Eye Explorer: A robotic endoscope holder for eye surgery

Abstract Background Holding endoscopes by hand when performing eye surgery reduces the dexterity of the surgeon. Methods A robotic endoscope holder called “Eye Explorer” is proposed to hold the endoscope and free the surgeon's hand. Results This device satisfies the engineering and clinical requirements of eye surgery. The force for manual operation is less than 0.5 N. The observable ranges inside the patient's eye considering horizontal and vertical perspectives are 118° and 97°, and the motion of the holder does not interfere with the surgeon's hand and other surgical devices. The self‐weight compensation can prevent the endoscope from falling when extra supporting force is released. When comparing the external force exerted on the eye by the Eye Explorer with that in case of manual operation, a decrease of more than 15% can be observed. Moreover, the consumption time of endoscope view adjustment using the Eye Explorer and manual operation does not significantly differ. Conclusion The Eye Explorer allows dual‐hand operation, facilitating a successful endoscopic eye surgery.


Preceyes surgical system developed by Preceyes BV has been
demonstrated to safely assist doctors in performing retinal vitreous surgery on humans, 6 receiving CE marking approval in 2019.
When using a co-manipulation system, the surgeon directly holds the surgery device connecting to the actuation part to do the operation. The motion of the device is mainly determined by the surgeon, but the actuation part also outputs auxiliary forces to reduce the hand tremors or constrain the motion within a safe area.
The Steady-Hand Robotic System 7 developed by John Hopkins University, is a representative co-manipulation system. Patel et al.

applied an adaptive control method on the Steady-Hand Robotic
System to make the sclera force or insertion depth of the tooltip be within the pre-defined safe trajectories. 8 The KU Leuven robotic system can be operated in both telemanipulation and co-manipulation modes. By using the KU Leuven robotic system, the first roboticassisted endovascular surgery was performed on a human patient in 2018. 9 Regardless of the operation method, a clear view of the operation field is crucial in case of eye surgery procedures. Microscopes are widely used to obtain a clear view. Some improvements on the microscope were developed for integrating them with a robotassisted eye surgery. Inoue et al. 10 developed a wide-angle viewing system to provide a wide-ranging view of the eye. Zhou et al. 11 developed an image guiding system for a surgeon looking through a microscope. Mukherjee et al. 12 developed a fast and accurate algorithm that can map the retinal vasculature and localize the retina with respect to the microscope. However, it is difficult to observe the intraocular area that is at the back of the endoscope because of the fixed perspective. Additionally, the condition of the patient's eye may affect the quality of the view. A surgeon may not be able to obtain a clear view if the cornea is muddy or the pupil is shrunken.
Optical coherence tomography (OCT) is used to compensate for the disadvantages associated with the microscope. OCT can be used to obtain detailed images of the operation area inside the eye, and the quality of the image is not affected by the condition of the patient's eye. Thus, some groups have investigated OCT-integrated robot-assisted eye surgery. Yu et al. evaluated the efficacy of robotassisted microsurgery with OCT guidance. 13 Chen et al. evaluated an OCT-guided robot-assisted cataract removal surgery. 14 Balicki and Yang respectively evaluated the utility of the OCT-guided eye surgery with respect to the co-manipulation robot systems. 15,16 However, the view range of an OCT image is limited, and the operation of real-time OCT for view alternation has been reported as 'complicated' by surgeons. Furthermore, the performance of the OCT images, considering real-time situations, is poor because the highest imaging speed is 10-20 volumes/s, 17 even when using a high-grade graphics processing unit.
In such cases, using an intraocular endoscope is an alternative choice. The intraocular endoscope can provide real-time images independent of the condition of the patient's eye. Moreover, the view range, location and zooming capacity of an endoscope can be altered as the surgeon deems necessary. In many cases, endoscopes and microscopes are used together. Therefore, the usage of an intraocular endoscope in eye surgery has increased, especially in case of the vitrectomy surgery, retinal vein occlusion, retinal detachment and intraocular tumour treatment.
However, the adoption of endoscopic eye surgery is still limited despite its benefits. This can be attributed to the fact that the surgeons have to use one hand to hold the endoscope, making it difficult to perform dual-hand operations. Whereas the operation dexterity of dual-hand operation is expected in complicated surgeries such as the inner limiting membrane detachment. 18 Tadano et al. developed a robotic laparoscope holder that allows the surgeon to intuitively adjust the laparoscope without the hand motion. 19 Therefore, the authors apply the same concept of robotic holder to an endoscopic vitrectomy surgery and propose a novel robotic endoscope holder that will facilitate the endoscopic eye surgery in terms of liberating the surgeon's hand to allow dual-hand operation. This device is expected to popularize endoscopic eye surgery. Figure 1 depicts the conceptual design of an endoscopic eye surgery performed using a robotic endoscope holder and a microscope. A slim-sized robot arm must approach the surgical field near the surgeons' hand, stably holding an endoscope and precisely controlling the viewpoint according to the surgeons' command.
The structure of this paper is as follows. In Section 2, we discuss the requirements and the design of the proposed endoscope holder as well as some functional specifications. Section 3 presents the experimental results concerning each clinical requirement, and Section 4 presents the discussions and conclusions. F I G U R E 1 Application of a robotic endoscope holder for eye surgery 2 | MATERIAL AND METHODS

