Abstract
The use of artificial intelligence (AI) is rapidly growing across many domains, of which the medical field is no exception. AI is an umbrella term defining the practical application of algorithms to generate useful output, without the need of human cognition. Owing to the expanding volume of patient information collected, known as ‘big data’, AI is showing promise as a useful tool in healthcare research and across all aspects of patient care pathways. Practical applications in orthopaedic surgery include: diagnostics, such as fracture recognition and tumour detection; predictive models of clinical and patient-reported outcome measures, such as calculating mortality rates and length of hospital stay; and real-time rehabilitation monitoring and surgical training. However, clinicians should remain cognizant of AI’s limitations, as the development of robust reporting and validation frameworks is of paramount importance to prevent avoidable errors and biases. The aim of this review article is to provide a comprehensive understanding of AI and its subfields, as well as to delineate its existing clinical applications in trauma and orthopaedic surgery. Furthermore, this narrative review expands upon the limitations of AI and future direction.
Cite this article: Bone Joint Res 2023;12(7):447–454.
Article focus
-
Comprehensive review of artificial intelligence (AI) and its subfields, as well as existing applications and its role in orthopaedic surgery.
-
Critical presentation of validated and evolving AI research in orthopaedic surgery.
-
Limitations of AI and the need to establish robust validation and reporting frameworks.
Key messages
-
AI is showing promise as a useful tool in healthcare research and data science, including all aspects of patient care pathways.
-
Existing applications in orthopaedic surgery have shown promise in highlighting implant malposition; detecting features of loosening; predicting length of hospital stay, costs involved, functional outcomes, and prognostic scores; and implant identification in arthroplasty.
-
Clinicians should remain cognizant of AI’s limitations and proceed cautiously, until external validity is proven within acceptable margins of error.
Strengths and limitations
-
This is a comprehensive literature review of recent advances and AI applications in orthopaedic surgery, detailing areas lacking validated research.
-
Our study does not entail quantitative synthesis of outcomes with the use of AI.
Introduction: artificial intelligence, time for clear nomenclature
The application of artificial intelligence (AI) is rapidly growing across many domains, with the field of medicine being no exception. Traditionally AI is an umbrella term, originally theorizing the replication of human intellect via computers.1 The broad definition of AI is the practical application of complex algorithms to generate useful output, excluding the need for human cognitive intelligence.2,3 AI is becoming an integral part of modern society, ranging from air flight autopilot to fraud detection, social media advertisements, and the seemingly omniscient capabilities of ChatGPT.4 It is estimated that AI could cut annual USA healthcare costs by $150 billion by 2026.5,6 A considerable component of the cost reduction stems from adopting a proactive health management approach, expected to result in fewer hospitalizations, fewer doctor visits, and reduced treatments.5,6 This may be attributed to early detection of disease with known cures, through automating the review of large volumes of data using AI with advanced individualized risk profiling.7
Owing to the exponentially expanding volume of patient information collected, known as ‘big data’, AI is showing promise as a useful tool in the healthcare research armamentarium. Algorithms based on clinical data sets (including electronic medical records, genomic level data, validated clinical scores and imaging etc.) for predicting patients’ clinical outcomes are rigorously being explored. This concept encompasses data sets so large that there is no conceivable way that humans could comprehend such a plethora of information without the use of technology.3,8 While AI can categorize and make sense of big data, it is still only as good as the data provided and thus human contribution is paramount. The accuracy of its function can be progressively refined, as it has intriguingly been likened to successive human learning, whereby sequential exposure reinforces comprehension.2,3,9
While still in its infancy, the application of AI in the field of orthopaedics is a new frontier of data science. Orthopaedic surgery is already home to some of the most innovative technologies, such as robotic-assisted surgery, of which AI is an ever-growing part.10-18 Recently, the orthopaedic and wider healthcare literature have witnessed a surge in studies using AI, which on many occasions employ methodology not very different to traditional prediction models. To describe the application of AI models in orthopaedic surgery, it is necessary to delineate the concepts of each architectural design. The basis through which each model is created describes the level of complexity, power, and importantly the limitations. It is therefore of paramount importance to differentiate between different types of AI (Figure 1), in order to achieve consistency and ensure transparency for the readers. To accomplish that, enhanced understanding of AI and its subcategories is necessary, while abandoning a focus on the umbrella term AI could be considered when it comes to orthopaedic research.
Fig. 1
Key artificial intelligence applications in trauma and orthopaedic surgery. ACL, anterior cruciate ligament; NLP, natural language processing.
The purpose of this narrative review article is to provide a comprehensive understanding of AI and its subfields, in addition to delineating its role in orthopaedic surgery and describing current existing applications. Furthermore, this review explores the current limitations and touches upon future direction.
