Autonomous Robotic Systems in Otolaryngology-Head and Neck Surgery

INTRODUCTION

Surgery, one of the most ancient forms of medicine, is also at the forefront of technological development and possibility with the novel use of robotic systems in the last 3 decades. The introduction of robotic assistance for medicine is seen in a variety of settings, from rehabilitation treatment to image guidance, to the operating theater.¹ In the field of otolaryngology-head and neck surgery (OHNS), transoral robotic surgery (TORS) with the da Vinci system has shown clear benefits in efficiency, patient outcomes, and expenditure compared to open methods and encouraged OHNS surgeons to incorporate these systems in their routine care.² As the interface between medicine and technology grows, and the burgeoning field of artificial intelligence (AI) crosses industry lines, surgeons are eager to understand what may be on the horizon in terms of robotic assistance in surgery. It is increasingly important to characterize the current roles of robotic assistance in OHNS within modern definitions and identify sectors with potential for improved utilization. This review will describe the current state of robotics in medicine and OHNS as well as future developments toward autonomous robotic systems.

BACKGROUND

The earliest record of a mechanical creation that could potentially step into a human role came with Leonardo da Vinci who drew metal-plated mechanical knights containing gears to create motion.³ The term robot originated from the 1920s’ Czech play entitled “Rossom’s Universal Robots,” morphing the Czech word “robota” meaning labor.⁴ In 1972, robots were given a standardized definition by the Robot Institute of America as a “reprogrammable, multifunctional manipulator designed to move materials, parts, tools or specialized devices through variable programmed motions for the performance of a variety of tasks.” In 1992, the MIT-Manus was unveiled as the first robotic device designed particularly for rehabilitation. Users place their arm into a brace and the machine will guide them to complete simple tasks with varying amounts of force depending on the patient’s ability.^5,6

Surgery presents the unique opportunity for a robot to combine the physical and mental acumen of a surgeon into automated actions. Nearly 2 decades ago, Dr Russ Taylor delineated the goals of surgical robots: to improve surgeons’ technical capability and promote surgical safety.⁷ Robotic assistance debuted in various fields in the 1980s starting with orthopedics (Arthrobot, 1983), neurosurgery (Programmable Universal Machine for Assembly, 1985), and urology (for transuretheral resection of prostate, 1988).³ The introduction of more autonomous robotic systems began with the ROBODOC for orthopedics in 1992, which assisted the surgeon in planning their entry orientation.⁸ In 1994, Computer Motion Inc, a group of researchers supported by a NASA grant, received Food and Drug Administration (FDA) approval for their Automated Endoscopic System for Optimal Positioning (AESOP) surgical system. The first of its kind to replace a human endoscope holder, it was controlled by foot pedals and voice commands and represented the true synergy that can be achieved between surgeons and robots in the operating room.⁹ However, in 1998, Intuitive Surgery Inc introduced the current monolith of robotic surgery, the da Vinci surgical system.

Most recent da Vinci models are made of 3 surgical instrument arms and 1 camera arm. da Vinci understands that surgeons can only operate as well as they can see and expands the ability to operate around corners with a custom endoscope for full stereotactic vision and high resolution. Containing discrete channels of viewing for each eye and safety checks to prevent inappropriate manipulation, the da Vinci system debuted in cardiothoracic surgery.¹⁰ In the following 2 years, da Vinci was applied to all general surgery procedures and received FDA approval.¹¹ The precise wrist movement with 7° of freedom, or axes of motion, encouraged OHNS surgeons at the University of Pennsylvania to pilot da Vinci for the first otolaryngologic procedure in 2005, termed TORS.¹² TORS utilizes da Vinci’s visualization abilities to access structures deep in the pharynx through the natural oral orifice. Since 2005, TORS has helped with oropharyngeal, laryngeal, and tongue base tumor resections; surgeries for obstructive sleep apnea; thyroid resections; neck and salivary gland dissections, and even select neuro-otologic applications.^13,14

