [{\cos }^{-1}(\frac{dl}{2l_0}+\cos {\theta }_1)\left]\right\},$$

(5)

$$dh=l_1·(\cos {\theta }_1-\cos {\theta }_0).$$

In Eq. (5), horizontal motion is defined according to the control length, and the gives the control rule for the actuator to maneuver the probe in parallel. In other words, the probe could be vertically moved downward from its initial state by translating the whole frame backward, where translational motion is carried out by the active cannulas. Figure 4 shows the results of two cases, namely without control and with control.

Fig. 4

Trajectories of system geometry: (a) uncontrolled trajectory, (b) controlled trajectory.

The controlled trajectory is attained by translating the point $C_1$ backward, while drawing the control point $C_3$ in every sampling interval. The target point $P_t$ could be vertically moved downward, so that the probe is guided to be parallel to the frontal surface. This analysis is applied to a real experiment in the same manner as that for control with the active cannulas. Other advantages of this system include its ability to generate another OCT scan in the lower side of the four bar linkages and to acquire 3-D information about the target by rotating around the middle axis, i.e., the line connecting $C_1$ to $C_3$, while maintaining vertical motion, i.e., spiral reconstruction of the target volume. The general procedure for measuring a sample is as follows:

• 1. The position and shape of the target sample is given.

• 2. The end-effector of the active cannulas is maneuvered to access the front of the target surface.

  • (a) The parallel guidance system is set up in the initial state.

  • (b) With the drawing and pushing of each joint, parallel scanning is performed on the sample surface.

• 3. After measurement, the guidance system is folded down to be inserted into the tube of the active cannulas.

4.

Experiments with Biological Samples

From the design concepts and geometrical motion analysis, a laboratory prototype of the guidance system is built to double the scale. Before scanning the sample, the active cannulas are controlled such that the end-effector is located near the target, and it is assumed that the position and orientation of the sample surface are known a priori. In the following, the OCT fiber is inserted inside the tube of the active cannulas, and the probe is loaded into two holders of two linkages. Between the holder and the probe, some slips are allowed to not fix either support position because the probe could be broken by the moment generated at the supports. Each linkage is made up of soft plastic materials that could pass through curved tubes. Figure 5 shows the configuration of the integrated system for experiments.

Fig. 5

Experimental setup for OCT guide system: (a) active cannulas and parallel guidance, (b) parallel guidance.

To validate and calibrate the integrated system, a coverglass sample having a flat surface and constant thickness was used. The obtained result was compared with the real value. The process was iteratively repeated until a reliable performance was attained. The result is shown in Fig. 6 where the horizontal axis shows the relative depth (A scan) in pixels and the vertical axis refers to the scan line (B scan). The relative depth is converted into the distance in $\mu \mathrm{m}$ by the refraction index and the relative pixel difference.

Fig. 6

Calibration result for OCT image.

After calibration, the microwell plate that has regular cylinder patterns was used as a target sample for verification of the system. The sample specification and its scan result are displayed in Fig. 6. Considering the refraction index of glass as a reference sample and poly dimethylsiloxane (PDMS) as the target sample, real distances in pixels are converted and analyzed. The height difference between the two layers is 54 pixels. Using the refraction index of glass as 1.5, each pixel has a distance of $4.33\mu \mathrm{m}$ and the measured thickness is $234\mu \mathrm{m}$, while the real value is $250\mu \mathrm{m}$ (difference $16\mu \mathrm{m}$). Similarly, the target sample can be analyzed in Fig. 7. In Fig. 7, the height difference between the stepped layers is 52 pixels. Using the refraction index of PDMS as 1.4 and with each pixel having a distance of $4.64\mu \mathrm{m}$, the measured embossed height is $241\mu \mathrm{m}$ and the measured value is $257\mu \mathrm{m}$ (a difference of $16\mu \mathrm{m}$). From the result of analysis, the measurement error is 6.4% for the glass sample and 6.2% for the PDMS sample. All results show that controlled scanning ensures a reliable accuracy of less than $100\mu \mathrm{m}$ for diagnosis of biological samples such as tissues or cells.

Fig. 7

Calibration by reference sample: (a) reference sample, (b) reference measurement.