Machine learning
AI encompasses a subfield called machine learning (ML) (Figure 1).19 ML can be described as harnessing the dimensions to ‘learn and adapt’ based on algorithms and input data, often surpassing human comprehension.20 Further subclassifications include supervised, unsupervised, and reinforcement ML. Supervised ML involves input data being labelled by humans and correcting the computer’s mistakes. For example, a computer is shown thousands of images of a normal radiograph (the computer recognizes all the peculiarities from pixels identified by human supervision) and then thousands of images of a broken bone. An AI algorithm dictates the recognition of what is labelled ‘broken’ or ‘not broken’. ML is only halfway complete and, following this process, the model must be refined or trained to validate accuracy before wider use. In the context of supervised ML, ‘ground truth’ data used to train the ML models are typically labelled or annotated by humans, who indicate the correct answer or outcome for a given input.19,20 Having high-quality, accurate ground truth data is crucial in validating ML models. Furthermore, when the model is trained to provide accurate ground truth, this allows for reuse of the model on new problems, which is referred to as ‘transfer learning’. In this way, the developer can avoid having to recreate entire new algorithms from new data sets, therefore saving huge amounts of time and money. Transfer learning has been applied to fracture recognition and osteoarthritis (OA) quantification, to name but a few.21
Unsupervised ML processes unlabelled training data, with a known outcome of interest, clustering them as known or unknown. In the aforementioned example, radiograph images of the same anatomical bone (e.g. a hip radiograph) allow the computer to ascertain what normal looks like. The computer then groups similar data and patterns them together (e.g. for radiographs of broken and unbroken bones), using an algorithm.22,23 The repetition of these functions can be used to fine-tune the algorithm and improve accuracy. This process (called an ‘epoch’) may be repeated up to a thousand times to achieve the accuracy required, before an algorithm can move beyond proof-of-concept and enter the validation phase. It is through this mechanism that a final algorithm can be established and applied to unknown data sets as required.8,19
Semi-supervised or reinforcement ML learns by exploration of the environment based on reward or punishment from certain actions (e.g. self-driving cars from Tesla).1,22,23 There is a growing body of research in deep reinforcement learning for the application of models in computer-assisted orthopaedic surgery, reporting capabilities of generating real-world, clinical-grade solutions without needing patient data for training.24 However, validated and reproducible high-quality tools are awaited.
Deep learning and neural networks
Deep learning (DL) is a more progressive and comprehensive subcategory of ML, comprising numerous, complex layers of algorithm, mirroring the neural networks seen in the brain through artificial neural networks (ANNs) (Figure 1).8,20,25,26 Where ML comprises thousands to millions of parameters, DL may have billions, with varying degrees and layers of complexity to broaden the function for which it is programmed. Akin to ML, much of DL requires human supervision to learn and be modified. However, there is an increasing research interest in DL models functioning without the need for human supervision. It operates with unlabelled and unstructured input, permitting the output of interest. An example of DL that is being explored in the world of orthopaedics is convolutional neural networks (CNNs), often purposed for imaging analysis and computer vision tasks.25,27
There has been a surge of research in CNNs pertaining to diagnostic and image recognition, classification and tumour detection, segmentation, and natural language processing (NLP), of which the field of orthopaedics is no exception. The mathematical architecture of CNNs may be thought of as an overlapping of grid patterns.25 The basic building blocks include convolutional layers (a combination of linear and non-linear operations used to extract image data), pooling layers (to reduce learnable parameters), and fully connected layers designed to automatically propagate image input and learn spatial hierarchies of features through a forward and backward propagation algorithm.25 For the most part, CNNs are made up of these three layers: the first two layers, convolution and pooling, perform feature extraction, whereas the third, a fully connected layer, maps the extracted features into final output, such as a classification.25
Natural language processing and electronic medical records
NLP describes computer comprehension of language.3 Functionally, it can scan clinical medical records and make sense of information such as operative notes and radiology reports. NLP algorithms have the potential to automate data collection for diagnostic elements, which could directly improve patient care and augment cohort surveillance.28 The implications of NLP may include aggregating and analyzing large databases abundant in information, usually too arduous for manual sorting, and reducing documentation time.3,29 It has already seen use in organizing relevant data from electronic medical health records during cases of periprosthetic fractures, and aiding the diagnosis of periprosthetic joint infections.28,30 For example, data elements that comprise the Musculoskeletal Infection Society (MSIS) criteria31 were manually extracted and used as the gold standard for validation. The NLP algorithm was applied to extract the presence of sinus tract, purulence, pathological documentation of inflammation, and growth of cultured organisms from medical records.
Image recognition and diagnostics
Fracture recognition
The past decade has seen imaging analysis become a considerable focal point among AI research.32 Numerous authors have studied the ability of CNNs to identify various upper and lower limb fractures on radiographs, such as hip, calcaneus, and radial fractures. Accuracy of up to 98% has been reported, as well as the potential of CNNs to outperform or perform non-inferiorly to humans.26,32-36 It has also been reported that DL models could recognize laterality, exam view, and body part in wrist, hand, and ankle radiographs. They have also shown promise in achieving more notoriously difficult diagnoses, such as scaphoid fractures, as effectively as human specialists.20,30,35,37 It should not be forgotten that the performance of CNNs needs to be validated both internally and externally before clinical adoption. While internal validation is proving quite successful, there are hurdles that make external validation difficult, e.g. for fracture classification. One reported issue is relating to different institutions using different labelling systems for radiographs or radiation dosages. In particular, if a given institution changes their protocols, previously validated algorithms may become invalid and problematic to translate.34
A recent paper by Oliveira E Carmo et al38 highlights the lack of external validity of CNNs for fracture detection in the literature. In a large systematic review, only four studies (11% of total studies identified) were found to show external validity, both temporal and geographical, beyond one hospital site. The authors recommend the use of standardized reporting guidelines in order to ascertain ground truth for CNNs in fracture recognition, such as the Clinical Artificial Intelligence Research (CAIR) checklist,39 to critically appraise performance of CNNs to facilitate eventual implementation into clinical practice.38
Tumour detection