The da Vinci system has since grown to be the standard of care in high-volume hospital systems; however, as an assistive robot, it cannot complete surgical tasks autonomously. Beyaz and colleagues⁸ described a distinction between the array of robotic systems currently available for surgical use: haptic versus active. Haptic systems are surgeon-guided and still require the mastery of a trained surgeon; like da Vinci, they are meant to improve surgical performance. Active systems can be considered autonomous and can follow a preset plan of operations executed by the robot with minimal to no input from a surgeon. These robots still require the surgeon to direct advancement from one stage of surgery to the next, identify target tissue, and distinguish between diseased and healthy anatomy. It is in this sphere that robotic systems have the greatest potential to grow into being a fully functioning, autonomous system.

LEVELS OF SURGICAL AUTONOMY

Current development of surgical robots focuses on creating semiautonomous machines that can independently perform discrete tasks under the purview and guidance of the surgeon, ultimately to enhance, not replace, the surgeon’s existing capability. The field of surgery is nascent in incorporating autonomous systems. There is a level of hesitancy from surgeons and patients with the concept of a robot stepping into the role of a surgeon.¹⁵ A recently published survey of American Rhinologic Society members found that over 80% of respondents would involve AI for clinical decision-making and diagnostics, such as using AI to analyze radiographic images. The use of AI for direct treatments like surgery was polarizing for the cohort, with only 50% interested in even exploring this possibility. However, select surgeons indicated their interest in using AI for preoperative planning and surgical navigation by segmenting images and recognition of tumor tissue and anatomy.^16,17

Analogous developments in the motor vehicle industry can demonstrate how AI could be used in surgery to improve efficiency and safety. For autonomous driving, there exist multiple levels of human versus machine control, ranging from the driver in full control of the car to a fully autonomous car requiring neither human hands nor vision, a level that has not yet been achieved. Within that spectrum, there are a variety of automated tools we have become accustomed to, such as lane assistance, which automatically detects roadway lanes and provides feedback to the car and driver to course-correct. Our representation of corollary levels of autonomy for robotic surgery in inspired by Yang and colleagues¹⁸, where level 1 is manual surgery, level 3 is the ability to complete tasks without direct surgeon intervention, and level 5 is a fully autonomous robotic surgeon—the material of science fiction (Fig. 1). At this stage, research and development of surgical robots remains at levels 2 to 3.

Fig. 1.

Levels of autonomy in robotic surgery.

Like lane assistance technology in motor vehicles, semiautonomous guidance in surgery would rely on the operator to perform the task while augmenting and improving performance through AI-guided identification of virtual fixtures, visual-tactile perception, and navigation. A semiautonomous system thereby requires less infrastructure and provides flexibility, as not every outcome needs to be programmed and individualized to each patient and procedure. Finally, if semiautonomous robotic guidance proves to be more acceptable to the general population, patient hesitance around surgical robots would be mitigated by the knowledge that the surgeon is the primary controller.

ARTIFICIAL INTELLIGENCE AND ROBOTICS IN OTOLARYNGOLOGY

For otolaryngology, the current state of AI is mainly split between oncologic, laryngologic, and neuro-otologic applications. A recent systematic review found that AI applications for head and neck cancer surgery primarily involved surgical margin assessment, complications assessment, and salvage surgery.¹⁹ The studies used machine learning to intraoperatively identify accurate margins to improve surgical efficiency or analyze large datasets to predict common postoperative adverse events and oncologic risk factors. Bur and colleagues²⁰ suggest implementing machine learning in high-stakes skull base surgery to label critical structures such as the carotid artery and optic nerve. In 2016, Carlson and colleagues²¹ used machine learning techniques to visualize glottic opening, eventually used in endotracheal intubation for real-time feedback to physicians. The AESOP surgical robot was the first AI platform that integrated the surgeon’s voice to control the position of a camera for endoscopic procedures.^22,23 Despite use in a variety of surgical procedures requiring endoscopy, including otolaryngology, widespread utilization is hindered by the need to constantly readjust the endoscope’s view.^24–26