| Engineering and clinical requirements
The engineering requirements are related to the quality of design, which can be given as follows: � Easy sterilization: As required by most of medical devices, the non-sterilizable part (endoscope) must be easily separated from the sterilizable part (holder) for sterilization and tool exchange � Easy manual operability: Similar to other surgical devices, the surgeon may have to manually move the end effector of the device; thus, it is expected that the device can be easily moved without large mechanical impedance The clinical requirements are related to the compatibility of the design with respect to real surgical cases and can be given as follows: � Sufficient observable angle: Because the surgeon uses an endoscope to view the places that cannot be viewed using a microscope, the robotic holder should allow the endoscope to provide views covering most of the inside area of the eyeball, regardless of the insertion direction. As a result, the required observable angles in the horizontal and the vertical perspectives are 90°and 80°, respectively.  Based on the data obtained with respect to the behaviour of an expert surgeon, a sclera force of less than 120 mN is considered to be safe. 20

| Arm unit
The arm unit has three DOFs of rotary joints. Joint J2 is a parallel link mechanism, whereas J3 is a serial wire-linkage mechanism (the stainless wires were pre-stretched before winding at the pulleys to eliminate the relative sliding between the wire and pulley).  Figure 4 also shows the pull-springs inside the arm unit.

| Holder unit
The holder unit is a passive gimbal mechanism with two DOFs, as shown in Figure 5. This passive mechanism was selected because of its compact and lightweight design without actuators. The benefits of the passive mechanism will be presented in Section 2.3. In addition, because of the linkage mechanisms of the arm unit, the angle of the yaw axis to the horizontal plane is maintained at 45°or À 45°, which cannot only provide a sufficient operation space to the surgeon's hands but can also avoid a singular posture at which the endoscope, the base of the holder unit, and the second link are parallel. The posture in which moving the endoscope view left and right will exert a large force on the eye of the patient.
The holder unit uses a clip to attach the endoscope on a simple drape cover, as shown in Figure 5. This makes it easy to separate the non-sterilizable part (endoscope) from the sterilizable part (holder unit), satisfying the engineering requirement of easy sterilization.

| Operation method
This device can be operated in two modes, that is, the passive mode and the active mode. Figure 6A, B depict the operation scheme of the Eye Explorer in these two modes, respectively. Table 1 presents the motion states of the three mechanical units in each operation mode.
In the passive mode, which is mainly considered before and after the surgery, the surgeon pulls the endoscope to the target position, inserts it into the eyeball through a trocar hole and adjusts the endoscope such that it points toward the area where the surgical operation will begin. In this mode, all motions are driven by the surgoen's hands. As shown in Figure 6A, the translation drive unit (XYZ stage) is locked in the passive mode, and the electromagnetic breaks in the arm unit are released, indicating that the arm unit can be freely moved by the surgeon. The arm unit was designed to allow easy manual positioning of the endoscope. As shown in Figure 3, the electromagnetic brakes and endoscopes for each joint were located at the base of the arm unit to reduce the weight of the movable section, and the released electromagnetic breaks have low mechanical impedance.
The surgeon will begin to operate on the patient's eye after inserting the endoscope into the eyeball and adjusting the view area.
During the operation, the surgeon can use the active mode to alter the view provided by the endoscope. As shown in Figure 6B, in this mode, the electromagnetic breaks in the arm unit are locked and the whole arm unit is rigid. The posture and position of the passive endoscope holder unit are only determined by the translation motion of the XYZ stage. F I G U R E 5 Holder unit mechanism and the manner in which it adapts to an endoscope Because the surgeon uses both hands to operate, we designed a footswitch ( Figure 6C) that will allow the surgeon to intuitively adjust the endoscope view by foot. The black square at the left side is movable, and it is connected to a potentiometer that measures its motion direction. Thus, the endoscope camera, which shows the view on a screen, will move towards the direction where the surgeon moves the black square. The joystick at the right side determines the zooming of the endoscope view. Thus, the endoscope view can be adjusted without interrupting the hand operation.
Furthermore, if necessary, the surgeon can adjust the view by hand by switching to the passive mode. The two modes are alternated with another footswitch (not shown in Figure 6), with the device remaining in passive mode if the surgeon continuously steps on the footswitch. Hence, mode changing can be achieved easily by slightly uplifting one foot.