The potential of AI has been shown to extend beyond fracture recognition. Park et al40 showed that a CNN was able to eclipse the accurate detection of proximal femur bone tumours compared to clinicians.40 Specifically, ML may prove useful for the diagnosis of more ambiguous primary bone and soft-tissue tumours, ones that are not clearly evident on plain radiographs. These applications have also been proposed to help predict patients’ prognosis, such as those with synovial sarcoma.20,23,41
Other diagnostic applications
AI has also shown promising results in several other diagnostic applications, ranging from developmental abnormalities to soft-tissue knee injuries. A proof-of-concept investigation by Xie et al42 tested a CNN-based algorithm to improve the quality of MRI scans in tibial plateau fractures with combined meniscal defects.43 The authors documented a sensitivity of 96.9%, specificity of 93.2%, and accuracy of 95.3%, respectively, when MRI diagnostics were compared with arthroscopic findings. The clearer, enhanced AI imaging produced by the CNN model led to a diagnosis that was consistent with intraoperative findings. This study is one of many that highlights feasible grounds for future research and advancements for current imaging modalities.42 Regarding congenital abnormalities, such as hip dysplasia, studies have also shown practicalities for radiological measurements in a quick and effective manner.44 AI-assisted diagnosis and classification of OA from radiographs have demonstrated similar accuracy to senior clinicians.20 Furthermore, CNNs for osteoporosis fracture recognition have been developed to directly evaluate bone mineral density from radiographs.45,46
AI image recognition may soon be a highly sought-after application in orthopaedics, corroborated in a study by Jang et al47 where CNNs were reported to identify bone and soft-tissue landmarks as objects on radiographs. Additionally, more accurate calculations using the DL model for knee alignment may provide the potential for preoperative planning in total knee arthroplasty (TKA).47 However, several limitations such as the established ground truths, radiograph quality, alignment, or rotation indicate the variability and, as such, these methods are not yet employed in preoperative planning for TKA.47
A recent scoping review by Gurung et al48 investigated the application of AI in analyzing postoperative radiographs following total hip arthroplasty (THA) and TKA to ensure adequate implant positioning, and reported > 90% accuracy. While the 12 individual studies were large, using up to 320,000 radiographs, their robustness was a point of contention. The authors concluded that there is currently insufficient evidence to use AI for said purposes in clinical practice.48
Automated identification of arthroplasty implants using DL has been reported to be a useful augment in revision surgery, enabling accurate planning of the operative technique and necessary extraction equipment.8,25,34,49 A study by Borjali et al49 assessed a novel, highly accurate, and fully automatic approach identifying the design of THA prosthesis from plain radiographs. An AI model able to identify prosthesis within milliseconds, versus 20 to 30 minutes, can have huge implications for patient safety.49 Furthermore, it has been shown that in 10% of cases, surgeons are unable to identify the prosthesis preoperatively and 2% intraoperatively.49 This has been shown to be associated with increased operating time, blood/bone loss, recovery time, and healthcare costs.49 A sensitivity up to 94% and specificity of 97% in identifying implant loosening following hip and knee arthroplasty using CNNs has been reported.20 Of note, the CNN algorithm outperformed the human counterpart from plain radiographs, illustrating its potential role in preventing serious complications and redistributing clinical time to improve patient care.20,50
Predictive algorithms
Recent literature has showcased the predictive value of AI models to calculate mortality rates, transfusion risk, and length of hospital stay following elective arthroplasty.8,34,51-53 This could be of particular benefit when considering patient care pathways, from preoperative optimization to recovery plans and resource allocation.25 It has also been reported that DL/ML models could predict, up to a decade in advance, knee and hip OA by means of bone texture analysis on the proximal femur and acetabulum, and clinical risk factors, with acceptable accuracy.8,21,54 Conceptually, this could act as a risk stratification tool, identifying individuals in need of early intervention.25 A recent study comparing a conventional ML, ANN model with traditional logistic regression of 28,742 patients from the National Surgery Quality Improvement Programme (USA) has demonstrated similar predictability of clinically important factors for safe same-day discharge post TKA using the ANN model.51
Multiple AI predictive models assimilating large amounts of patient data to improve healthcare outcomes have been described. Examples in orthopaedic surgery include AI models predicting suitable patients for nerve blocks following anterior cruciate ligament (ACL) reconstruction.32 Kim et al55 developed a DL algorithm to predict the mortality and morbidity risk following spinal fusion, and found this to be more accurate compared to the traditionally used scoring system by the American Society of Anesthesiologists.56 Another interesting application of AI is showcased by Kumar et al,57 who developed a ML algorithm predicting patient outcomes in shoulder arthroplasty. The input comprises shoulder range of motion, demographic data, American Shoulder and Elbow Surgeons (ASES) scores,58 and visual analogue scale (VAS) pain scores, to assess prognosis and range of motion up to seven years post-treatment, with up to 82% reported accuracy.8,57,59 A recent study involving a total of 111,147 patients undergoing primary shoulder arthroplasty reported 73.1% to 91.8% accuracy using ANN in predicting length of stay, hospital costs, and discharge disposition for both chronic/degenerative and acute/traumatic conditions.60 From a recent retrospective multicentre analysis of nearly 2,000 patients following total shoulder arthroplasty, a model to predict two-year ASES scores has been developed and validated. The model was reported to be accurate within the minimal clinically important difference in 85% of patients.22
The role of AI in surgical training
AI could play a pivotal role in orthopaedic surgical training, where repetition and the existence of a training framework are imperative to acquiring competence.61 Through ML and computer vision, AI now has the capacity to gather data and provide meaningful, personalized feedback on surgical abilities. Lavanchy et al62 created a ML algorithm capable of assessing the skill of laparoscopic cholecystectomies, which demonstrated 87% accuracy in identifying the kinematics of surgical instruments as a surrogate measure of efficiency.14-16,61 This provided constructive feedback to the operator and represents a system that could feasibly be translated into orthopaedics.62 The integration of AI systems (such as the Virtual Operative Assistant) into virtual reality (VR) and augmented reality (AR) can help to attain objective critique without depending on the typical ‘apprenticeship’ learning modality.61 Siemionow et al63 provided an example of successful AI incorporation into AR. The researchers developed a ML system enabling the overlay of a 3D spinal image onto cadavers, facilitating accurate metal probe placement into lumbar vertebrae.63 The overarching advantage of these technologies is patient safety, given surgical trainees can acquire experience while mitigating risk to patients.