Autonomous robots could alleviate this stress by having an intuition to follow the surgeon’s workflow and focus the camera on the appropriate area. In order for the AESOP to be incorporated with AI, the endoscope should be able to optimize the viewpoint without any guidance or readjustments from the surgeon. In 2008, the AutoLap (MST, Israel) system, which uses image analysis and computer-based instrument recognition to tag the surgeon’s instrument and shift the view based off the surgeon’s movement, was piloted for clinical use as the first foray into AI integration.²⁷ In 2020, a joint group from Italy and Johns Hopkins introduced the System for Camera Autonomous Navigation (SCAN) ability which tracks the tools and adjusts the endoscope position independent of surgeon guidance.^28,29 The researchers compared SCAN’s autonomous to manual camera control modalities, and found shorter task times with autonomous camera control mode and surgeons reported greater ease of use. As such, it appears that the future of automated endoscopic manipulation might be on the horizon.

ROSA machines (Medtech) are a series of platforms for orthopedic surgery that are seen as more autonomous extensions of the original ROBODOC for an arthroplasty and can perform the entire procedure without direct surgeon guidance.³⁰ Succeeding versions of this technology include the Robotic Arm Interactive Orthopedic System (RIO) (MAKO Surgical Corp), CASPAR, and NAVIO which all act as active-autonomous systems that are still controlled by the surgeon and have been documented to improve surgical outcomes.^31–34 The ROSA robot took these a step further as a semiautonomous system, FDA-approved in 2019, which calibrates itself to the femoral and tibial landmarks and identifies a “best scenario” surgical plan based on intraoperative measurements.³⁵ The ROSA confirms these measurements and proceeds with the planned incisions, all independently.³⁶ ROSA is a newer technology which has shown both improved patient outcomes with reduced operative time, yet still with dissatisfying accuracy in surgical planning.^36,37

ROSA ONE (Zimmer Biomet) is a stereotactic robot for neurosurgery that has been proven successful and comprises a robotic arm, touchscreen, robot stand, and navigating telescope arm.^38–40 The ROSA ONE uses preoperative images to register the surgical field and plan for multiple surgical paths. According to Zimmer Biomet, the ROSA ONE can ensure “instruments are placed in the planned area while avoiding critical structures.”⁴¹ They have developed robotic systems for both the brain and spine with the purpose of functional and stereotactic endoscopic open skull procedures or pedicle screw placement, respectively.^41,42 While the ROSA ONE effectively implements the concepts of registration and virtual fixtures, this does not qualify as an AI-driven surgical robot.

Applications of active-autonomous systems into otolaryngology are mainly in the field of cochlear implant (CI) insertion. The majority of CI surgery is vastly successful and follows a manual approach with a mastoidectomy bordering the facial nerve. Minimally invasive robotic assistance can minimize risks with the mastoid approach while preserving optimal outcomes seen with the traditional method.⁴³ The first robotic-assisted CI insertion used image guidance from a preoperative computed tomography (CT) scan for segmentation and path planning to facilitate stereotactic drilling. Labadie and colleagues⁴³ present the novel use of this robot on 7 patients with 6 successful insertions. Two years later, Caversaccio and colleagues⁴⁴ presented HEARO, a robotic platform with an arm as the first semiautonomous drill, for CI insertion. It is capable of segmenting critical anatomy and registration of surgical plan using surgical screws as references, and plan execution. Preoperative imaging was used to guide the robot and the safety of the drilling path was verified with intraoperative imaging and various sensors ensuring safe drilling. This system has since been proven as feasible and efficacious for CI insertion in patients, even those with complications.^45–47

iotaMotion (Iowa City, IA, USA) created the iotaSOFT Insertion System for electrode array insertions in CI surgeries consisting of a single-use sterile drive unit. iotaSOFT can help in preserving the delicate inner ear structures and minimizing damage to anatomy crucial for optimal hearing after insertion.^48,49 This robot has been especially helpful in inserting the electrode array at a slow and steady pace, something human surgeons are physiologically incapable of. iotaSOFT, however, requires significant surgeon involvement in locking the drive head tip to a location within 5 to 10 mm of the round window. The subsequent movement direction and speed in the inner ear is controlled entirely by a foot pedal operated by the surgeon, emphasizing the lack of robotic autonomy.⁵⁰

