Journal of ISAKOS 9 (2024) 227–233

Contents lists available at ScienceDirect

Journal of ISAKOS
journal homepage: www.elsevier.com/locate/jisakos

Current Concepts Review

Artificial intelligence and the orthopaedic surgeon: A review of the

literature and potential applications for future practice: Current concepts
Al-Achraf Khoriati, MBBS, BSc, FRCS (Tr and Orth) a, Zuhaib Shahid, MBBS, MRCS a,
Margaret Fok, MBChB (Otago), FRCSEd (Ortho), FHKAM (Orthopedic Surgery) b, c,
Rachel M. Frank, MD d, 1, Andreas Voss e, Pieter D'Hooghe f, 2, Mohamed A. Imam, MD PhD
FRCS a, g, *
a
Rowley Bristow Orthopaedic Centre, Ashford and St Peter's NHS Foundation Trust, Chertsey, KT106PZ, UK
b
Department of Orthopaedics and Traumatology, Queen Mary Hospital, The University of Hong Kong, Pok Fu Lam Rd, High West, Hong Kong, China
c
Asia Pacific Orthopaedic Association, 57000, Malaysia
d
Department of Orthopaedic Surgery, Joint Preservation Program, University of Colorado School of Medicine, 12631 E 17th Ave, Mail Stop B202, Aurora, CO 80045, USA
e
Sporthopaedicum Regensburg, Street, Hildegard-von-Bingen-Straße 1, 93053, Regensburg, Germany
f
Aspetar Orthopedic and Sports Medicine Hospital, Aspire Zone, Sportscity Street 1, P.O. Box 29222, Doha, Qatar
g
Smart Health Centre, University of East London, University Way, London, E16 2RD, United Kingdom

Current Concepts

Artificial Intelligence in medicine (AIM) and surgery (AIS) should be considered as two related but distinct entities. The combined use of AI and
deep learning along with human interpretation for automated measurements in orthopaedics yields excellent results.

AI has been used successfully to facilitate decision making when it comes to prognostication. These models still require human oversight due to
the complex nature and variables involved.

Robotic-assisted arthroplasty improves implant positioning in both hip and knee arthroplasty, there is less conclusive evidence to support
improvement in functional outcomes or long-term survival of these implants.

Simulation technology is on the rise and is has been increasingly used as an adjunct to traditional models. These models cannot be used as a
substitute to traditional training.

Future Perspectives

Abstract concepts such as intuition which are difficult to impart to a machine in the form of computer code remain elusive and further work is
needed to refine these processes to a point where human oversight is minimal or redundant.

AI driven prognostication models remain in their infancy. More work is needed to guide treatment pathways and formulate strategies to guide
preventative medicine.

Whilst robotic assisted surgery and Virtual reality has improved surgery in numerous domains, this has not yet translated to an improvement in
patient outcomes. Until these are achieved, further development may be required into the optimisation of these technologies.

* Corresponding author. Rowley Bristow Orthopaedic Unit, Ashford and St Peter's Hospitals, UK.
E-mail addresses: a.khoriati@nhs.net (A.-A. Khoriati), Zuhaib.shahid@nhs.net (Z. Shahid), margaret_fok@yahoo.com (M. Fok), rachel.frank@cuanschutz.edu
(R.M. Frank), voss@sporthopaedicum.de (A. Voss), m.imam1@nhs.net (M.A. Imam).
1
Tel.: þ1 303 724 2927; fax: þ1 303 724 1593.
2
Tel.: þ974 66890368.

https://doi.org/10.1016/j.jisako.2023.10.015
Received 15 December 2022; Received in revised form 28 October 2023; Accepted 30 October 2023
Available online 8 November 2023
2059-7754/© 2023 The Author(s). Published by Elsevier Inc. on behalf of International Society of Arthroscopy, Knee Surgery and Orthopedic Sports Medicine. This is
an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
A.-A. Khoriati et al. Journal of ISAKOS 9 (2024) 227–233

of supervised learning is predictive modelling. This is a statistical tech-

INTRODUCTION
nique to predict future behaviour. Predictive modelling solutions are a
form of data-mining technology that analyses historical and current data
Over the last two decades, the application of Artificial Intelligence
in order to generate models, which in turn help predict future outcomes.
(AI) to the field of surgery has developed exponentially, with the number
An example of unsupervised learning is cluster analysis which involves
of PubMed search results for “artificial intelligence” more than doubling
applying clustering algorithms with the goal of finding hidden patterns or
between 2019 and 2021, according to Chen et al. [1]. The surge of in-
groupings in a data set.
terest in the orthopaedic applications for AI is also noticeable [2]. The
DL is essentially a more refined subtype of ML that studies compu-
number of total publications and citations whose title, abstract, and/or
tational models (deep neural networks) exposed to large datasets [4]. DL
keywords refer to the field of AI in orthopaedics has grown by a factor of
is capable of unsupervised learning from unstructured data, filtering out
ten between 2017 and 2021 [3]. Improvements in computer science,
data input from variables of low relevance to a prediction of interest.
processor speeds, and associated technologies have seen AI drive
The applications of AI in surgery are potentially limitless, and an in-
increasing applications relevant to orthopaedic field [3]. Advances in
depth investigation into all of these is beyond the scope of this article. For
muskuloskeletal imaging [4], arthroplasty planning [5], robotics, and
ease of understanding, we have chosen to cover several key topics most
computer navigated surgery [6] have proved to be useful tools in the
relevant to the modern orthopaedic surgeons. These are medical imaging,
orthopaedic armarium. The rapid rise in the application of these tech-
prognostication, robotics in surgery, and rehabilitation as well as VR and
nologies has not always been followed by an uptake in use, with many
simulation.
remaining reluctant to engage with the technological tools available
fully. The causes for this phenomenon remain complex and multifacto-
rial. The associated cost, required retraining, and lack of thorough edu- DIAGNOSTICS AND ENHANCED DECISION SUPPORT
cation on the benefits of AI may all contribute in part to this slow uptake
of use. This current concepts review aims to synthesise the available The assignment of AI-guided tasks was met with initial success.
literature on the subject, facilitating the understanding of this complex Computer programs easily mastered complex calculations and mathe-
field and its application, relevance, and usefulness to the orthopaedic matical predictions that were difficult for humans to perform, particu-
surgeons. larly with the advent of advanced computer processors. Paradoxically,
To facilitate the reader's understanding of the literature, we have simpler tasks—such as image or object recognition [1]—were more
broken down the subject matter into 5 broad categories: difficult for AI to master as they involved abstract concepts such as
intuition which are difficult to impart to a machine in the form of com-

Diagnostics and enhanced decision support. puter code.

