Artificial Intelligence-assisted Surgery

Final Report

Technical Report Writing ENG 1112[C]

Professor: Lynda Morrissey

Student Name: Niyol Jha

Student Number: 300247600

Date: December 20, 2022

Executive Summary

The purpose of this investigation was to determine whether robotic-assisted surgery is an

effective, efficient, and safe alternative to traditional surgical procedures.

The methodology consisted of secondary research only. Sources used to conduct research
included scientific journals and medical peer-reviewed articles. The results were categorized into
the following topics: robot-assisted surgery (RAS) machines, the efficiency of robotic surgery
compared with traditional surgical methods, deep and machine learning models in surgery, and
the risks and effects of technical failure in surgical robots.

Research studies conducted on robotic-assisted surgical (RAS) machines revealed that

RAS systems have been used to feasibly perform surgeries without compromising accuracy, with
operations of the da Vinci surgical system alone reaching 1.25 million global operations in 2020
(Moglia 2).

Research further revealed that RAS systems were mostly used to perform minimally
invasive surgery, a procedure involving a surgeon making tiny incisions in the abdomen and
pelvis to limit the amount of pain the patient may feel. This surgical procedure is also known as
laparoscopic surgery (Lee 2). In a study assessing AI Computer vision algorithms to understand
the content of digital video frames in surgery, an 83.3% accuracy level of the assessment of
surgical performance for the da Vinci surgical system was reported.

Since laparoscopic surgery involves making small, precise cuts, RAS systems were found
to have advantages in conducting these types of surgical procedures due to their motion scaling
abilities (Prasad 866). While analyzing the efficiency of surgical robots in comparison to
traditional surgical methods, the Hugo and Zeus RAS systems were found to apply motion
scaling: conversion of the surgeon’s hand movements into precise movements by the surgical
instrument.

In a study conducted to investigate whether motion scaling in a robotic-assisted platform

improved accuracy and enhanced performance when compared with a traditional surgical
platform, laparoscopic surgical drills were performed on manual surgical platforms and robot-
assisted platforms using the dominant hand first, and then the non-dominant hand. After
comparing the performance of the dominant hand against the non-dominant hand across both the
manual platforms and the robot-assisted platforms configured with motion scaling, significant
gains were observed in the performance of the non-dominant hand with robotic-assistance
(Prasad 867). This indicated that the addition of motion scaling could allow for improvement in
accuracy while allowing for ambidextrous utilization of the robotic-assisted platform, thereby
allowing an efficient and accurate surgical procedure.

The application of Machine and Deep learning models in surgery indicate that they can
accurately be used in the form of object-detection and tracking algorithms to identify digital
content in images and video. An example of such a model is a segmentation algorithm.
Segmentation algorithms are features of computer vision, an AI subfield closely related to DL
that attempts to reproduce the capability of human vision by understanding the content of digital
video frames in surgery (Moglia 14). In a study evaluating surgeons’ skills in performing robotic
surgery procedures, segmentation and tracking algorithms are used to identify the type of
surgical instrument and track its trajectory. The results of the study revealed that the DL tracking
algorithm determined the trajectories of surgical instruments during the minimally-invasive
procedure with an accuracy of 83% (Lee 11).

In analyzing the risks and effects of technical failure, the da Vinci, Hugo RAS, and Zeus
surgical systems investigated in this report were found to be safe. However, the criteria for
determining their safety varied. In a study to investigate technical failure in da Vinci systems, a
technical failure rate of 10.9% (37 occurrences out of 340 surgical procedures) was reported.
However, despite encountering a number of technical errors with the da Vinci system, it was
clinically tested, had low rates of technical failure and conversions to manual surgery, and
reported no direct patient injuries (Nayyar and Gupta 1713). The Hugo RAS system was also
reported to be safe in cadaveric models, but unlike the da Vinci system, it was tested in a
preclinical setting. This meant that further research in the clinical setting was needed to confirm
its safety in terms of technical failure and intraoperative complications.