Rehabilitation and postoperative care
The postoperative phase has been highlighted as a key area of AI interest.64 A growing body of studies have reported the use of smartphones to gather continuous, remote data on a patient’s vitals and rehabilitation progress following TKA.3,8 ML-based algorithms allow tracking of physiotherapy engagement and exercise participation, and can alert healthcare professionals if patient milestones are not met.65 Similarly, the surveillance of patients’ vitals, wellbeing, and complications, such as deep vein thrombosis, has been documented extensively in the literature.64,66,67 These AI features have been documented to reduce readmission rates following TKA and THA. However, no statistically significant difference in the rate of hospital discharge without remote monitoring has been reported.68,69 Interestingly, it has also been proposed that ML algorithms could prove a useful augment in rehabilitation following ACL surgery, by using biomechanical data to assess for asymmetries in gait analysis.32,70
DL has been touted by multiple studies to be capable of predicting the risk of complications leading to revision surgery, using postoperative hip arthroplasty radiographs. Rouzrokh et al71 found that a DL algorithm trained on over 90,000 postoperative images predicted implant dislocation within five years of surgery. This model had a rather high negative predictive value. However, this may still provide a useful ‘ruling out’ method for high-risk patients, and demonstrates the potential role of AI in guiding pre-emptive interventions.8,30,71
Limitations to AI in orthopaedics
AI is associated with considerable capital costs and financial burden on healthcare systems, potentially impeding its widespread adoption.1,65 Notwithstanding this, carefully designed cost-benefit analyses could delineate whether its utility in orthopaedics results in cost-effective interventions.72-74 The risk of breaching patient confidentiality is inherent with large data sets, and therefore should be treated as a prominent ethical consideration.1,3 As with any research being generalized to the wider clinical setting, AI models must go through a rigorous process of validation. Norgeot et al75 proposed a minimum set of documentation to bring similar levels of transparency and utility to the application of AI in medicine and surgery: minimum information about clinical AI modelling (MI-CLAIM). These guidelines involve six areas that require attention when appraising CNN models, and aim to inform clinical adoption of AI models: 1) study design; 2) separation of data into partitions for model training and model testing; 3) optimization and final model selection; 4) performance evaluation; 5) model examination; and 6) reproducible pipeline.75
The application of AI models outside of the data or institution of which it is designed (external validity) should be carefully considered. Systematic errors within algorithms could lead to negligent and widespread implications for patients. Accordingly, a systematic approach to designing and validating models using proven concepts is required to avoid such errors before translation into clinical practice. To mitigate this risk, AI is intended as an adjunct to the clinical decision-making process, not a substitute. Clinicians should remain cognisant of AI’s limitations and proceed cautiously, until external validity is proven within acceptable margins of error.3,37
AI is as good as its data, and the development of robust reporting frameworks is vital to preventing avoidable errors.70,76-78 Guidelines for establishing models are necessary, such as the Transparent Reporting of a multivariate prediction model for Individual Prognosis or Diagnosis (TRIPOD) initiative,77 which has already been used in validating ML in orthopaedics. The complexity of CNNs usually depends upon the complexity of the input data. The more convoluted the input data are, the more comprehensive mathematical algorithms are required to deliver the desired output. An inherent problem described repeatedly in the literature is the generation of complex CNNs, solely reflecting the data they are set up to evaluate.70 Overfitting refers to a model that fails generalizability well after model training, and is particularly common with models that are nonparametric/nonlinear and have more flexibility when learning the target function. The lack of generalizability may be attributed to the model learning the random fluctuation details and noise as part of the training data. Therefore, when translated to external data sets the model is unable to recognize the new patterns as efficiently. Furthermore, this issue occurs when the model mirrors and focuses entirely on minor characteristics present in the training data set instead of perceiving more generalized patterns beneath the data, and therefore requires continuous learning with larger volumes of data.34,77,78 It is vital for clinicians to be aware of this risk, and a collective effort is needed from multiple stakeholders to ensure appropriate collection, curation, and annotation of data that are validated beyond a given institution.70
Conclusion and future considerations
The use of AI in orthopaedics bears the potential to improve patient outcomes and reduce the workload of healthcare professionals. An auspicious future development is the innovative ‘digital twin’ pertaining to a virtual representation of oneself. This is thought to be at the cornerstone of precision medicine, able to predict diseases, treatment outcomes, and preventive interventions tailored to the individual patient phenotype, even down to the genome level. The effect this could have on the evolution of orthopaedic surgery and medicine is almost incomprehensible. AI in orthopaedic surgery shows promise in identifying hip and knee implants, highlighting implant malposition, detecting features of loosening, and predicting length of hospital stay, costs involved, functional outcomes, and prognostic scores. The current state of AI technology requires a coordinated effort to effectively progress from proof-of-concept into clinical practice. In this vein, the establishment of systematic and robust validation and reporting frameworks is of utmost importance to allow a safe adoption of this technology.