Predictive analytics. Nevertheless, AI has revolutionised each stage of the imaging

Robotics and its use in surgical planning and augmentation. pathway, with improvements in imaging acquisition, interpretation,

Rehabilitation. reconstruction, and analysis [9]. Incorporating patient records, clinical

Teaching and training (including virtual reality [VR]). findings, and laboratory results has led to improved patient algorithms
capable of optimising patient-specific and appropriate imaging protocols.
This allows the reader to understand the impact of AI on the whole of AI has also improved the speed of imaging acquisition. This is particu-
the patient care pathway. This starts at the level of diagnosis and imaging larly relevant to imaging modalities such as magnetic resonance imaging
of pathology, follows through to prognostication and prediction models. (MRI) [10]. Investigations that are reliant on ionising radiation have also
The preoperative planning phase is covered as well as the impact of AI benefitted, with the overall dose associated with scans being reduced by
technology on the operative process itself. Finally, the postoperative AI optimisation [11].
management of patients (with an emphasis on bespoke rehabilitation and
telerehabilitation) is covered, with a final emphasis on the use of AI in Plain radiographs
surgical training.
The application of AI to radiograph interpretation still requires
KEY DEFINITIONS improvement. Most plain radiographs are reported by healthcare pro-
fessionals with narrative descriptions of the images being assessed. The
When examining the literature regarding AI, it is clear that multiple potential for AI to relieve some of the burden of routine image inter-
terms are often used interchangeably. This in itself may be confusing; it is pretation may in time reduce the burden of workload, eliminating the
essential to understand and define these terms clearly. risk of stress-induced interpreter error, which has been reported to be as
John McCarthy coined AI as a theory that computers could eventually high as 40% [1]. Inter- and intra-observer variability and the learning
learn to perform tasks through pattern recognition and with minimal to curve associated with image reporting is also a problem in radiograph
no human involvement [7]. AI is the theory and development of com- interpretation. Examples of this phenomenon can be seen in the mea-
puter systems able to perform tasks normally requiring human surement of acetabular component positioning—with DL measurement
intelligence. tools being developed to facilitate this process [12].
The application of AI to medicine and surgery should be considered as AI-assisted estimation of bone age is more effective than diagnosis by
two related but distinct entities [8], with the former being used to a radiologist operating alone—though the best results can be achieved
manage or treat patients without a specific interventional procedure as with the refinement of the technique with human assistance [13].
the intended result of prognostic or diagnostic investigation. The use of AI and DL in interpreting automated measurements (leg
The terms AI, machine learning (ML), and deep learning (DL) are alignment, joint orientation and leg length) is of equal accuracy but more
often used interchangeably [4]; however, there are differences in the time-effective than with human eyes alone [1]. It is important to note
meaning of each term. that in these studies, severe deformities and poor image quality were
Whereas AI refers to technology that enables computers to mimic exclusion criteria and that human supervision remains an essential
human intelligence, ML is a subset of AI that allows machines to improve component of using these techniques.
their performance by developing experience with the help of statistical The use of ML and DL has excellent potential for application to
and mathematical tools. ML algorithms can learn from examples to fracture detection. The recognition of common fracture patterns can be
enhance the accuracy of synthesised predicted models. ML algorithms imparted algorithmically to AI. Several studies have compared DL algo-
can also be trained in a supervised or unsupervised manner. An example rithms to human performance when recognising fracture patterns [14].