While traditional surgical procedures are effective, the method of robot-assisted surgery
was also found to be an effective, efficient, and safe alternative to the former. This was
evidenced by the feasibility of RAS systems in performing minimally-invasive surgical
procedures without compromising accuracy, the addition of motion scaling in RAS systems to
enhance accuracy and efficiency over manual surgical methods, and the application of innovative
deep learning algorithms allowing for accurate detection of objects in surgical views. Despite
more research being needed to confirm the safety of novel RAS systems such as the Hugo RAS,
current RAS systems in operation like the da Vinci and Zeus surgical systems are modern, yet
safe alternatives for performing traditional minimally-invasive surgical procedures.

Introduction

In 1950, the mathematician Alan Turing pioneered the concept of using computers to
simulate intelligent behavior and critical thinking. Turing used a simple test, which later became
known as the “Turing test”, to establish whether computers were capable of human intelligence.
Six years later, computer scientist John McCarthy described the term artificial intelligence (AI)
as “the science and engineering of making intelligent machines”. Over several decades, AI has
advanced to include complex algorithms that can perform functions similar to the human brain
(Kaul 1).
 AI contains subfields such as machine learning (ML), deep learning (DL), and artificial
neural networks (ANN). ML is a prediction tool that identifies patterns to “learn” from a
particular situation. It applies that information dynamically to clinical decision-making to
customize patient care rather than follow a static algorithm. ML has evolved into what is now
commonly known as DL, a composition of algorithms that create an ANN. An ANN can learn
and make decisions on its own, similar to the human brain (Kaul 1).

The applications of AI in surgery can be dichotomized into two subtypes: virtual and
physical (Malik 1). Examples of virtual applications of AI in surgery can include DL-based
guidance in tracking surgical instruments in a complex surgical view (Lee). AI’s physical
applications include robots assisting in performing surgeries, intelligent prostheses for
handicapped people, and elderly care (Malik 1).

A common application of AI in surgery is the use of robotic-assisted surgery (RAS)

machines. Robotic machines are driven by AI using powerful computing systems and large
datasets to perform tasks that normally require human intervention, thus reducing human effort.
A Da Vinci surgical system is an example of a RAS machine. It incorporates rotating and
positioning surgical instruments using AI technology. The robotic arms of the system mimic a
surgeon's hand movements with better precision, containing magnification options that allow the
surgeon to perform minute incisions (Malik 2). The utilization of Da Vinci RAS machines
worldwide increased to 1.25 million in 2020 (Moglia 2).

An example of a novel RAS device is the HugoTM RAS system. It includes innovative
features such as four independent robotic arm-carts that improve the system’s modularity, and
customizable motion scaling settings. In robotic surgery, motion scaling converts the surgeon’s
hand movements into precise movements by the surgical instrument.

The advent of developments in deep learning has allowed significant features of AI in

surgery, such as object detection and tracking to advance. Object detection and tracking are key
features of an AI subfield known as Computer vision. These features are driven by the goal of
understanding the content of digital images and videos in surgery. Computer vision involves
developing methods that reproduce the abilities of human vision (Moglia 14).

The objective of this investigation was to determine if AI-driven robotic-assisted surgery

is an accurate, efficient alternative to traditional surgical procedures. Several topics were
examined to assess AI robotic-assisted surgery: robot-assisted surgery (RAS) machines, the
efficiency of surgical robots compared with traditional surgical procedures, deep and machine
learning models in surgery, and the risks and effects of technical failure in surgical robots.

There were a few constraints that affected the methodology of this report. The first
constraint was the lack of studies across multiple surgical fields researching the efficiency of
robotic surgery in contrast to the efficiency of traditional surgical procedures. As a result, one
study that compared robotic-assisted surgery and traditional surgery in a minimally-invasive
surgical procedure, and another study that investigated a novel RAS system was used to compare
efficiency. The second constraint was the pre-clinical nature of novel RAS machines such as the
Hugo RAS system, and the lack of research conducted in the clinical setting to confirm its safety
in terms of technical failure.

Methodology

The methodology of this report involved researching robot-assisted surgery (RAS)

machines, the efficiency of surgical robots compared with traditional surgical procedures, deep
and machine learning models in surgery, and the risks and effects of technical failure in surgical
robots. The methodology for this report involved secondary research only.

Robot-assisted surgery (RAS) machines such as the HugoTM RAS system and the Zeus
Robotic Surgical system, were researched through articles belonging to renowned peer-reviewed
medical journals such as BJU International, Journal of the American College of Surgeons, and
the International Journal of Surgery. These articles were used to define the capabilities of RAS
machines and provide examples of RAS systems to determine what level of accuracy, precision,
and efficiency they performed surgeries with.

