Abstract

Robotics has made rapid inroads into specialised fields of surgery like neurosurgery. The robot has many advantages like increasing accuracy, eliminating muscle fatigue and physiological tremor during long operations, all of which improve outcomes and avoid potential complications in high risk neurosurgery procedures. At present, we use pre-programmed robots to guide the surgeon accurately localising the brain anatomy. This allows the surgeon to place electrodes in the brain or pedicle screws in the spine or take precise biopsies. Use of robots in Neurosurgery are presently limited to these tasks. In future, robots will have to perform these tasks and other complex procedures without the involvement of the surgeon.

For this to happen, robots will need be equipped with algorithms and software to help them to learn without being programmed, that is, via machine learning. At the core of this process is the ability of the robots to exploit large amounts of data based on their computational power and then translate it to actions that would mimic the surgeon. Progress on this front has been slow, mainly due to ethical (data access) and safety issues. It is felt that with the ability of newer generation computers to absorb and analyse tremendous amounts of data (e.g. thousands of operations of a particular kind), machines will learn to copy the actions of the surgeon and make safe choices so they can be trusted to perform the surgery and manage any unforeseen events which may arise from the surgery.

The two main challenges that hinder complete automation are surgical perception and tissue manipulation. Neurosurgery operates within severe spatial constraints and the consequence of any tissue injury is likely to be catastrophic. The tactical nous and fine sense acquired by an experienced surgeon in his hands can be learned by the machine through the development of haptic feedback through tissue resistance. Long distance robotic surgery supervised by a distant surgeon (Telesurgery) can only be safe if used over a dedicated network. This will ensure telecommunication is uninterrupted during the procedure and information exchange is concurrent. The risk to safety increases significantly with latencies above 200ms and exponentially above 1000ms.
In this article, we discuss the current situation, the future possibilities and the factors which are an obstacle to the quick adoption of AI and ML in neurosurgery.

Machine learning in surgery, past, present and the future

The potential for machines and other artificial forms of intelligence such as robots to enhance the ability of the surgeon to perceive, act, and extend their capabilities beyond current limitations has been a major contributor to the ever-increasing presence of robotics in neurosurgery and surgery as a whole.1 Nevertheless, the complete automation of the robots is a significant challenge, principally due to the problems associated with machine learning. Surgeons traditionally rely on their experience and expertise to perform operations; in contrast, robots and other artificial intelligence forms must be equipped with algorithms and software to help them to learn without being programmed, that is, via machine learning. At the core of the process is the ability of the robots to exploit data based on their computational power and translate it to actions that would mimic the typical surgical procedure performed by a human being, via the algorithms that make them equipped to postulate various problem-solving models.2

Globally, the last few decades have seen the invention, integration, and incorporation of various machines into the practice of neurosurgery. For example, common devices used in the last thirty years in neurosurgery include, the PUMA 200, NeuroMate, Pathfinder, Neuroarm, Spine assist, Renaissance, the Steady hand system, Neurolocate, iArms, EXPERT System, iSYS1 Robot, Spinal robotics, Augmented Reality systems, Neurosurgical lasers, the Da Vinci robot, SOCRATES and ROSA. Clinically, each robot has different applications in surgery. For example, the earliest robot, the PUMA 200, aids in performing of stereotactic surgery, where surgeons use CT guided biopsy needle to access tissues in the brain; Neuromate is a stereotactic system that is useful in doing various deep brain procedures such as stereo-encephalography, with six degrees of freedom and is considerably safer than the PUMA 200 for biopsies in surgery; Renaissance is a relatively newer image-guided system that is applicable in keyhole neurosurgery and uses an automated system that relies on MRI/CT scan images to insert needles, catheters, and drill the skull; The iArmS is equally a later invention that follows the movement of the surgeon. In operations, it prevents fatigue and trembling during microscopic procedures.3 As a result, it is beneficial in long and time-consuming procedures which are typical of neurosurgery.

However, to date, no completely automated machines have been deployed in the field of neurosurgery. Instead, optimisation of present technology with an increased master-slave relationship between human and machines has dominated present integration of artificial intelligence into neurosurgery.4 Nonetheless, it is critical to note that modern technology has allowed for the use of various miniaturised systems uniquely designed to serve specific elements in operations. The outcome evident with robots such as the neuroMate, SpineAssist, Renaissance, steady hand system, and Neurolocate indicate a bright future for machine learning and use in the field of neurosurgery.3

The current thought is that, although the industry will likely continue to experience a growth in adoption of robots in surgery, complete automation will not happen over the next two decades.5 It is argued that, at present, two main challenges hinder complete automation that would allow patients to have surgeries in hospitals without any human interaction. They include perception and manipulation. More than any other field, neurosurgery operates within severe spatial constraints and the potential consequences for even minor deviations may be catastrophic. For full automation to be realised, machines would have to learn to analyse the digital data such as images about soft tissues and then act on it appropriately without the aid of a surgeon.6 Accordingly, it is logical to expect that although machine learning will allow for greater inventions with more processing capacity and independence levels, the near future of artificial intelligence in neurosurgery will still include the participation of surgeons as either observers or active participants, or to rescue and limit brain injury when there has been an unforeseen event during surgery. An example of such a situation would be the ability to tackle unexpected injury to a blood vessel in the brain. The robot may cauterise the vessel and stop the bleeding. This however, may result in a stroke with profound neurological deficit or death for the patient. Where safety margins are low with potential serious consequences, the focus must be on prevention and safety.