228
A.-A. Khoriati et al. Journal of ISAKOS 9 (2024) 227–233

The accuracy of fracture detection was high. The sensitivity and speci- Clinical decision support systems have also been used to provide rec-
ficity of hip fracture detection are as high as 97.1 % and 96.7 %, ommendations on the diagnosis and treatment of lower back pain [31],
respectively [15]. In fracture localisation, performance was lower, with Hill et al. designing a screening tool which identified at-risk sub-
ranging from 95.8 to 20% depending on fracture location. groups of patients and guided the provision of early secondary prevention
Liu et al. compared the performance of orthopaedic surgeons with AI in primary care. AI may, therefore, be useful in efficiently allocating ser-
at detecting tibial plateau fractures [16]. The accuracy of the recognition vices and improving referral pathways. These pathways must factor in a
algorithm was found to be comparable to human performance. However, number of variables, including age, gender, comorbidities, and ethnicity.
the main benefit was found to be in speed, with AI found to be 16 times
faster than orthopaedic surgeons. ROBOTICS AND ITS USE IN SURGICAL PLANNING AND
There may be a role for AI in the detection of more specialised frac- AUGMENTATION
tures, which are difficult for the generalist orthopaedic surgeon to detect,
such as vertebral fractures. The rate of missed vertebral fractures can be The advent of the robot and its application to the field of orthopaedics
as high as 30 % on plain films [17]. Deep convolutional neural networks has developed rapidly over the last two decades. Robotic surgery utilises
designed to detect vertebral fractures are as accurate as orthopaedic the advantages of complex computer calculations to optimise surgical
surgeons in detecting vertebral fractures. However, these were less ac- performance, be it in the implantation of prostheses or implants, fracture
curate than spine specialists [18], indicating room for improvement in reduction, or in the rehabilitation of orthopaedic patients.
this field. The rationale behind robotically augmented surgery lies in the basis
Other body areas studied include the wrist, femur, hand, and prox- that the knowledge and experience of correct prosthetic implantation lie
imal humerus [14]. In general, the accuracy of fracture detection is high, ultimately with the surgeon. The ability to apply this skill consistently
ranging from 83 to 98%. With fracture classification, the accuracy ranges and accurately may be deficient due to human error. Several generations
from 70 to 90% in the limited studies available [14]. Some studies have of robotically assisted tools have been developed to improve consistency
assessed the use of AI in measuring the curvature of the spine in scoliosis among arthroplasty surgeons to improve implant position and alignment
[19,20]. AI has subsequently been used to detect disc herniation [21]. and, ultimately, patient outcomes (function and implant survival).
Computer programming and planning of implant position all revolve
Advanced imaging around the accurate imaging of affected body parts, consideration of limb
alignment, and soft tissue tension. This, in turn, should theoretically
Studies have been performed on both MRI and computed tomogra- translate to correct bony preparation, precise cuts, and restoration of the
phy, particularly in the setting of trauma [22]. The accuracy and speed of physiological function of the limb. Inaccuracy of this process inevitably
detecting rib fractures are more accurate when radiologists employ the leads to implant malposition and, ultimately, failure [32].
assistance of a DL model. The use of AI-assisted diagnostics with MRI has Robotic systems may be known as “Closed” or “Open”. The former is
facilitated the detection of injuries to the anterior cruciate ligament compatible only with the type of implant associated with the robot's
(ACL), Menisci and cartilage within the knee [1], with a systematic re- manufacturer. The latter allows for a broader range of implants. It is
view by Siouras et al. [23] suggesting that the use of AI in MRI has the ultimately up to the surgeon to weigh the pros and cons of each type of
potential to be on par with human-level performance, showing a pre- robot and whether the features of an individual model outweigh the
diction accuracy of 72.5–100%. restrictions of its use and the subsequent impact on surgical freedom.
Overall, limited studies show that AI performance is comparable to Robotic systems may be image-based or imageless, with the former
human interpreters. These studies are limited for several reasons, notably system reliant on the preoperative visualisation of a patient's anatomy
their design. They are often based on one image projection. In reality, the and key mapping points used as reference points for device implantation
patient studied will have multiple views available, combined with a [33]. Preoperative imaging (CT or MRI) is crucial to this process. The
history and clinical examination. All standards of pattern recognition image-based approach allows for better preoperative preparation. Still, it
within these studies are set by human standards and, therefore, subject to comes with the disadvantages of increased cost, radiation exposure (in
human error. Finally, the overall number of these studies could be higher the case of CT), and reliance on imaging which must be taken close to the
and of better quality. This fact, combined with the potential for publi- time of surgery.
cation bias, means that the potential for the use of AI may currently be With imageless surgery, the detection and registration of the required
overplayed. A greater number of higher quality studies is needed. landmarks and surfaces directly on the patient's bones occur after expo-
sure intraoperatively. The advantages of this approach are the lower cost,
PREDICTIVE ANALYTICS avoidance of preoperative radiation, and temporal flexibility of operative
intervention. These must be weighed against the disadvantages of 1. less
AI can be used to facilitate decision-making with the recognition of flexibility in the application of orthopaedic condition, of which all the
complex results of analyses such as risk predictions, prognostications, landmarks have to be constant e.g. arthroplasty but not fractures and 2.
and treatment algorithms. This can guide the patient's pathway within an more insufficient preparation, which may impede a surgeon's ability to
appropriate clinical context [24], though ultimately the treating surgeon preselect appropriate implants and ensure their availability, particularly
and patient must interpret any data and use it to guide a shared in more complicated surgeries where the anatomy may require patient-
decision-making process. This decision-making process can predict the specific or rare implants.
clinical outcome of patients based on clinical datasets, genomic infor- Robotic systems may be known as active, passive, or semi-active.
mation, and medical images. Kim et al. were able to use ML to predict the Active robotic systems are pre-programmed by the surgeon, but after
complication rate of adults undergoing spinal deformity corrective sur- registration, the level of human interaction is the lowest as the robot
gery [25]. performs autonomously [33]. Passive robots work oppositely, with the
ML has been used to predict minimal clinically important differences in robot merely guiding the surgical process, with the surgeon mainly in
patient-reported outcomes following osteochondral graft transplantation control of the resection, with the robot providing a positioning guide
in knee surgery [26]. This process has also been applied to based on pre-planning. Some systems allow for the measurement of soft
decision-making regarding surgical outcomes and expectations in hip tissue tension intraoperatively, permitting further verification of the
arthroscopy [27], the progression of knee arthritis [28] leading to performed bony resection [34].
arthroplasty, the need for hospital admission following ACL surgery [29], Semi-active systems follow a hybrid approach between the afore-
or the need for prolonged postoperative analgesic use following arthros- mentioned surgical techniques, allowing for surgical planning followed
copy [30]. by surgeon-controlled resection. This resection is augmented by haptic