References
1. Han X-G , Tian W . Artificial intelligence in orthopedic surgery: current state and future perspective . Chin Med J . 2019 ; 132 ( 21 ): 2521 – 2523 . Crossref PubMed Google Scholar
2. Hashimoto DA , Rosman G , Rus D , Meireles OR . Artificial intelligence in surgery: Promises and perils . Ann Surg . 2018 ; 268 ( 1 ): 70 – 76 . Crossref PubMed Google Scholar
3. Myers TG , Ramkumar PN , Ricciardi BF , Urish KL , Kipper J , Ketonis C . Artificial intelligence and orthopaedics: An introduction for clinicians . J Bone Joint Surg Am . 2020 ; 102-A ( 9 ): 830 – 840 . Crossref PubMed Google Scholar
4. Bernstein J . Not the last word: ChatGPT can’t perform orthopaedic surgery . Clin Orthop Relat Res . 2023 ; 481 ( 4 ): 651 – 655 . Crossref PubMed Google Scholar
5. Soldozy S , Young S , Yağmurlu K , et al. Transsphenoidal surgery using robotics to approach the sella turcica: Integrative use of artificial intelligence, realistic motion tracking and telesurgery . Clin Neurol Neurosurg . 2020 ; 197 : 106152 . Crossref PubMed Google Scholar
6. Kalis B , Collier M , Fu R . 10 Promising AI Applications in Health Care . Harvard Business Review . 2018 . https://hbr.org/2018/05/10-promising-ai-applications-in-health-care ( date last accessed 23 June 2023 ). Google Scholar
7. Magan AA , Kayani B , Chang JS , Roussot M , Moriarty P , Haddad FS . Artificial intelligence and surgical innovation: lower limb arthroplasty . Br J Hosp Med (Lond) . 2020 ; 81 ( 10 ): 1 – 7 . Crossref PubMed Google Scholar
8. Kurmis AP , Ianunzio JR . Artificial intelligence in orthopedic surgery: evolution, current state and future directions . Arthroplasty . 2022 ; 4 ( 1 ): 9 . Crossref PubMed Google Scholar
9. Sun T , He X , Song X , Shu L , Li Z . The digital twin in medicine: A key to the future of healthcare? Front Med . 2022 ; 9 : 907066 . Crossref PubMed Google Scholar
10. Kayani B , Konan S , Huq SS , Ibrahim MS , Ayuob A , Haddad FS . The learning curve of robotic-arm assisted acetabular cup positioning during total hip arthroplasty . Hip Int . 2021 ; 31 ( 3 ): 311 – 319 . Crossref PubMed Google Scholar
11. Kayani B , Konan S , Thakrar RR , Huq SS , Haddad FS . Assuring the long-term total joint arthroplasty . Bone Joint J . 2019 ; 101-B ( 1_Supple_A ): 11 – 18 . Crossref PubMed Google Scholar
12. Fontalis A , Epinette J-A , Thaler M , Zagra L , Khanduja V , Haddad FS . Advances and innovations in total hip arthroplasty . SICOT J . 2021 ; 7 : 26 . Crossref PubMed Google Scholar
13. Fontalis A , Raj RD , Kim WJ , et al. Functional implant positioning in total hip arthroplasty and the role of robotic-arm assistance . Int Orthop . 2023 ; 47 ( 2 ): 573 – 584 . Crossref PubMed Google Scholar
14. Fontalis A , Kayani B , Asokan A , et al. Inflammatory response in robotic-arm-assisted versus conventional jig-based TKA and the correlation with early functional outcomes: Results of a prospective randomized controlled trial . J Bone Joint Surg Am . 2022 ; 104-A ( 21 ): 1905 – 1914 . Crossref PubMed Google Scholar
15. Kayani B , Tahmassebi J , Ayuob A , Konan S , Oussedik S , Haddad FS . A prospective randomized controlled trial comparing the systemic inflammatory response in conventional jig-based total knee arthroplasty versus robotic-arm assisted total knee arthroplasty . Bone Joint J . 2021 ; 103-B ( 1 ): 113 – 122 . Crossref PubMed Google Scholar
16. Chang JS , Kayani B , Wallace C , Haddad FS . Functional alignment achieves soft-tissue balance in total knee arthroplasty as measured with quantitative sensor-guided technology . Bone Joint J . 2021 ; 103-B ( 3 ): 507 – 514 . Crossref PubMed Google Scholar
17. Karasavvidis T , Pagan Moldenhauer CA , Haddad FS , Hirschmann MT , Pagnano MW , Vigdorchik JM . Current concepts in alignment in total knee arthroplasty . J Arthroplasty . 2023 ; S0883-5403(23)00080-3 . Crossref PubMed Google Scholar
18. MacDessi SJ , Oussedik S , Abdel MP , Victor J , Pagnano MW , Haddad FS . The language of knee alignment: updated definitions and considerations for reporting outcomes in total knee arthroplasty . Bone Joint J . 2023 ; 105-B ( 2 ): 101 . Crossref PubMed Google Scholar
19. Farrow L , Zhong M , Ashcroft GP , Anderson L , Meek RMD . Interpretation and reporting of predictive or diagnostic machine-learning research in Trauma & Orthopaedics . Bone Joint J . 2021 ; 103-B ( 12 ): 1754 – 1758 . Crossref PubMed Google Scholar
20. Farhadi F , Barnes MR , Sugito HR , Sin JM , Henderson ER , Levy JJ . Applications of artificial intelligence in orthopaedic surgery . Front Med Technol . 2022 ; 4 : 995526 . Crossref PubMed Google Scholar