The efficiencies of surgical robots compared with traditional surgical methods were
researched using articles belonging to the peer-reviewed medical journals BJU International and
the Journal of the American College of Surgeons. These articles were consulted for information
on whether surgical robots could improve on accuracy while operating on multiple surgical
targets. A peer-reviewed medical book was used to assess the definition of higher-quality surgery
for the purpose of determining whether robot-assisted surgery provided an enhanced surgical
quality

The examination of Machine learning (ML) and Deep learning (DL) models in robotic
surgery were provided by an article in the Journal of Clinical Medicine, a peer-reviewed medical
journal. Articles in the medical journals International Journal of Surgery and Gastrointestinal
Endoscopy were used to provide definitions of AI subfields like ML, DL, and Computer vision
pertaining to robotic surgery. Secondary definitions for the dichotomy of applications of AI were
determined using an article belonging to the Journal of Family Medicine and Primary Care.

The risks and effects of technical failure in surgical robots were researched through an
article in the medical journal BJU International. This article studied the technical problems
associated with RAS systems and their effects, and examined whether RAS systems were a safe
mode of surgical treatment.

Finally, the results were determined for each of these topics by investigating the
information pertaining to robot-assisted surgery and looking for key factors that distinguished
robot-assisted surgery from traditional surgical procedures.
Findings and Observations

Robot-assisted surgical machines

In the analysis of AI robotic-assisted surgery, the first point investigated was whether
robotic-assisted surgery (RAS) machines can feasibly perform surgeries without compromising
accuracy.

RAS machine systems have witnessed a significant evolution in the last two decades
(Sarchi 671). In a study conducted by the peer-reviewed medical journal BJU International, the
HugoTM RAS system was used to perform a robot-assisted radical prostatectomy (RARP) on a
cadaver, a robotic surgical procedure that entails removing the prostate glands (Sarchi 671).

The Hugo RAS system contained innovative features, such as four independent arm-
carts, and customizable motion scaling settings. The system’s independent arm-carts allowed for
an optimized setup of the working space in the operating room (Sarchi 672). The system’s
modularity created different tilt and docking angles for the arm-carts, optimizing access to the
surgical field (Sarchi 672).

Customizable motion scaling settings are a key feature of the Hugo RAS system. In
robotic surgery, motion scaling involves reducing large movements by the surgeon’s hand into
microscopic movements at the tips of the surgical instruments (Prasad 864). In a study conducted
by the Journal of the American College of Surgeons, the aim was to evaluate whether robotic-
assistance enhanced surgical precision and if motion scaling in a robotic-assisted platform
improved accuracy and enhanced performance when compared with a traditional surgical
platform (Prasad 864). The results of the study concluded that motion scaling could not only be
used to feasibly perform surgeries but also significantly improved accuracy over traditional
surgical instrumentation by 20% to 30% (Prasad 868).

However, the study conducted by BJU International also indicated that the Hugo RAS
system’s modular robotic platform could raise concerns about the event of a possible collision
between the independent robotic arms inside the abdomen. To feasibly perform the surgery,
proper orientation of the robotic arms using the tilt and docking angles is pivotal (Sarchi 674).
The Hugo RAS system was found to be safe and feasible in cadaveric models and was boosted
by accuracy-enhancing features such as motion scaling, and ‘pistol-grip’ handles that allow the
surgeon to move instruments with added stability without compromising lightness (Sarchi 674).

The Da Vinci surgical system is a popular RAS machine in the surgical field that operates
the movement of surgical instruments using AI technology, surpassing more than 1.25 million
operations globally in 2020 (Moglia 2). In a study conducted by International Journal of
Surgery, Da Vinci surgical systems were used to assess surgical performance during a pelvic
lymph node dissection (PLND) procedure. The system utilized AI Computer vision algorithms
such as Edge Detection and Segmentation. These algorithms were utilized to understand the
content of digital video frames in surgery (Moglia 8). In Computer vision, examples of
understanding the content of digital images may involve extracting a description from the image,
which may be an object or a text description (Moglia 14).
 Figure 1. Table showing an 83.3% accuracy level of the assessment of surgical performance for the da Vinci surgical system during a Pelvic
Lymph Node Dissection (PLND) surgical procedure (Moglia et al.).