229
A.-A. Khoriati et al. Journal of ISAKOS 9 (2024) 227–233

feedback and safety measures limiting deviation from the defined sur- Overall, the available quality of evidence reviewed was considered
gical plan. The robot will regulate certain aspects of the resection, but low, with a high risk of bias. It is difficult to find any directly tangible
these features may be overridden by the surgeon, who remains in ulti- benefits to the patient with the available body of evidence, especially
mate control [33]. considering the increased cost of robotic surgical equipment. More work
is needed to justify the use of robotics more firmly in the future.
Robotics in arthroplasty
Robotics in rehabilitation
Most advances in robotics have occurred in lower limb arthroplasty,
representing over 90% of the implant market [33]. Robotic and sensor-based neurologic rehabilitation programmes are
While it has been well established that robotic-assisted arthroplasty well established and recommended for upper [47] and lower [48] limb
has been proven to improve implant positioning in both hip and knee rehabilitation. The importance of rehabilitation following trauma or
arthroplasty, there is less conclusive evidence to support improvement in elective procedures is proven, and the increasing paucity of available
functional outcomes or long-term survival of these implants [35,36]. An rehabilitation resources may mean that clinicians should prove innova-
economic analysis by Pierce et al. [37] revealed that robotic-assisted tive to cope with an increasing clinical burden.
surgery was associated with shorter length of stay, reduced utilisation
of services, and reduced 90-day costs compared with non-robotic-assisted Robotic treatment of the lower extremity focuses primarily on promoting
surgery. From a technical perspective, robotic-guided surgery has been prescribed gait patterns
found to reduce the learning curve in the implantation of uni-
compartmental knee arthroplasty [38,39]. One must consider that au- The treatment of upper limb injuries remains much more complex.
thors associated with the studies mentioned carry conflicts of interest. This is partly due to the complexity of upper limb movement (there are
Further evidence is needed with more research into the long-term out- 27 degrees of freedom in the upper limb). Both the variety and
comes of robotic-assisted arthroplasty. complexity of tasks required by the upper limb further complicate the
rehabilitative process. It has been mainly used in the training for prep-
Robotics in spinal surgery aration of the myeloelectric prosthesis for upper limb amputees. This
enables patients to perform more intuitive movements when their pros-
The most common focus of robotic surgery in spinal orthopaedics is thesis are available and in turn encourage compliance of the use of
the use of computers to guide the placement of pedicle screws [40]. prosthesis. In a recent pioneering study [47], Jakob et al. designed a
Freehand placement techniques have been historically used but are matrix-like approach to treating upper limb injuries using integrated
associated with component misplacement and subsequent complications, robotic and sensor-based devices to address distal and proximal training.
including neurological and vascular complications. Further advances in Patients were stratified by level of disability.
the field will focus more on more complex fusion procedures such as In a multicentre randomised controlled trial, robotic group therapy
higher cervical fusions and S2-sacral-iliac screw placement [40]. was found to reduce costs by 50% with equivalent outcomes.
The most extensively studied robotic spinal systems revolve around Much work remains to be done—and it should be noted that the
several key steps [41]. The first is preoperative planning, where CT imaging initial equipment and training costs may be high. However, any initial
is uploaded to pre-programmed software, and the optimal implant trajec- expense or investment may eventually be offset by savings accrued by the
tory is calculated. A small robot is then mounted on the spine. long-term economic benefits of computer-assisted rehabilitation without
Three-dimensional syncing occurs whereby the preoperative imaging adversely affecting patient outcomes.
is matched to the patient's anatomy via intraoperative fluoroscopic imaging.
Finally, a robotic arm is used to guide the trajectory of instrumentation. REHABILITATION
Future innovation in this field will revolve around augmented reality
as well as machine-guided image surgery which allows the operator to There are further uses for AI in orthopaedic rehabilitation that extend
perform surgery without the associated risk of radiation and will help beyond robotics. Wearable technology offers a source of rich, epidemi-
address line of sight issues which may hamper instrument tracking [40]. ological data through surveillance of physical behaviour [49]. Smart
wearables employ AI to monitor behaviour, activity recognition, and
Robotics in trauma pattern recognition. This allows treating physician or physiotherapist to
monitor exercise adherence and accuracy, which can often be poor. Burns
Most of the existing literature concerning the use of robotics in or- et al. [50] tested performance accuracy on individuals who performed a
thopaedics involves robotically assisted elective procedures, as most of rotator-cuff exercise protocol whilst wearing an Apple Watch. Various
these procedures have standardized technique and landmarks. Never- methods of supervised learning were used to classify exercise accuracy.
theless, some studies have been performed on trauma patients. A recently Simple interventions such as these which are easily adapted by patients
published systematic review [42] outlining the key benefits of are promising, though further research on such techniques is warranted
robotic-assisted fracture reduction has been used in several settings. The as they are relatively novel.
review focused on the following parameters: planning time, operating Though the topic of augmented reality will be covered in more detail
time, fluoroscopy time/frequency, screw placement accuracy, intra- further on, its use in the process of patient rehabilitation has increased in
operative blood loss, postoperative physical performance/functional recent years, with the development of technologies such as the Cave
outcomes and wound/fracture healing time. Automatic Virtual Environment. This system consists in a square room
Overall, a robotic intervention was found to have a net positive typically composed by either 4 or 6 six back projected screens which are
impact on trauma surgery, with reduced operating [43]/fluoroscopy combined with glasses for 3D vision. This in turn provides a continuous
times [44] and fluoroscopy frequency [44]. Improvements in screw projection surface. A linked head-tracking device allows display of real-
placement accuracy were reported in the fixation of pelvic fractures [45]. time images according to the participant's point of view, while the
Although intraoperative blood loss was reduced, no current consensus audio stimuli are delivered by speakers positioned around the device
exists on the definition of a clinically relevant volume. The Standardised [51]. Such devices are not only useful in helping create a controlled
Endpoints for Perioperative Medicine collaborative is currently con- environment where patient rehabilitation can be tested but they may also
ducting a review to reach a consensus on this matter [46]. Postoperative allow rehabilitators to assess patient confidence and slowly build it up in
physical performance and functional outcomes were not enhanced in the a measured, observable manner without subjecting the patient to undue
studies performed, and fracture healing times were unaffected. risk out in the community.