ARTIFICIAL INTELLIGENCE–DRIVEN ROBOTICS

The role of an AI-guided robot is not solely mechanical; rather, these systems can integrate information about the surgical field like anatomic location, feel, and compliance to enhance control of the robot. Machine learning algorithms have been developed to identify structures (image segmentation) for intraoperative guidance (image navigation) and improve haptic feedback to improve surgeon drilling ability.^51–53

It can be argued that the skull base is the ideal surgical field for the development of autonomous robotic tools. The skull base is difficult to access, containing important neurovascular and neurologic structures separated by mere millimeters, and pathologies in this space often require long surgeries and extensive training for competency. The potential consequences of a complication in skull base surgery that may arise from an injury are very high.⁵⁴ In comparison to other anatomic areas, the skull base is entirely fixed in bone, and anatomic structures do not move relative to one another, making surgical scene segmentation and image navigation easier to accomplish than in soft tissue surgery. In this space, we can independently address several problems in parallel to create a full robotic system for surgery.

Haptics is one of these problems. Robotics has resulted in haptic feedback being largely diminished or entirely removed from the hands of the surgeon. Haptics encompasses a number of different sensory inputs, including “kinesthetic sensation,” that is, where the body is relative to itself with (for example, joint angles in limb movement), and “tactile sensation,” which is information about the physical characteristics of objects outside of the body. In contrast to vision, haptic information regarding an object’s physical properties requires the body to come into contact with the object. Research has been focusing on restoring haptic information sensing that can detect tool to tissue forces and tissue deformation, augmenting the human ability to sense this information (for example, with a force sensor that is more sensitive to compliance changes than the human hand), and then presenting this information to a surgeon in real time without temporal delay. Robotic systems must be able to provide kinematic precision (tools move to the surgeon’s command accurately), kinetic fidelity (haptic information is accurately measured), and with minimal temporal lag. When this is achieved, we may be able to do force scaling and augment sensory information to feel, for example, the force involved in causing tip foldover in a CI electrode. We can also apply force limits to avoid errors, protect sensitive tissue, and alert a surgeon when critical thresholds are approached. We can also generate higher fidelity surgical simulators for training that can feel more like real surgery.

Largely regarded as one of the pitfalls of robotic-assisted surgery, lack of tactile and haptic sensation can result in tissue damage and comorbidities. Visual cues can only partially compensate for the reduction in kinetics.⁵⁵ Laparoscopic surgeons use visual cues in the absence of haptic feedback to estimate force on tissue. Endoscopic surgeons subconsciously estimate depth with small motions of the hand and camera despite a monocular field of view. In the field of orthopedics, it took nearly 15 years of development before introducing RIO, which was equipped with advanced haptic technology using tactile feedback from the robot to the surgeon. RIO has grown to be the most widely used robot for certain orthopedic procedures.⁵⁶ The RIO system uses preoperative CT to create a 3-dimensional model of the target bone and set a preset cutting zone for the robot. As such, final manipulation and actions are still determined by the surgeon who operates the robot. This preoperative planning ensures that the surgeon stays within the boundaries of the surgical zone while maintaining complete control over the tool. The haptic feedback allows the surgeon to use the robot as they would use a regular drill with the added advantage of having these preset anatomic guard rails. The robot can identify in real time where the surgeon is operating and will constrain the force-controlled tip of the instrument to only function within the predefined drilling zone or resection limits.^57,58 The concept of “no-go zones” is not novel for autonomous robots and has been discussed for nearly 2 decades.⁷