The results from the study showed that the da Vinci RAS surgical system applied AI
Computer vision algorithms to assess the PLND surgical procedure with an 83.3% accuracy rate.
In the studies conducted by BJU International and the International Journal of Surgery, the
results indicated that RAS systems could be applied in surgical procedures while maintaining a
high degree of accuracy. The results also exhibited that RAS systems with motion scaling
capabilities could be used to enhance accuracy levels, exceeding the accuracy levels of
traditional surgical instrumentation by 20% to 30% (Prasad 868).

Efficiency of robotic-surgery compared with traditional surgical methods

The main focus of this topic was to determine whether surgical robots had a higher rate of
efficiency in conducting minimally invasive surgery than traditional surgical procedures. In the
case of surgical robots, efficiency is a measure of the ability of the robot to perform surgery by
operating on multiple surgical targets while maintaining a high degree of accuracy.

Minimally invasive surgery entails a surgeon making tiny incisions in the abdomen and
pelvis to limit the amount of pain the patient may feel. This procedure is traditionally known as
Laparoscopic surgery (Lee 2). In a study conducted by the Journal of the American College of
Surgeons, the Zeus Robotic Surgical System was used to conduct a microsurgical drill created to
simulate drills used in both manual laparoscopic and robotic-assisted laparoscopic surgery
(Prasad 867). In this drill, subjects used a needle locked into a piece of surgical equipment
known as a tissue grasper. The subjects used the needle from a fixed start position to pierce six
targets 1.2 centimetres away, alternating from the start position back to each successive target.
After completing the six targets with their dominant hand, the experiment was performed again
with their non-dominant hand (Prasad 865).

The drill was conducted with each hand thrice for two manual laparoscopic traditional
surgical platforms (M1 and M2) and three robotic-assisted laparoscopic platforms (R1, R2, R3).
configured with motion scaling (Prasad 865). In surgery, motion scaling converts the surgeon’s
hand movements into precise movements by the surgical instrument. The robotic-assisted
laparoscopic platforms also utilized tremor filtration, a method to remove hand tremors during
surgery. After comparing the performance of the dominant versus the non-dominant hand across
both manual and robot-assisted platforms, significant gains were observed in the performance of
the non-dominant hand with robotic-assistance (Prasad 867).
Figure 2. Table showing a comparison of the performance of the dominant versus the non-dominant hand across all manual laparoscopic surgical
platforms and robotic-assisted laparoscopic platforms (Prasad et al.).

The results revealed that advancements in motion scaling could allow for improvement in
accuracy without increasing surgery operating times, allowing for an efficient and accurate
surgical procedure (Prasad 868). In this case, the study suggested that the addition of motion
scaling to the robotic platform may reduce operative times over tremor filtration alone (Prasad
868). Motion scaling was found to equalize the performance of the dominant and non-dominant
hands, allowing for ambidextrous use of the robotic-assisted platform (Prasad 868). This means
that the ambidextrous utilization of the RAS platform could possibly allow for operation on
multiple surgical targets without compromising on operation time.

In another study, the Hugo RAS system allowed for modified docking angles of the
independent robotic arm-carts in a laparoscopic surgical procedure, thereby improving the ability
to reach different anatomical targets and suggesting an increase in surgical efficiency over
conventional laparoscopic procedures (Sarchi 674).

The results indicated that robotic laparoscopic platforms configured with motion scaling
were more efficient and accurate in comparison with manual laparoscopic platforms, as the
robot-assisted platforms maintained a higher degree of accuracy than the manual platforms while
exhibiting ambidextrous use, with an accuracy value of 0.80 compared to 0.03 for the manual
platform.

Deep and Machine learning models in Surgery

In the investigation of deep and machine learning models in surgery, the focus of the
investigation was to examine whether the application of AI algorithms in surgical procedures
could enhance surgical quality.

Machine learning (ML) and Deep learning (DL) are subfields of Artificial Intelligence.
ML is the application of specific traits to identify patterns that can be used to analyze a specific
situation. The machine then “learns” from and applies that information to future similar scenarios
so it can improve with experience (Kaul 1). DL is an advanced subdivision of ML composed of
algorithms that can learn and make decisions on their own, similar to the human brain (Kaul 1).