21. Leung K , Zhang B , Tan J , et al. Prediction of total knee replacement and diagnosis of osteoarthritis by using deep learning on knee radiographs: Data from the Osteoarthritis Initiative . Radiology . 2020 ; 296 ( 3 ): 584 – 593 . Crossref PubMed Google Scholar
22. Baessler AM , Brolin TJ , Azar FM , et al. Development and validation of a predictive model for outcomes in shoulder arthroplasty: a multicenter analysis of nearly 2000 patients . J Shoulder Elbow Surg . 2021 ; 30 ( 12 ): 2698 – 2702 . Crossref PubMed Google Scholar
23. Kumar V , Patel S , Baburaj V , Vardhan A , Singh PK , Vaishya R . Current understanding on artificial intelligence and machine learning in orthopaedics - A scoping review . J Orthop . 2022 ; 34 : 201 – 206 . Crossref PubMed Google Scholar
24. Ackermann J , Wieland M , Hoch A , et al. A New Approach to Orthopedic Surgery Planning Using Deep Reinforcement Learning and Simulation . In de Bruijne M , Cattin PC , Cotin S , et al. Medical Image Computing and Computer Assisted Intervention – MICCAI 2021 . Vol 12904 . Cham : Springer , 2021 : 540 – 549 . Crossref Google Scholar
25. Purnomo G , Yeo S-J , Liow MHL . Artificial intelligence in arthroplasty . Arthroplasty . 2021 ; 3 ( 1 ): 37 . Crossref PubMed Google Scholar
26. Beyaz S , Açıcı K , Sümer E . Femoral neck fracture detection in X-ray images using deep learning and genetic algorithm approaches . Jt Dis Relat Surg . 2020 ; 31 ( 2 ): 175 – 183 . Crossref PubMed Google Scholar
27. Yamashita R , Nishio M , Do RKG , Togashi K . Convolutional neural networks: an overview and application in radiology . Insights Imaging . 2018 ; 9 ( 4 ): 611 – 629 . Crossref PubMed Google Scholar
28. Fu S , Wyles CC , Osmon DR , et al. Automated detection of periprosthetic joint infections and data elements using natural language processing . J Arthroplasty . 2021 ; 36 ( 2 ): 688 – 692 . Crossref PubMed Google Scholar
29. Wyatt JM , Booth GJ , Goldman AH . Natural language processing and its use in orthopaedic research . Curr Rev Musculoskelet Med . 2021 ; 14 ( 6 ): 392 – 396 . Crossref PubMed Google Scholar
30. Yi PH , Mutasa S , Fritz J . AI MSK clinical applications: orthopedic implants . Skeletal Radiol . 2022 ; 51 ( 2 ): 305 – 313 . Crossref PubMed Google Scholar
31. Parvizi J , Gehrke T , International Consensus Group on Periprosthetic Joint Infection . Definition of periprosthetic joint infection . J Arthroplasty . 2014 ; 29 ( 7 ): 1331 . Crossref PubMed Google Scholar
32. Corban J , Lorange J-P , Laverdiere C , et al. Artificial intelligence in the management of anterior cruciate ligament injuries . Orthop J Sports Med . 2021 ; 9 ( 7 ): 23259671211014210 . Crossref PubMed Google Scholar
33. Pranata YD , Wang K-C , Wang J-C , et al. Deep learning and SURF for automated classification and detection of calcaneus fractures in CT images . Comput Methods Programs Biomed . 2019 ; 171 : 27 – 37 . Crossref PubMed Google Scholar
34. Innocenti B , Radyul Y , Bori E . The use of artificial intelligence in orthopedics: Applications and limitations of machine learning in diagnosis and prediction . Applied Sciences . 2022 ; 12 ( 21 ): 10775 . Crossref Google Scholar
35. Langerhuizen DWG , Bulstra AEJ , Janssen SJ , et al. Is deep learning on par with human observers for detection of radiographically visible and occult fractures of the scaphoid? Clin Orthop Relat Res . 2020 ; 478 ( 11 ): 2653 – 2659 . Crossref PubMed Google Scholar
36. Gyftopoulos S , Lin D , Knoll F , Doshi AM , Rodrigues TC , Recht MP . Artificial intelligence in musculoskeletal imaging: Current status and future directions . Am J Roentgenol . 2019 ; 213 ( 3 ): 506 – 513 . Crossref PubMed Google Scholar
37. Langerhuizen DWG , Janssen SJ , Mallee WH , et al. What are the applications and limitations of artificial intelligence for fracture detection and classification in orthopaedic trauma imaging? A systematic review . Clin Orthop Relat Res . 2019 ; 477 ( 11 ): 2482 – 2491 . Crossref PubMed Google Scholar
38. Oliveira E Carmo L , van den Merkhof A , Olczak J , et al. An increasing number of convolutional neural networks for fracture recognition and classification in orthopaedics: are these externally validated and ready for clinical application? Bone Jt Open . 2021 ; 2 ( 10 ): 879 – 885 . Crossref PubMed Google Scholar
39. Olczak J , Pavlopoulos J , Prijs J , et al. Presenting artificial intelligence, deep learning, and machine learning studies to clinicians and healthcare stakeholders: an introductory reference with a guideline and a Clinical AI Research (CAIR) checklist proposal . Acta Orthop . 2021 ; 92 ( 5 ): 513 – 525 . Crossref PubMed Google Scholar