For otolaryngology, a combined engineering-OHNS team at the start-up Galen Robotics has developed a cooperative control robot platform that provides surgeons with a steady hand known as the robotic ENT microsurgery system (REMS).⁵⁹ The Galen REMS has 5° of freedom. Unlike the da Vinci system, which uses teleoperation with manual surgeon manipulation of the robotic arms outside the surgical field; cooperative control platforms like the REMS or RIO allow the surgeon to hold and operate with their instrument of choice in tandem with the robot. REMS can augment surgeon control with hand tremor cancellation, which can be crucial for the tight anatomic regions otolaryngologists operate in.⁶⁰ REMS also uses virtual fixtures to define the surgical space and prevent the surgeon from moving beyond that. Studies have demonstrated novices successfully performing a mastoidectomy in 5 trials using anatomic virtual fixtures enforced by REMS, with no training or coaching, in a mean time to completion of w3.6 minutes.⁶¹ This was a great step toward the implementation of semiautonomous robotic systems in surgery, as the operative time with fully autonomous procedures was reported to be around 90 minutes.⁶² A force-sensing drill has also been developed that can be used in conjunction with a cooperative robot such as REMS.⁶³ Force information can be integrated with positional data to enforce force thresholds that the surgeon cannot exceed when touching critical structures with haptic feedback.⁶⁴

In addition to providing safety barriers, another crucial area for AI-guided robotics is accurate and automated image segmentation using computer vision. Manually segmented imaging studies are used to train machine learning algorithms to automatically segment, localize, and track anatomy and objects on scans and video streams of surgery. The goal of an automated robot is to register imaging to anatomic structures accurately and consistently while quickly adjusting to new structures uncovered as the surgical view changes. Sinha and colleagues⁶⁵ presented an algorithm using AI machine learning techniques to register surrounding anatomy to submillimeter accuracies while conducting an active clinical endoscopy. This potentiates applications for structures without a reference preoperative CT scan.

Ding and colleagues⁶⁶ report a novel method for automated registration-based segmentation of temporal bone anatomy without the need for a preoperative CT scan. They created a standard template for the average temporal bone after compiling standard temporal bone CTs with salient anatomy labeled. Their findings presented a more rapid, automated process for segmentation of the temporal bone. Thus, patient outcomes benefit from the use of segmentation to develop virtual fixtures and protect vulnerable middle and inner ear structures. One year later, the same group published findings supporting the use of deep learning networks to autonomously segment temporal bone anatomy to expedite the preoperative planning process. The utilized deep learning neural network, nnU-Net, resulted in submillimeter accuracy for manual versus automated labeled structures of interest.⁶⁷ Automatic registration has since expanded to other fields using neural network–produced anatomic annotation. Intraoperative landmark localization was achieved for fluoroscopic navigation of the hip, suggesting that this technology, with further research and surgical validation studies to refine registration techniques can be used for image-guided otolaryngologic surgery in the skull base.^68,69

OTHER CONSIDERATIONS

With all these areas of burgeoning research, robotic surgery and semiautonomous systems can provide new tools on the horizon to enhance a surgeon’s performance. Ideally these advancements would also optimize efficiency in the operation room and reduce costs. A major consideration before implementing any robotic system is the cost to the health system, as robotic surgery is consistently more expensive than laparoscopic or open methods.⁷⁰ However, robotic assistance has proved its worth and established itself as a standard of care for certain centers. These hubs have welcomed the use of robots and found that for high case volumes, robotic assistance can improve postoperative quality of life with increased cost-effectiveness.⁷¹ For example, the cost of RIO orthopedic robot as a barrier to adoption. Although robotic hip and knee arthroplasty had higher fixed costs and longer operative times, a Medicare analysis showed cost-effectiveness when considering the lower revision rate, shorter length of stay, and fewer complications when compared to open surgery with implementation at high-volume centers.⁷²