In a study conducted by the Journal of Clinical Medicine, DL was used to track the tips
of surgical instruments during a minimally-invasive laparoscopic surgical procedure. The study
used computer vision features such as segmentation and object-tracking. Segmentation and
tracking are algorithms of computer vision, an AI field closely related to DL that attempts to
reproduce the capability of human vision by understanding the content of digital video frames in
surgery (Moglia 14). In this study, segmentation and tracking are used to identify the type of
surgical instrument and track its trajectory (Lee 3).

Figure 3: Figure showing the qualitative results of the segmentation algorithm. The results show that each surgical instrument can be identified
using a different colour and tracked in multiple surgical views (Lee et al.).

The study revealed that when compared with traditional manual methods of tracking
surgical instruments, the DL tracking algorithm determined the trajectories of surgical
instruments during the minimally-invasive procedure with an accuracy of 83% (Lee 11).
However, the accuracy rate of traditional tracking methods for surgical instruments in
laparoscopic procedures was not provided. This result indicates that the use of AI algorithms can
deliver a higher-quality surgery for patients, as high-quality surgery is defined by greater
accuracy and a minimally invasive procedure, so that the amount of pain the patient may feel is
reduced (Grespan 8).

Risks and effects of technical failure in surgical robots

In analyzing the risks and effects of technical failure, the risks of robotic-assisted surgery
and problems associated with the use of RAS systems were investigated to determine whether
RAS systems were safe alternatives to conventional manual surgical procedures.
 In a study conducted by BJU International on technical failure in da Vinci RAS systems,
some risks of technical failure were found to originate from the machines themselves, such as
fused bulbs or locked robotic arms. However, other risks of technical failure might be related to
human-machine interaction, such as a surgeon overdoing a movement beyond the range of the
surgical instrument (Nayyar and Gupta 1712). Another risk of technical failure are problems that
can cause patient injury, thereby affecting the surgical procedure by making the surgical steps
harder to perform and increasing operative time as an effect (Nayyar and Gupta 1710).

This study utilized a da Vinci RAS system to help conduct a minimally-invasive

laparoscopic surgery and reported the number of technical problems. The study reported 23
robotic instrument-related technical problems out of 340 cases of robotic surgery (Nayyar and
Gupta 1711). This meant replacing the robotic instrument with a new instrument during the same
surgery, thus increasing the surgical operation time by 20 minutes as an effect.

The study presented 37 total occurrences of technical failure out of 340 surgical
procedures for a failure rate of 10.9% (Nayyar and Gupta 1712). However, 28 of those 37
occurrences could be corrected with additional operating time and total conversions back to
manual laparoscopic surgery due to technical robotic failure were 2 out of 340 surgeries, for a
rate of 0.6% (Nayyar and Gupta 1712). The study found that no complications of the RAS
system arose, and none of the patients were directly harmed as a result of the technical faults in
the robotic system (Nayyar and Gupta 1712).

In a review conducted by the FDA (Federal Department of Agriculture) on surgeries

carried out by surgical robots, 189 technical failures were reported out of 50000 surgeries for a
device failure rate of 0.38% (Nayyar and Gupta 1712). The failure rates of the da Vinci and Zeus
RAS systems investigated in this report were included in the review. However, the study
concluded that despite the number of technical errors encountered, the da Vinci RAS system
remained a safe mode of treatment in surgery, with low technical failure and manual conversion
rates, and no reported direct patient injury (Nayyar and Gupta 1713).

In a separate study, the operation of the Hugo RAS system was determined to be safe for
use in cadaveric models, with no reports of instrument collision, intraoperative complications, or
technical system failure (Sarchi 674). However, the pre-clinical nature of the Hugo RAS system
means that further research in the clinical setting is needed to confirm its safety in terms of
technical failure and intraoperative complications.