Master-slave robot relationship (Telesurgery)

In this situation, the surgeon controls the robot. The introduction of the da Vinci surgical system marked a new generation of robots, especially since it had seven degrees of freedom and four arms, making it distinct from its predecessors. It revolutionised robotic surgery through a functional design improvement of previous machines.7

With modern development, the relationship between machines and surgeons has also progressed. Typically, robotics is classified either according to their level of autonomy or functional design. In this vein, robotic systems applicable in neurosurgery can be broadly categorised into three (handheld shared/controlled systems, the telesurgical, and the supervisory surgeon-controlled robot), according to the level and relationship between machines and the surgeons. For instance, telesurgery robots employ a master-slave relationship where the surgeon takes control of a surgery by controlling the robot’s actions remotely. The NeuroArm is one of the most applied tele-surgical robots and was developed in 2001 as a refined master-slave system to aid in neurosurgery, which allows surgeons to perform microsurgical and stereotactic procedures using data collected from real-time MRI and is structurally designed to withstand strong magnetic fields of MRI without altering the procedure quality.8 The system can also perform functions such as cutting, needle insertion, irrigation, and microsurgical cauterisation.

Safety issues in robot assisted surgery; focus on telesurgery

However, the safety of telesurgery is dependent on the experience of the surgeon and the performance capacity of the machine used. For instance, a study investigated the efficacy of the use of telesurgery in the removal of a phantom pituitary tumour in Nashville, by controlling a robot that was located about 800 kilometres away from the hospital. With a video latency of less than 100ms for the robot, all the surgeons involved in the operation gave it a perfect subjective safety score. However, they also noted that the operation used a dedicated network to ensure telecommunication is uninterrupted during the duration of the operation.9 This is important, because intrinsic latency in telecommunication network determines the potential for risks, slips in operation, and other safety issues, even with expert surgeons, as a delay in relaying real-time video data could result in an incision made earlier or later than expected.10 Whilst researchers note that expert surgeons could adapt to the delays, the safety of telesurgery deteriorates mildly at latencies of up to 200ms and increases exponentially to become fatally unsafe at above 1000ms.11,12 Thus, prevention of damage to the nervous system during telesurgery require a dedicated high capacity network to ensure videos are displayed within the latency of 0-100ms and the surgeon is skilled in the procedure.

Safety mechanisms through machine learning: perception, haptic feedback, collision detection

The surgeon’s perception is a major challenge to surgical safety. Typically, most experienced surgeons use the tactical sense between their instruments and the bones or tissue to guide them in neurosurgery. Yet the use of the sense of touch is eliminated with telesurgery and requires the surgical procedure to be steered purely with vision. Especially for the beginner, but also during complex procedures, this is an aspect that could increase the risk of slips or misdirection of the robot. How then, can this setback be overcome?

Although presently the technology is in the testing and experimentation state, haptic feedback should be integrated into telesurgery to address the issue of substituted contact for pure vision.6 In practice, the haptic feedback is achievable through a combination of collision detection algorithms for the calculation of depth penetration, and through coordinate transformation between the various systems in the machine that is translated into a virtual wall which the surgeon can visualise on a monitor. Presently, the technology allows robots to have an ‘intrinsic’ force or deflection-based sensing mechanism that mimics haptic feedback, with the development of robots such as SIROMAN system. This uses haptic guided telesurgery for tumour removal. The robot was developed to allow surgeons to access tumour tissue located deep in the brain and remove it with minimal damage by the creation of a virtual wall-based haptic guidance. Haptic mechanisms restrict the autonomy of surgeons by solely allowing machine movement within the virtual walls. In other words, machine learning has made it potentially possible for modern robots to perform surgeries with the same accuracy as human beings. Thus, in the next two decades, higher safety standards could be guaranteed in telesurgery.