40. Park C-W , Oh S-J , Kim K-S , et al. Artificial intelligence-based classification of bone tumors in the proximal femur on plain radiographs: System development and validation . PLoS ONE . 2022 ; 17 ( 2 ): e0264140 . Crossref PubMed Google Scholar
41. Han I , Kim JH , Park H , Kim H-S , Seo SW . Deep learning approach for survival prediction for patients with synovial sarcoma . Tumour Biol . 2018 ; 40 ( 9 ): 1010428318799264 . Crossref PubMed Google Scholar
42. Xie X , Li Z , Bai L , et al. Deep learning-based MRI in diagnosis of fracture of tibial plateau combined with meniscus injury . Scientific Programming . 2021 ; 2021 : 1 – 8 . Crossref Google Scholar
43. Cheng K , Guo Q , He Y , et al. Artificial intelligence in sports medicine: Could GPT-4 make human doctors obsolete? Ann Biomed Eng . 2023 ; Epub ahead of print . Crossref PubMed Google Scholar
44. Archer H , Reine S , Alshaikhsalama A , et al. Artificial intelligence-generated hip radiological measurements are fast and adequate for reliable assessment of hip dysplasia: an external validation study . Bone Jt Open . 2022 ; 3 ( 11 ): 877 – 884 . Crossref PubMed Google Scholar
45. Nguyen TP , Chae DS , Park SJ , Yoon J . A novel approach for evaluating bone mineral density of hips based on Sobel gradient-based map of radiographs utilizing convolutional neural network . Comput Biol Med . 2021 ; 132 : 104298 . Crossref PubMed Google Scholar
46. Al-Hourani K , Tsang STJ , Simpson AHRW . Osteoporosis: current screening methods, novel techniques, and preoperative assessment of bone mineral density . Bone Joint Res . 2021 ; 10 ( 12 ): 840 – 843 . Crossref PubMed Google Scholar
47. Jang SJ , Kunze KN , Brilliant ZR , et al. Comparison of tibial alignment parameters based on clinically relevant anatomical landmarks: a deep learning radiological analysis . Bone Jt Open . 2022 ; 3 ( 10 ): 767 – 776 . Crossref PubMed Google Scholar
48. Gurung B , Liu P , Harris PDR , et al. Artificial intelligence for image analysis in total hip and total knee arthroplasty: a scoping review . Bone Joint J . 2022 ; 104-B ( 8 ): 929 – 937 . Crossref PubMed Google Scholar
49. Borjali A , Chen AF , Muratoglu OK , Morid MA , Varadarajan KM . Detecting total hip replacement prosthesis design on plain radiographs using deep convolutional neural network . J Orthop Res . 2020 ; 38 ( 7 ): 1465 – 1471 . Crossref PubMed Google Scholar
50. Borjali A , Chen AF , Muratoglu OK , Morid MA , Varadarajan KM . Detecting mechanical loosening of total hip replacement implant from plain radiograph using deep convolutional neural network . Cornell University . 2019 . https://arxiv.org/abs/1912.00943v2 ( date last accessed 23 June 2023 ). Google Scholar
51. Wei C , Quan T , Wang KY , et al. Artificial neural network prediction of same-day discharge following primary total knee arthroplasty based on preoperative and intraoperative variables . Bone Joint J . 2021 ; 103-B ( 8 ): 1358 – 1366 . Crossref PubMed Google Scholar
52. Harris AHS , Kuo AC , Weng Y , Trickey AW , Bowe T , Giori NJ . Can machine learning methods produce accurate and easy-to-use prediction models of 30-day complications and mortality after knee or hip arthroplasty? Clin Orthop Relat Res . 2019 ; 477 ( 2 ): 452 – 460 . Crossref PubMed Google Scholar
53. Jo C , Ko S , Shin WC , et al. Transfusion after total knee arthroplasty can be predicted using the machine learning algorithm . Knee Surg Sports Traumatol Arthrosc . 2020 ; 28 ( 6 ): 1757 – 1764 . Crossref PubMed Google Scholar
54. Hirvasniemi J , Gielis WP , Arbabi S , et al. Bone texture analysis for prediction of incident radiographic hip osteoarthritis using machine learning: data from the Cohort Hip and Cohort Knee (CHECK) study . Osteoarthritis Cartilage . 2019 ; 27 ( 6 ): 906 – 914 . Crossref PubMed Google Scholar
55. Kim JS , Arvind V , Oermann EK , et al. Predicting surgical complications in patients undergoing elective adult spinal deformity procedures using machine learning . Spine Deform . 2018 ; 6 ( 6 ): 762 – 770 . Crossref PubMed Google Scholar
56. McDonnell JM , Evans SR , McCarthy L , et al. The diagnostic and prognostic value of artificial intelligence and artificial neural networks in spinal surgery: a narrative review . Bone Joint J . 2021 ; 103-B ( 9 ): 1442 – 1448 . Crossref PubMed Google Scholar
57. Kumar V , Roche C , Overman S , et al. What is the accuracy of three different machine learning techniques to predict clinical outcomes after shoulder arthroplasty? Clin Orthop Relat Res . 2020 ; 478 ( 10 ): 2351-2363 . Crossref PubMed Google Scholar
58. King GJ , Richards RR , Zuckerman JD , et al. A standardized method for assessment of elbow function. Research Committee, American Shoulder and Elbow Surgeons . J Shoulder Elbow Surg . 1999 ; 8 ( 4 ): 351 – 354 . PubMed Crossref Google Scholar