Conclusions

The applications of AI within surgery have driven a rise in robotic surgical technology.
With the emergence of the robotic da Vinci surgical system that includes applications such as
moving robotic arms with motion scaling abilities that mimic precise human movement, and
deep learning algorithms for the purpose of detecting content from video and images that are
meant to simulate human vision, it is clear AI has advanced over several decades to perform
functions similar to the human brain.
 In this report, the applications of robot-assisted surgery were discussed to determine if it
was an effective and safe alternative to traditional surgical procedures. The results of this
investigation were divided into four categories: The use of robotic-assisted surgery (RAS)
machines, the efficiency of robotic-surgery compared with traditional surgical methods, Deep
and Machine learning models in surgery, and the risks and effects of technical failure in surgical
robots.

The observations revealed that robotic-assisted surgery (RAS) machines could be used to
feasibly perform surgeries without compromising accuracy. Popular RAS machines in use such
as the da Vinci and Zeus surgical systems were found to include features such as motion scaling,
which provided significant gains in accuracy during minimally-invasive (laparoscopic) surgical
procedures. As a result, due to their gains in accuracy and precision over traditional manual
laparoscopic surgical procedures, RAS systems are primarily used to perform laparoscopic
surgery. Laparoscopic surgery involves making precise, tiny cuts in the abdomen and pelvis to
limit the amount of pain post-surgery, thereby increasing recovery time for the patient.

Furthermore, RAS systems equipped with motion scaling were found to equalize the
performance of the dominant and non-dominant hands during surgery, granting ambidextrous use
of the robotic-assisted platform which can increase surgical efficiency by potentially allowing
operation on different surgical targets. Additionally, advances in motion scaling could allow
further improvements in accuracy without compromising operation time, providing an accurate
and efficient surgical procedure. The application of Deep learning algorithms can help surgical
robots learn and identify objects in images, and use pattern recognition to predict and digitally
track objects in surgical procedures with a high rate of accuracy.

While the safety of the patient emerges from the correct interaction between surgical
robots and humans, in regards to robot-assisted surgical (RAS) systems, the rate of technical
failure is low, with no direct patient injury attributed to RAS systems reported in any of the
studies researched in this investigation.

The investigation of this report concludes that while traditional surgical procedures are
effective, the method of AI-driven robot-assisted surgery was also found to be an effective,
efficient, and safe alternative to the former. This was evidenced by the feasibility of RAS
systems in performing minimally-invasive surgical procedures without compromising accuracy,
the addition of motion scaling in RAS systems to enhance accuracy and efficiency over manual
surgical methods, and the application of innovative deep learning algorithms allowing for
accurate detection of objects in surgical views. Despite more clinical research being needed to
confirm the safety of novel RAS systems such as the Hugo RAS, current RAS systems in
operation like the da Vinci and Zeus surgical systems are modern, yet safe alternatives for
performing traditional minimally-invasive surgical procedures.
 Works Cited

Grespan, Lorenzo., et al. The Route to Patient Safety in Robotic Surgery. 1st ed. 2019., Springer

International Publishing, 2019.

Kaul, Vivek, et al. “History of Artificial Intelligence in Medicine.” Gastrointestinal Endoscopy,

vol. 92, no. 4, 2020, pp. 807–12.

Lee, Dongheon, et al. “Evaluation of Surgical Skills During Robotic Surgery by Deep Learning-

Based Multiple Surgical Instrument Tracking in Training and Actual Operations.”

Journal of Clinical Medicine, vol. 9, no. 6, 2020, p. 1964–.

Malik, Paras, et al. “Overview of Artificial Intelligence in Medicine.” Journal of Family

Medicine and Primary Care, vol. 8, no. 7, 2019, pp. 2328–31.

Moglia, Andrea, et al. “A Systematic Review on Artificial Intelligence in Robot-Assisted

Surgery.” International Journal of Surgery (London, England), vol. 95, 2021, pp.

106151–106151.

Nayyar, Rishi, and Narmada P. Gupta. “Critical Appraisal of Technical Problems with Robotic

Urological Surgery.” BJU International, vol. 105, no. 12, 2010, pp. 1710–13.

Prasad, Sunil M., et al. “Surgical Robotics: Impact of Motion Scaling on Task Performance.”

Journal of the American College of Surgeons, vol. 199, no. 6, 2004, pp. 863–68.

Sarchi, Luca, et al. “Robot‐assisted Radical Prostatectomy Feasibility and Setting with the

HugoTM Robot‐assisted Surgery System.” BJU International, vol. 130, no. 5, 2022, pp.

671–75.