230
A.-A. Khoriati et al. Journal of ISAKOS 9 (2024) 227–233

Widespread advances in telecommunication technology have VR IN SIMULATION

increased our ability to deliver rehabilitation via the internet (i.e. tele-
rehabilitation). The use of such technology in conjunction with the VR technology has come a long way in its application over the last two
aforementioned VR and wearable technologies will no doubt broaden the decades. Initial application from its use in aerospace technology has
access of patients to rehabilitation, particularly when they are located in translated to use in multiple professional and recreational fields,
remote areas which are poorly served by local healthcare services. This in including engineering, gaming, military technology, and medical sci-
turn may ensure improved continuity of care as well as patients moni- ence. The technology in question has been used for prosthetic sizing/
toring and postoperative counselling [52]. Several studies have shown placement, remote surgery, phantom limb pain therapy, physical ther-
that telerehabilitation is effective to improve clinical outcomes in apy, joint injection as well as mobile app-based education [58].
disabling conditions [52]. In a systematic review of the literature con- The use of VR models can not only be employed in surgical training
ducted by Agostini et al. [52], a strong positive effect was found for but also within the operating theatre itself, with the ultimate aim being
patients following orthopaedic surgery, suggesting that the increased aims to increase operative accuracy and improve safety by decreasing
intensity provided by telerehabilitation holds promise as a method of procedure-related complications [58].
rehabilitation. AI can be used in conjunction with VR to enhance the personalization
and adaptation of VR interventions. AI can also improve the interactivity
SIMULATION, TEACHING, AND TRAINING IN ORTHOPAEDICS and realism of VR experiences by enabling natural language processing,
computer vision, and ML capabilities [59].
The traditional orthopaedic approach to training has been one of VR can be broken down into 3 subcategories [58].
apprenticeship, whereby orthopaedic trainees are slowly guided towards
a more complex skill set based on operating time spent effectively 1) Full visual immersion: This is in an artificial, computer-generated
practising on patients under supervision. This training model bears sig- environment. Artificial sounds and other stimuli may also be gener-
nificant drawbacks regarding time efficiency and patient safety and relies ated. This can be used in preoperative planning, patient education,
on the goodwill of surgical trainers within the context of adequate service and surgical training.
provision. 2) Augmented reality: A digital display overlay on real-world surfaces,
The concept of virtual surgical training has evolved to bypass some of which allows for depth perception. This may be used in preoperative
these obstacles. The aim is a more efficient surgical training model, planning, intraoperative guidance, and training
making optimal use of available technological resources to improve a 3) Mixed reality: This technology uses a digital display overlay com-
surgeon's skillset with maximal training opportunities at a minimal cost bined with interactive projected holograms. The surgeon views the
to the patient. The first surgical simulators can be traced back to the early real world while manipulating digital content generated by the device
nineties [53]. However, the relatively primitive technology resulted in a using commands and hand gestures. This may be used in preoperative
reluctance to adopt technology as a surgical training tool. planning, intraoperatively for guidance and in training.
With the advent of more modern simulation technology, research has
focused on adapting technology to allow surgeons to develop skills in The theory is that visual and retinal displays worn on a surgeon's face
several fields, from procedural (arthroscopic/arthroplasty) to sensory in the form of goggles or glasses may relay information to the operator in
(haptic feedback technology). real-time—displaying both geometric guidance and the accuracy of in-
The application of AI technology to simulators enhances the training strument placement. Such technology has excellent synergy with other
experience by providing personalised feedback to the user, while also technological advances, including robotic surgery, and the two tech-
automating an immersive surgical experience for visualisation of patient niques are increasingly used in conjunction with one another.
anatomy [54]. AI is able to enhance surgical training simulators by Retinal displays and the use of mixed reality may also be used to view
evaluating a subject's performance and providing individualised feed- 3D representations of preoperative imaging, allowing surgeons to better
back to the end user [55]. orientate themselves according to the patient's native anatomy.
The value of retention of surgical skills gained in simulation has been There is level I evidence supporting the use of VR in surgical training
demonstrated at six months post training [56], though this finding is with increased procedural accuracy and the completion of tasks
contradicted in other parts of the literature [57]. It is important to note demonstrated in medical students using the technology compared to
that studies on the subject of retention are rare and usually involve those using guides [60]. VR is thought to augment learning the proce-
different procedures, so drawing any firm conclusions on the matter is dural workflow and movements required to perform surgical tasks.
complex. Simulators fail to simulate the stressful conditions present in A recent study comparing VR arthroscopic simulation [61] with
emergency surgery. However, one could argue that the purpose of the cadaveric models demonstrated superiority in task completion time.
surgical simulation is to develop motor skills that are second nature, to be There is scant evidence in the literature comparing the two training
deployed unconsciously in times of stress. In addition, it helps surgeons techniques. The benefits of simulated surgery lie primarily in cost savings
to maintain the skills e.g. in the time of COVID – 19 where there is a lack compared to the expense and sparsity of cadaveric training models.
of cases. A potential benefit of augmented reality that has yet to be fully
For surgical simulators to be used successfully, they should be explored is the potential reduction in radiation exposure to the patient
deployed within a predetermined training framework containing several and operating theatre staff, with sensors being able to direct the surgeon
key steps. These begin with a sound theoretical understanding of the to place metalwork without the need for potentially harmful X-rays [62].
techniques used, followed by simple simulators and more complex tasks. Although the initial costs associated with the use of VR technology
These are followed by a cadaveric test run before the surgeon in training may seem prohibitive, surgeons considering its use should consider the
can operate on human patients. While this model is sound in theory, the fact that these costs may be offset by the gains improved performance,
realities of the associated costs mean that such a model is likely to see an training and ultimately the patient.
uptake in general surgical use once the resources to do so become widely
and more cheaply available. CONCLUSIONS
The future of arthroscopic simulators lies in the employment of haptic
feedback devices, though these are only widely available. Current active The field of AI in orthopaedics is exponentially growing. It enhances
haptic technology, which employs motors to simulate tactile feedback, surgeons’ performance in many different areas, including diagnosis, and
does not demonstrate sufficient face validity or match the sophistication precision of surgery. This in turn aims to improve the outcome of patients
of passive haptic systems in high-fidelity arthroscopy simulators. care. The ability of surgeons to keep pace with this rapidly evolving field