In addition to cost, the environmental effects of robotic systems should be considered. Minimally invasive procedures contribute to global carbon emissions more than laparoscopic or open methods. Recent analyses show that robotic gynecologic procedures have around 40% greater greenhouse gas (GHG) emissions and create nearly 25% more waste than laparoscopic procedures.⁷³ The reimbursement model for many medical device companies in high-income countries relies on single-use instruments that contribute to a high carbon footprint. Robotic systems also use single-use instruments, yet costs can be higher due to the complexity of these instruments.⁷⁴ Analyses of cost-effectiveness and environmental impacts for autonomous procedures are still insufficient due to low surgical volumes, and research in this area is limited. A 2022 study analyzing potential GHG emissions for an AI diabetic eye examination found an 80% expense reduction as opposed to in-person eye examinations. The implications for surgery are vastly different; however, machine learning can be utilized to encourage the autonomous system to choose environmentally efficient options while still upholding a high quality of patient care. Furthermore, part of the purpose of implementing autonomous systems is to improve operative efficiency and patient outcomes, which can significantly reduce travel-related or complication-related carbon emissions.⁷⁵

SUMMARY

Semiautonomous robotic systems are an emerging technology with a potential for augmenting a surgeon’s capabilities. AI is being integrated into these robotic platforms to help perform higher level tasks. Several critical components of surgery are being simultaneously automated and refined, including surgical field segmentation, image segmentation, navigation and registration, tool tracking, and tool-to-tissue force sensing. These elements can enforce virtual boundaries, provide enhanced force sensing, and allow for customized, high-fidelity surgical simulations for planning and training. As these systems are increasingly refined, the machine-physician interface can become more seamless to enhance a surgeon’s abilities.

Fig. 1 and Video 1 detail the different stages that an autonomous surgical robot would progress through. Level 0 represents the lowest level of autonomy where the robot has no intrinsic ability to enhance the procedure outside of its mechanical purpose. Level 5 represents the highest level achievable for an autonomous system where the robot essentially does take the place of a surgeon. Current AI-driven robotics for surgery sit at level 1 or 2.

Supplementary Material

SUPPLEMENTARY DATA

Supplementary data to this article can be found online at https://doi.org/10.1016/j.otc.2024.05.004.

KEY POINTS.

• It is increasingly important to characterize the current roles of robotic assistance in otolaryngology-head and neck surgery (OHNS) within modern definitions and identify sectors with potential for improved utilization.

• The goal of incorporating autonomous systems is not so much to create robots to replace surgeons, but rather to aid the surgeon by providing improved efficiency, consistency, safety, and outcomes.

• Semiautonomous guidance in surgery would rely on the operator to perform the task while augmenting and improving performance with artificial intelligence (AI)–guided identification of virtual fixtures, visual-tactile perception, and navigation.

• The role of an AI-guided robot is not solely mechanical with holding tools or increasing stability; rather, these systems can integrate information about the surgical field like anatomic location, feel, and compliance to enhance control of the robot—aspects that can be well developed within the field of skull base surgery.

• Cost to the health system should be considered, as robotic surgery is consistently more expensive than laparoscopic or open methods, as well as environmental impacts since minimally invasive procedures contribute to global carbon emissions more than laparoscopic or open methods.

CLINICS CARE POINTS.

• Including AI into surgery is not meant to replace the surgeon but enhance the surgeon’s skill and support achievement of efficient and effective methods for optimal patient outcomes.

• Haptic feedback, virtual fixtures, force-feedback, tremor control, and automated image segmentation are all existing starting points for AI-driven robotics in OHNS, which serve to improve surgical outcomes.

• The current state of autonomous systems in surgery is far from being entirely independent. Rather, these machines are just starting to broach the level of being semiautonomous. This means the surgeon initiates and oversees the robot actions, which are limited to smaller tasks that it may have been instructed to do. The surgeon also maintains complete control over these actions and can take over or terminate the process at any time.