Robots in neurosurgery in the near future

The role and function of robots in neurosurgery in the future is dependent on the advancements of main stakeholders such as engineers, healthcare administrators, surgeons, entrepreneurs, and the public. Although the pace of technological development could theoretically cause the ultimate replacement of surgeons, it is unlikely to happen soon.13 The fact that the potential interaction between the computer, the patient, and surgeon has not been exploited, is mainly due to ethical and safety issues, which will also delay the replacement of the human factor in operations in the near future.14 Even if biomedical engineers develop fully automated robots that can replace surgeons, the failure to change the public perception about the safety issues associated with machines compared to the decision-making processes of surgeons as human beings, could obstruct their adoption in hospitals;15 furthermore, with the development of fully autonomous systems, fewer practicing surgeons have updated their practice and knowledge to be able to operate the machines currently, a trend that could hinder complete automation in the future.16 Moreover, existing regulatory bodies in neurosurgery focus on the human behaviour rather than the success rate of robots, chiefly due to the difficulty involved in defining and classifying robotics. This signifies that surgeons will remain relevant in operations for the foreseeable future, with the next two decades witnessing a change in neurosurgery founded on the surgeon-robot-patient axis, with each stakeholder assuming joint custodianship of surgical operations.

References

  1. Senders JT, Zaki MM, Karhade AV, Chang B, Gormley WB, Broekman ML, Smith TR, Arnaout 0. (2018). An introduction and overview of machine learning in neurosurgical care. Acta Neurochir (Wien). 2018;160(1):29-38.
  2. Mattei TA, Rodriguez AH, Sambhara D, Mendel E. Current state-of-the-art and future perspectives of robotic technology in neurosurgery. Neurosurg Rev. 2014;37(3):357-366; discussion 366.
  3. Devito DP, Kaplan L, Dietl R, Pfeiffer M, Horne D, Silberstein B, Hardenbrook M, Kiriyanthan G, Barzilay Y, Bruskin A, Sackerer D, Alexandrovsky V, Stuer C, Burger R, Maeurer J, Donald GD, Schoenmayr R, Friedlander A, Knoller N, Schmieder K, Pechlivanis I, Kim S, Meyer B, Shoham M. Clinical acceptance and accuracy assessment of spinal implants guided with SpineAssist surgical robot: retrospective study. Spine (Phila Pa 1976) 2010;35(24): 2109-2115.
  4. Madhavan K, Kolcun JPG, Chieng LO, Wang MY. Augmented-reality integrated robotics in neurosurgery: are we there yet? Neurosurg Focus 2017;42(5):E3.
  5. Anna S. Will the machines take over surgery? The Bulletin of the Royal College of Surgeons of England 2017;99(3):88-90.
  6. Seung S, Choi H, Jang J, Kim YS, Park JO, Park S, Ko SY. Virtual wall-based haptic-guided teleoperated surgical robotic system for single-port brain tumor removal surgery. Proc Inst Mech Eng H 2017;231(1):3-19.
  7. Theodore N, Arnold PM, Mehta AI. (2018). Introduction: the rise of the robots in spinal surgery. Neurosurg Focus 2018;45(VideoSuppl1):Intro.
  8. Sutherland GR, Lama S, Gan LS, Wolfsberger S, Zareinia K. (2013). Merging machines with microsurgery: clinical experience with neuroArm. J Neurosurg 2013;118(3):521-529.
  9. Wirz R, Torres LG, Swaney PJ, Gilbert H, Alterovitz R, Webster RJ, 3rd. Weaver KD, Russell PT, 3rd. An experimental feasibility study on robotic endonasal telesurgery. Neurosurgery 2015;76(4):479-484;discussion 484.
  10. Hung AJ, Chen J, Shah A, Gill IS. Telementoring and Telesurgery for Minimally Invasive Procedures. J Urol 2018;199(2):355-369.
  11. Anvari M, McKinley C, Stein H. Establishment of the world’s first telerobotic remote surgical service: for provision of advanced laparoscopic surgery in a rural community. Ann Surg. 2005;241(3):460-464.
  12. Sebajang H, Trudeau P, Dougall A, Hegge S, McKinley C, Anvari M. The role of telementoring and telerobotic assistance in the provision of laparoscopic colorectal surgery in rural areas. Surg Endosc 2006;20(9):1389-1393.
  13. Mirnezami R, Ahmed A. (2018). Surgery 3.0, artificial intelligence and the next-generation surgeon. Br J Surg. 2018;105(5):463-465.
  14. Weinstein RS, Lopez AM, Joseph BA, Erps KA, Holcomb M, Barker GP, Krupinski EA. (2014). Telemedicine, telehealth, and mobile health applications that work: opportunities and barriers. Am J Med. 2014;127(3):183-187.
  15. Wehner M, Truby RL, Fitzgerald DJ, Mosadegh B, Whitesides GM, Lewis JA,Wood RJ. An integrated design and fabrication strategy for entirely soft, autonomous robots. Nature. 2016;536(7617):451-455.
  16. Miehle J, Ostler D, Gerstenlauer N, Minker W. (2017). The next step:intelligent digital assistance for clinical operating rooms. Innovative Surgical Sciences. 2017;2(3):159-161. Retrieved 6 Jul. 2018, from doi:10.1515/iss-2017-0034.