231
A.-A. Khoriati et al. Journal of ISAKOS 9 (2024) 227–233

will be vital to exploiting future technological developments. Future [21] Oktay AB, Albayrak NB, Akgul YS. Computer aided diagnosis of degenerative
intervertebral disc diseases from lumbar MR images. Comput Med Imag Graph
accessibility and education remain key to the achievement of this goal. A
2014;38(7):613–9. https://doi.org/10.1016/j.compmedimag.2014.04.006.
significant volume of research contained within the engineering litera- [22] Meng XH, Wu Di, Wang Z, Ma X, Dong X, Liu A, et al. A fully automated rib fracture
ture may not be readily accessible to orthopaedic surgeons and may not detection system on chest CT images and its impact on radiologist performance.
reach readers with a clinical background [63]. Future integration of AI, Skeletal Radiol 2021;50(9):1821–8. https://doi.org/10.1007/s00256-021-03709-8.
[23] Siouras A, Moustakidis S, Giannakidis A, Chalatsis G, Liampas I, Vlychou M, et al.
robotic, and VR-related education is the solution. Changes must be Knee injury detection using deep learning on MRI studies: a systematic review.
implemented to medical and surgical curricula at an early stage in order Diagnostics 2022;12(2):537. https://doi.org/10.3390/diagnostics12020537.
to ensure a future optimised for the highest quality of patient care. [24] Hashimoto DA, Rosman G, Rus D, Meireles OR. Artificial intelligence in surgery:
promises and perils. Ann Surg 2018;268(1):70–6. https://doi.org/10.1097/
SLA.0000000000002693.
Declaration of competing interest [25] Kim JS, Arvind V, Oermann E, Kaji D, Ranson W, Ugoku C, et al. Predicting surgical
complications in patients undergoing elective adult spinal deformity procedures
using machine learning. Spine Deform 2018;6(6):762–70. https://doi.org/
All authors declare that they have no conflicts of interest. 10.1016/j.jspd.2018.03.003.
[26] Ramkumar PN, Karnuta J, Haeberle H, Owusu-Akyaw K, Warner T, Rodeo S, et al.
References Association between preoperative mental health and clinically meaningful
outcomes after osteochondral allograft for cartilage defects of the knee: a machine
learning analysis. Am J Sports Med 2021;49(4):948–57. https://doi.org/10.1177/
[1] Chen K, Stotter C, Klestil T, Nehrer S. Artificial intelligence in orthopedic
0363546520988021.
radiography analysis: a narrative review. Diagnostics 2022;12(9):2235. https://
[27] Kunze KN, Polce EM, Clapp I, Nwachukwu BU, Chahla J, Nho SJ. Machine learning
doi.org/10.3390/diagnostics12092235.
algorithms predict functional improvement after hip arthroscopy for
[2] Lisacek-Kiosoglous AB, Powling AS, Fontalis A, Gabr A, Mazomenos E, Haddad FS.
femoroacetabular impingement syndrome in athletes. J Bone Joint Surg Am 2021;
Artificial intelligence in orthopaedic surgery. Bone Joint Res 2023;12(7):447–54.
103(12):1055–62. https://doi.org/10.2106/JBJS.20.01640.
https://doi.org/10.1302/2046-3758.127.BJR-2023-0111.R1.
[28] Pareek A, Parkes CW, Bernard CD, Abdel MP, Saris DBF, Krych AJ. The SIFK score: a
[3] Farhadi F, Barnes MR, Sugito HR, Sin JM, Henderson ER, Levy JJ. Applications of
validated predictive model for arthroplasty progression after subchondral
artificial intelligence in orthopaedic surgery. Front Med Technol 2022;4:995526.
insufficiency fractures of the knee. Knee Surg Sports Traumatol Arthrosc 2020;
https://doi.org/10.3389/fmedt.2022.995526.
28(10):3149–55. https://doi.org/10.1007/s00167-019-05792-w.
[4] Gyftopoulos S, Lin D, Knoll F, Doshi AM, Rodrigues TC, Recht MP. Artificial
[29] Lu Y, Forlenza E, Cohn M, Lavoie-Gagne O, Wilbur R, Song B, et al. Machine
intelligence in musculoskeletal imaging: current status and future directions. AJR
learning can reliably identify patients at risk of overnight hospital admission
Am J Roentgenol 2019;213(3):506–13. https://doi.org/10.2214/AJR.19.21117.
following anterior cruciate ligament reconstruction. Knee Surg Sports Traumatol
[5] Li W, Xu S-M, Zhang D-B, Bi H-Y, Gu G-S. Research advances in the application of AI
Arthrosc 2021;29(9):2958–66. https://doi.org/10.1007/s00167-020-06321-w.
for preoperative measurements in total knee arthroplasty. Life 2023;13(2):451.
[30] Lu Y, Forlenza E, Wilbur R, Lavoie-Gagne O, Fu M, Yanke A, et al. Machine-learning
https://doi.org/10.3390/life13020451.
model successfully predicts patients at risk for prolonged postoperative opioid use
[6] Picard F, Deakin AH, Riches PE, Deep K, Baines J. Computer assisted orthopaedic
following elective knee arthroscopy. Knee Surg Sports Traumatol Arthrosc 2022;
surgery: past, present and future. Med Eng Phys 2019;72:55–65. https://doi.org/
30(3):762–72. https://doi.org/10.1007/s00167-020-06421-7.
10.1016/j.medengphy.2019.08.005.
[31] Hill JC, Dunn K, Lewis M, Mullis R, Main C, Foster N, et al. A primary care back pain
[7] Myers TG, Ramkumar PN, Ricciardi BF, Urish KL, Kipper J, Ketonis C. Artificial
screening tool: identifying patient subgroups for initial treatment. Arthritis Rheum
intelligence and orthopaedics. J Bone Joint Surg Am 2020;102(9):830–40. https://
2008;59(5):632–41. https://doi.org/10.1002/art.23563.
doi.org/10.2106/JBJS.19.01128.
[32] Ritter MA, Davis KE, Meding JB, Pierson JL, Berend ME, Malinzak RA. The effect of
[8] Gumbs AA, Frigerio I, Spolverato G, Croner R, Illanes A, Chouillard E, et al.
alignment and BMI on failure of total knee replacement. J Bone Joint Surg Am
Artificial intelligence surgery: how do we get to autonomous actions in surgery?
2011;93(17):1588–96. https://doi.org/10.2106/JBJS.J.00772.
Sensors 2021;21(16):5526. https://doi.org/10.3390/s21165526.
[33] Innocenti B, Bori E. Robotics in orthopaedic surgery: why, what and how? Arch
[9] Han X-G, Tian W. Artificial intelligence in orthopedic surgery: current state and
Orthop Trauma Surg 2021;141(12):2035–42. https://doi.org/10.1007/s00402-