| Control method
When the surgeon operates the foot switch in the active mode, the motion of the XYZ stage determines the velocity of the arm base (V base ). The velocity of the gimbal centre at the holder unit relative to the coordinate system of the arm base (V gim ) equals to V base because the arm unit with locked joints can be considered as a rigid body. Therefore, a necessary V base required to generate the expected V view can be calculated by the following function: where R r1 is a transition matrix that changes the coordinate system of the endoscope to that of the base. Meanwhile, R r2 is a transition matrix that changes the coordinate system of the screen into that of the endoscope. Furthermore, R r1 is derived from the rotation angles at each joint, which are measured by the encoders (arm unit: MAH-19-524288N1; holder unit: MAH-19-524288N1, MTL lnc.).
We implemented a velocity control of the XYZ stage in each direction using V base in Equation (1)

| Working range and compatibility
The Eye Explorer is mounted on a movable stand (wheels of the stand can be braked) next to the surgical table; the height of the movable stand can be adjusted according to that of the surgical table. The endoscope can access the objective eye from an angle of 45°. Figure   10 indicates angle of the endoscope in Figure 10A and the yaw angle in Figure   10B are maintained constant. The observable ranges of the endoscope in the vertical and horizontal perspectives are 118°and 97°, respectively.
In cases that the endoscope holder and objective eye are on different sides of the patient's nose, the required and real observation range considering both perspectives are 118°and 85°, as shown in Figure 11 and the angle of the yaw axis of the holder unit relative to the horizontal plane is À 45. Furthermore, the observable ranges considering both perspectives are larger than those required (90°and 80°), regardless of the access manner.
The red sectors in Figures 10 and 11

| Performance of self-weight compensation
The needle-like endoscope must not fall into the patient's eye.
However, this may easily occur if the surgeon accidentally releases the device in the passive mode and could also occur in the active mode when the power to the electromagnetic brakes is cut off. Therefore, we developed self-weight compensation for the device using the mechanism of the arm unit described in This section introduces an experiment to evaluate the adopted self-weight compensation. We measured the necessary force to keep the positon of the endoscope without external supporting. Figure 12 presents

| External load exerted on the eye
When adjusting the endoscope view in the active mode, the interaction between the endoscope and insertion hole determines the external load exerted on the eye of the patient. This load should be small to ensure the security of the eye of the patient.
To measure the force exerted on the patient's eye, a virtual eyeball with a diameter of 24 mm was made by 3D printing, on which an insertion hole with a diameter of 1.0 mm was set for the 23G needle-like endoscope. Figure 14 depicts the designed experimental apparatus. A force sensor (MIRCO 4/20-A, BL. AUTOTEC. LTD., resolution: 0.04 N) was mounted onto the virtual eyeball. Four view fields were assigned inside the virtual eyeball, as shown in Figure 15.
A black ring was attached on the monitor screen as a sighting mark for view adjustment.
Five test operators moved the endoscope along the designated route, as shown in Figure 15, by operating the footswitch shown in

| Evaluation of the operating time
Although a robotic endoscope holder may provide benefits, an increase in operation time would likely cause usability and implementation issues. Hence, we investigated the operating time of endoscope manipulation. Figure 17 shows the comparison of the time spent to complete the view adjustment task between the manual and robotic operations described in Section 3.3.2. The pair corrections between the consumption time of two operation manners are high because every pair of data was generated from the same test operator. Hence, the paired student's t-test was feasible despite the small sample size. 22 The null hypothesis of the paired student's t-test was: the average consumption time of the manual operation t man equals that of the robotic operation t rob ðH 0 : t man À t rob ¼ 0Þ. The significant level was 5%. The option of standard deviation was set unknown.
The testing result showed that |t| ¼ 1.41, which is out of the rejection region of |t| ≥ 2.57. Hence, we can accept the null hypothesis, which means there are no difference in consumption time between the manual and robotic operations. Moreover, the moving speed of the Eye Explorer was set to 15 mm/s at the endoscope tip; the operators were instructed to move at normal speed during manual operation.