In Fig. 8, two reference samples are tested for calibration of the parallel guidance in the vertical scale. The scanning probe was vertically moved from top to bottom, and its result is plotted from its axis origin. The first result displays the expanded vertical range by discriminating seven patterns which amount to a vertical range of 3 cm. Even though the guidance is double the scale of the original design, the expanded view for the inter diagnostics could considerably reduce the measurement time and the cost.

Fig. 8

Measurement result for artificial sample: (a) on Microwell plate, (b) on Scotch tape.

However, there are some weaknesses in the result. During 10 scans from the start, some noises could be found due to actuator start-up motions. Actually, this phenomenon is not desirable, and it could be reduced by lowering the startup acceleration with elaborate manufacturing of the mechanical parts. Another noticeable finding is that the clarity and accuracy gets worse and this originates from the orientation changes of the probe tip due to other nonlinearities within the mechanical components. For this drawback, it is considered that roughness of the surface has a significant role in the measurement result. To compare the dependency on the roughness of the surface, a Scotch (St. Paul, Minnesota) tape was tested. As seen in Fig. 8, the case for (b) shows clearer layer boundaries than (a). From the results, it is considered that there is diffuse reflection instead of specular reflection since the tape surface is rough, and, therefore, the obtained image quality was robust to the orientation changes in the microwell plate. The result confirms the proposed measurement is effective for the measurement of most biological samples as its surface is not smooth. The pixel distance between the outer lines is $25(10+15)$ resulting in $146\mu \mathrm{m}$, and it matches well with the datasheet of the Scotch mask tape No. 214, where the thickness is $147\mu \mathrm{m}$.

Figure 9 shows the measurement results on some biological samples, specifically onion bark and a mouse dermis fixed on a coverglass (70–95 pixels). In Fig. 9(a), the thickness and density of the media in the bark could be acquired and Fig. 9(b) also shows some information about the layers and cilias distribution.

Fig. 9

Measurement results for biological samples: (a) on an onion bark, (b) on a mouse dermis.

Figure 10 presents a measurement result on the near part of retina and optic nerve in a mouse eyeball, which was described in Ref. 24. In Fig. 10(a), a part of the mouse eyeball is measured by an in vitro setup and Fig. 10(b) depicts its zoom-in result with the real depth scale in $\mu \mathrm{m}$ and with the same intensity pixel level (color bar) as used from Figs. 6 to 9.

Fig. 10

Measurement results for mouse eyeball part: (a) on a mouse optic nerve, (b) magnified result.

Note that in all the figures displayed, the horizontal axis of the depth is converted by a corresponding refraction ratio $n$ by scaling $6.5/n\mu \mathrm{m}$ per each pixel. In Figs. 9 and 10, both have a scaling of $4.89\mu \mathrm{m}/\text{pixel}$ where $n$ equals 1.33.

In the result, the biological information on media such as density, layers, spots, or variations of the density could be given for diagnosis. As an advantage for the system, it also has fast speed for scanning a line within only a few seconds where the speed is slowed to 35 ms in each A-scan, while still avoiding any vibration on the tip; however, its speed is faster than another comparable system as in Ref. 7. It is notable that the measurement speed for the proposed device largely depends not only on the sampling time per scan, but also on the moving speed of the parallel guidance, such that the imaging processing could be increased by the latest computing technology of multiprocessing or GPU processors while maintaining the measurement time; this could be performed as a future work.

5.

Conclusions

A parallel guidance endoscopic OCT system integrated with active cannulas was proposed and a prototype was tested for procedural implementation for inspecting internal organs or complex structures accessible through small passages. The system can scan the frontal surface in parallel, and it has larger view, thus overcoming the diametrical limitation of previous endoscopic OCT systems. It also has the capability to use simultaneous scanning by dual probes on the upper and lower side linkages. By spiral scanning guidance motion, the 3-D volumetric information could be attained in a single measurement with a reduced time.

The efficient structure of the proposed system could accommodate various types of microscale operations for physical examination, and its expanded view of internal diagnostics could considerably reduce the measurement time and cost. The system could also be used with other successive interventional operations with active cannulas such as laser ablation, biopsy, and needle insertion. More accurate pose control and increased scanning performance are required for commercial development. Also, a systematic and target media-independent performance needs to be attained in future work.