future perspective. Chin Med J 2019;132(21):2521–3. https://doi.org/10.1097/
021-04046-0.
CM9.0000000000000479.
[34] Parratte S, Price AJ, Jeys LM, Jackson WF, Clarke HD. Accuracy of a new robotically
[10] Hammernik K, Klatzer T, Kobler E, Recht MP, Sodickson DK, Pock T, et al. Learning
assisted technique for total knee arthroplasty: a cadaveric study. J Arthroplasty
a variational network for reconstruction of accelerated MRI data. Magn Reson Med
2019;34(11):2799–803. https://doi.org/10.1016/j.arth.2019.06.040.
2018;79(6):3055–71. https://doi.org/10.1002/mrm.26977.
[35] Ruangsomboon P, Ruangsomboon O, Pornrattanamaneewong C, Narkbunnam R,
[11] Gupta RV, Kalra MK, Ebrahimian S, Kaviani P, Primak A, Bizzo B, et al. Complex
Chareancholvanich K. Clinical and radiological outcomes of robotic-assisted versus
relationship between artificial intelligence and CT radiation dose. Acad Radiol
conventional total knee arthroplasty: a systematic review and meta-analysis of
2022;29(11):1709–19. https://doi.org/10.1016/j.acra.2021.10.024.
randomized controlled trials. Acta Orthop 2023;94:60–79. https://doi.org/
[12] Rouzrokh P, Wyles C, Philbrick K, Ramazanian T, Weston A, Cai J, et al. A deep
10.2340/17453674.2023.9411.
learning tool for automated radiographic measurement of acetabular component
[36] Chen X, Xiong J, Wang P, Zhu S, Qi W, Peng H, et al. Robotic-assisted compared
inclination and version after total hip arthroplasty. J Arthroplasty 2021;36(7):
with conventional total hip arthroplasty: systematic review and meta-analysis.
2510–2517.e6. https://doi.org/10.1016/j.arth.2021.02.026.
Postgrad Med 2018;94(1112):335–41. https://doi.org/10.1136/postgradmedj-
[13] Tajmir SH, Lee H, Shailam R, Gale H, Nguyen J, Westra S, et al. Artificial
2017-135352.
intelligence-assisted interpretation of bone age radiographs improves accuracy and
[37] Pierce J, Needham K, Adams C, Coppolecchia A, Lavernia C. Robotic arm-assisted
decreases variability. Skeletal Radiol 2019;48(2):275–83. https://doi.org/10.1007/
knee surgery: an economic analysis. Am J Manag Care 2020;26(7):e205–10.
s00256-018-3033-2.
https://doi.org/10.37765/ajmc.2020.43763.
[14] Langerhuizen DWG, Janssen S, Mallee W, van den Bekerom M, Ring D, Kerkhoffs G,
[38] Lonner JH. Indications for unicompartmental knee arthroplasty and rationale for
et al. What are the applications and limitations of artificial intelligence for fracture
robotic arm-assisted technology. Am J Orthop 2009;38(2 Suppl):3–6.
detection and classification in orthopaedic trauma imaging? A systematic review.
[39] Coon TM. Integrating robotic technology into the operating room. Am J Orthop
Clin Orthop Relat Res 2019;477(11):2482–91. https://doi.org/10.1097/
2009;38(2 Suppl):7–9.
CORR.0000000000000848.
[40] Jiang B, Azad T, Cotrill E, Zygourakis C, Zhu A, Crwford N, et al. New spinal robotic
[15] Yu JS, Yu S, Erdal B, Demirer M, Gupta V, Bigelow M, et al. Detection and
technologies. Front Med 2019;13(6):723–9. https://doi.org/10.1007/s11684-019-
localisation of hip fractures on anteroposterior radiographs with artificial
0716-6.
intelligence: proof of concept. Clin Radiol 2020;75(3):237.e1–9. https://doi.org/
[41] Ghasem A, Sharma A, Greif DN, Alam M, Maaieh MA. The arrival of robotics in
10.1016/j.crad.2019.10.022.
spine surgery: a review of the literature. Spine 2018;43(23):1670–7. https://
[16] Liu P, Lu L, Chen Y, Huo T, Xue M, Wang H, et al. Artificial intelligence to detect the
doi.org/10.1097/BRS.0000000000002695.
femoral intertrochanteric fracture: the arrival of the intelligent-medicine era. Front
[42] Schuijt HJ, Hundersmarck D, Smeeing DPJ, van der Velde D, Weaver MJ. Robot-
Bioeng Biotechnol 2022;10:927926. https://doi.org/10.3389/fbioe.2022.927926.
assisted fracture fixation in orthopaedic trauma surgery: a systematic review. OTA
[17] Berry GE, Adams S, Harris M, Boles C, McKernan M, Collinson F, et al. Are plain
Int 2021;4(4):e153. https://doi.org/10.1097/OI9.0000000000000153.
radiographs of the spine necessary during evaluation after blunt trauma? Accuracy
[43] Lan H, Tan Z, Li K-N, Gao J-H, Liu T-H. Intramedullary nail fixation assisted by
of screening torso computed tomography in thoracic/lumbar spine fracture
orthopaedic robot navigation for intertrochanteric fractures in elderly patients.
diagnosis. J Trauma 2005;59(6):1410–3. https://doi.org/10.1097/
Orthop Surg 2019;11(2):255–62. https://doi.org/10.1111/os.12447.
01.ta.0000197279.97113.0e.
[44] Long T, Li K, Gao J, Lui T, Mu J, Wang X, et al. Comparative study of percutaneous
[18] Murata K, Endo K, Aihara T, Suzuki H, Sawaji Y, Matsouka Y, et al. Artificial
sacroiliac screw with or without TiRobot assistance for treating pelvic posterior ring
intelligence for the detection of vertebral fractures on plain spinal radiography. Sci
fractures. Orthop Surg 2019;11(3):386–96. https://doi.org/10.1111/os.12461.
Rep 2020;10:20031. https://doi.org/10.1038/s41598-020-76866-w.
[45] Liu H-S, Duan S-J, Xin F-Z, Zhang Z, Wang X-G, Liu S-D. Robot-assisted minimally-
[19] Duong L, Cheriet F, Labelle H. Three-dimensional classification of spinal deformities
invasive internal fixation of pelvic ring injuries: a single-center experience. Orthop
using fuzzy clustering. Spine 2006;31(8):923–30. https://doi.org/10.1097/
Surg 2019;11(1):42–51. https://doi.org/10.1111/os.12423.
01.brs.0000209312.62384.c1.
[46] Bartoszko J, Vorobeichik L, Jayarajah M, Karkouti K, Klein A, Lamyet A, et al.
[20] Ramirez L, Durdle NG, Raso VJ, Hill DL. A support vector machines classifier to
Defining clinically important perioperative blood loss and transfusion for the
assess the severity of idiopathic scoliosis from surface topography. IEEE Trans Inf
Standardised Endpoints for Perioperative Medicine (StEP) collaborative: a protocol
Technol Biomed 2006;10(1):84–91. https://doi.org/10.1109/titb.2005.855526.