| DISCUSSION
As introduced in Section 3.1, the required force applied by the operator to move the endoscope holder in the passive mode is similar to that used to hold a tool of 50 g, which would be considered lightweight by most people. Therefore, the engineering requirement of easy manual operability can be satisfied.
By calculating the visible area through the endoscope when using this device, the endoscope view covered 70% of the intraocular area, regardless of the insertion manner. Thus, the clinical requirement of a sufficient observable range is satisfied.
Considering the compatibility of the device, first, the width of the XYZ stage (337 mm) was less than two-third of the surgical table  requirements of the device workspace considering the vertical direction (185-235 mm) when using a surgical microscope; therefore, the motion of this device will not interfere with the operation.
Moreover, the space above the head of the patient, which is occupied by the device, was smaller than that occupied by the device developed by Nassari, 2 the IRISS, 3 the KU Leuven robotic system 6 and the Preceyes surgical system, 7 all of which are representative surgical robot systems for eye surgery. Therefore, the clinical requirement of compatibility is satisfied.
An ideal self-weight compensation requires that the position of the endoscope is preserved in situations of power off. In this case, the forces measured by the force sensor should be zero. However, it is difficult to realize the ideal self-weight compensation because of the weight uncertainties (cable, tiny parts, etc.). Hence, we overestimated the weight in the design of the equipment. The real effect of the selfweight compensation in the upward forces is shown in Figure 13. The upward forces may retract the endoscope out from the patient eye; the force exerted on the eye during the retraction be focused in future works because the upward force may let the endoscope retrieve from the patient's eye.
The comparison between the force exerted on the eye in manual and robotic view adjustment is shown in Figure 16. The Eye Explorer reduced the maximum and mean forces exerted on the eye and fluctuation of force on the insertion hole. Furthermore, Table 2 shows that all indexes were improved by at least 15% for all test operators by using the Eye Explorer. Moreover, the endoscope view shown on the screen showed no noticeable vibration because of the vibration magnitude of the XYZ stage (0.005 mm at each axis), which is lower than that of a surgeon's hand (approximately 0.1 mm 2 ).
Therefore, the clinical requirement of safety is satisfied.
The force shown in Figure 16 is the sclera force during the operation, which is an important safety quantity in eye surgery. Table   2 shows that the average sclera forces are lower or near the safe threshold of 0.12 N. 21 However, the maximum forces are larger than 0.12 N. Note that the elasticity of the 3D printed eye used in this experiment is different from that of a real eye. The elasticity of tissue may reduce the maximum forces, which will be confirmed in future works. Moreover, the sclera force is related to the motion velocity of this device, which can be reduced whenever necessary.
Regarding the operation efficiency, the comparison result of the operation time only considered the process of the endoscope manipulation, not the total operation time of eye surgery. Now that this device does not impair the efficiency of the endoscope manipulation, and the overall operational efficiency is believed to be improved by considering the improved dexterity of the dual-hand operation realized by the Eye Explorer.

| CONCLUSION AND FUTURE WORKS
In this study, we developed a novel robotic endoscope holder for conducting eye surgeries named Eye Explorer that can hold an endoscope instead of the surgeon's hand. Using the Eye Explorer, the endoscope view can be adjusted without using the surgeon's hand.
Hence, this device is expected to enable a dual-hand operation during endoscopic eye surgeries and popularize the endoscopic eye surgery.
The experimental results demonstrated that the Eye Explorer satisfies the engineering requirements. Here, the mechanism makes it easy to place a sterilized cover on the robotic arm and separate the non-sterilizable part (endoscope) from the sterilizable part (holder unit). The lightweight mechanical design allows the operator to manipulate this device in the passive mode using forces that are lower than 0.5 N.
The device satisfied the established clinical requirements. The size of the base was less than two-third of the conventional instruments, such as the surgical table. Furthermore, the observable range of the endoscope is larger than that required in both the vertical and horizontal perspectives, ensuring that the endoscope view covered 70% of the intraocular area. The motion of the endoscope holder is not likely to interfere with the motion of the surgeon's hands and microscope. The safety measures were also considered, with the self-weight compensation mechanism, preventing the needle-like endoscope from falling and avoiding unintended damage to the retina. Moreover, the external load exerted on the eyeball when adjusting the endoscope view was considerably reduced.
Furthermore, the efficiency of the endoscope view adjustment by using this device did not differ from that of manual manipulation.
In future work, we intend to make several improvements to this device for clinical applications. First, an automatic instrument tracking system will be incorporated into the Eye Explorer to realize a fully automated assistive function and improve the direct user control interface. Second, we will consider more safety measures, including emergency measures when unexpected movement between the patient's eye and the endoscope happens (the patient's head moves or a large knocking is exerted on the arm unit), endoscope insertion depth control, 23 workspace registration, and a motion restriction function. Furthermore, we will evaluate the leaning process of using this device by analysing the feedback from real surgeons.
F I G U R E 1 7 Comparison of the time spent to complete the view adjustment task in the manual and robotic operation methods