232
A.-A. Khoriati et al. Journal of ISAKOS 9 (2024) 227–233

for a scoping review. BMJ Open 2017;7(6):e016743. https://doi.org/10.1136/ Comput Aided Surg 2008;13(2):63–81. https://doi.org/10.3109/
bmjopen-2017-016743. 10929080801957712.
[47] Jakob I, Kollreider A, Germanotta M, Benetti F, Cruciani A, Padua L, et al. Robotic [56] Jackson WFM, Khan T, Alvand A, Al-Ali S, Gill H, Price A, et al. Learning and
and sensor technology for upper limb rehabilitation. Pharm Manag PM R 2018;10(9 retaining simulated arthroscopic meniscal repair skills. J Bone Joint Surg Am 2012;
Suppl 2):S189–97. https://doi.org/10.1016/j.pmrj.2018.07.011. 94(17):e132. https://doi.org/10.2106/JBJS.K.01438.
[48] Zhang X, Yue Z, Wang J. Robotics in lower-limb rehabilitation after stroke. Behav [57] Howells NR, Auplish S, Hand GC, Gill HS, Carr AJ, Rees JL. Retention of
Neurol 2017;2017:3731802. https://doi.org/10.1155/2017/3731802. arthroscopic shoulder skills learned with use of a simulator. Demonstration of a
[49] Tack C. Artificial intelligence and machine learning | applications in learning curve and loss of performance level after a time delay. J Bone Joint Surg
musculoskeletal physiotherapy. Musculoskelet Sci Pract 2019;39:164–9. https:// Am 2009;91(5):1207–13. https://doi.org/10.2106/JBJS.H.00509.
doi.org/10.1016/j.msksp.2018.11.012. [58] Verhey JT, Haglin JM, Verhey EM, Hartigan DE. Virtual, augmented, and mixed
[50] Burns DM, Leung N, Hardisty M, Whyne CM, Henry P, McLachlin S. Shoulder reality applications in orthopedic surgery. Int J Med Robot Comput Assist Surg
physiotherapy exercise recognition: machine learning the inertial signals from a 2020;16(2):e2067. https://doi.org/10.1002/rcs.2067.
smartwatch. Physiol Meas 2018;39(7):075007. https://doi.org/10.1088/1361- [59] Combalia A, Sanchez-Vives MV, Donegan T. Immersive virtual reality in
6579/aacfd9. orthopaedics—a narrative review. Int Orthop Aug 2023. https://doi.org/10.1007/
[51] Tieri G, Morone G, Paolucci S, Iosa M. Virtual reality in cognitive and motor s00264-023-05911-w [Online ahead of print].
rehabilitation: facts, fiction and fallacies. Expet Rev Med Dev 2018;15(2):107–17. [60] Orland MD, Patetta MJ, Wieser M, Kayupov E, Gonzalez MH. Does virtual reality
https://doi.org/10.1080/17434440.2018.1425613. improve procedural completion and accuracy in an intramedullary tibial nail
[52] Agostini M, Moja L, Banzi R, Pistonni V, Tonin P, Venneri A, et al. Telerehabilitation procedure? A randomized control trial. Clin Orthop Relat Res 2020;478(9):2170–7.
and recovery of motor function: a systematic review and meta-analysis. J Telemed https://doi.org/10.1097/CORR.0000000000001362.
Telecare 2015;21(4):202–13. https://doi.org/10.1177/1357633X15572201. [61] Huri_ G, Gülşen MR, Karmiş EB, Karagüven D. Cadaver versus simulator based
[53] Satava RM. Virtual reality surgical simulator. The first steps. Surg Endosc 1993; arthroscopic training in shoulder surgery. Turk J Med Sci 2021;51(3):1179–90.
7(3):203–5. https://doi.org/10.1007/BF00594110. https://doi.org/10.3906/sag-2011-71.
[54] Park JJ, Tiefenbach J, Demetriades AK. The role of artificial intelligence in surgical [62] Multi-modal imaging, model-based tracking, and mixed reality visualisation for
simulation. Front Med Technol 2022;4:1076755. https://doi.org/10.3389/ orthopaedic surgery - PMC’. Accessed: May 24, 2023. [Online]. Available: http
fmedt.2022.1076755. s://www.ncbi.nlm.nih.gov/pmc/articles/PMC5683202/.
[55] Sewell C, Morris D, Blevins N, Dutta S, Agarwal S, Barbagli F, et al. [63] Federer SJ, Jones GG. Artificial intelligence in orthopaedics: a scoping review. PLoS
Providing metrics and performance feedback in a surgical simulator. One 2021;16(11):e0260471. https://doi.org/10.1371/journal.pone.0260471.

233