59. Kumar V , Schoch BS , Allen C , et al. Using machine learning to predict internal rotation after anatomic and reverse total shoulder arthroplasty . J Shoulder Elbow Surg . 2022 ; 31 ( 5 ): e234 – e245 . Crossref PubMed Google Scholar
60. Karnuta JM , Churchill JL , Haeberle HS , et al. The value of artificial neural networks for predicting length of stay, discharge disposition, and inpatient costs after anatomic and reverse shoulder arthroplasty . J Shoulder Elbow Surg . 2020 ; 29 ( 11 ): 2385 – 2394 . Crossref PubMed Google Scholar
61. Guerrero DT , Asaad M , Rajesh A , Hassan A , Butler CE . Advancing surgical education: The use of artificial intelligence in surgical training . Am Surg . 2023 ; 89 ( 1 ): 49 – 54 . Crossref PubMed Google Scholar
62. Lavanchy JL , Zindel J , Kirtac K , et al. Author Correction: Automation of surgical skill assessment using a three-stage machine learning algorithm . Sci Rep . 2021 ; 11 ( 1 ): 8933 . Crossref PubMed Google Scholar
63. Siemionow KB , Katchko KM , Lewicki P , Luciano CJ . Augmented reality and artificial intelligence-assisted surgical navigation: Technique and cadaveric feasibility study . J Craniovertebr Junction Spine . 2020 ; 11 ( 2 ): 81 – 85 . Crossref PubMed Google Scholar
64. Alsareii SA , Raza M , Alamri AM , et al. Machine learning and internet of things enabled monitoring of post-surgery patients: A pilot study . Sensors . 2022 ; 22 ( 4 ): 1420 . Crossref PubMed Google Scholar
65. Batailler C , Shatrov J , Sappey-Marinier E , Servien E , Parratte S , Lustig S . Artificial intelligence in knee arthroplasty: current concept of the available clinical applications . Arthroplasty . 2022 ; 4 ( 1 ): 17 . Crossref PubMed Google Scholar
66. Kim KM , Yefimova M , Lin FV , Jopling JK , Hansen EN . A Home-Recovery Surgical Care Model Using AI-Driven Measures of Activities of Daily Living . NEJM Catal . 2022 . 10.1056/CAT.22.0081 . https://catalyst.nejm.org/doi/full/10.1056/CAT.22.0081 ( date last accessed 23 June 2023 ). Google Scholar
67. Breteler MJM , Numan L , Ruurda JP , et al. Wireless remote home monitoring of vital signs in patients discharged early after esophagectomy: Observational feasibility study . JMIR Perioper Med . 2020 ; 3 ( 2 ): e21705 . Crossref PubMed Google Scholar
68. Mehta SJ , Hume E , Troxel AB , et al. Effect of remote monitoring on discharge to home, return to activity, and rehospitalization after hip and knee arthroplasty: A randomized clinical trial . JAMA Netw Open . 2020 ; 3 ( 12 ): e2028328 . Crossref PubMed Google Scholar
69. Ramkumar PN , Haeberle HS , Ramanathan D , et al. Remote patient monitoring using mobile health for total knee arthroplasty: Validation of a wearable and machine learning-based surveillance platform . J Arthroplasty . 2019 ; 34 ( 10 ): 2253 – 2259 . Crossref PubMed Google Scholar
70. Kunze KN , Orr M , Krebs V , Bhandari M , Piuzzi NS . Potential benefits, unintended consequences, and future roles of artificial intelligence in orthopaedic surgery research: a call to emphasize data quality and indications . Bone Jt Open . 2022 ; 3 ( 1 ): 93 – 97 . Crossref PubMed Google Scholar
71. Rouzrokh P , Ramazanian T , Wyles CC , et al. Deep learning artificial intelligence model for assessment of hip dislocation risk following primary total hip arthroplasty from postoperative radiographs . J Arthroplasty . 2021 ; 36 ( 6 ): 2197 – 2203 . Crossref PubMed Google Scholar
72. Rajan PV , Khlopas A , Klika A , Molloy R , Krebs V , Piuzzi NS . The cost-effectiveness of robotic-assisted versus manual total knee arthroplasty: A Markov model-based evaluation . J Am Acad Orthop Surg . 2022 ; 30 ( 4 ): 168 – 176 . Crossref PubMed Google Scholar
73. Close A , Robertson C , Rushton S , et al. Comparative cost-effectiveness of robot-assisted and standard laparoscopic prostatectomy as alternatives to open radical prostatectomy for treatment of men with localised prostate cancer: a health technology assessment from the perspective of the UK National Health Service . Eur Urol . 2013 ; 64 ( 3 ): 361 – 369 . Crossref PubMed Google Scholar
74. Sandhu J . 'Robosurgeons vs. robosceptics’: can we afford robotic technology or can we afford not to? Journal of Clinical Urology . 2019 ; 12 ( 4 ): 285 – 295 . Crossref Google Scholar
75. Norgeot B , Quer G , Beaulieu-Jones BK , et al. Minimum information about clinical artificial intelligence modeling: the MI-CLAIM checklist . Nat Med . 2020 ; 26 ( 9 ): 1320 – 1324 . Crossref PubMed Google Scholar
76. Prijs J , Liao Z , Ashkani-Esfahani S , et al. Artificial intelligence and computer vision in orthopaedic trauma: the why, what, and how . Bone Joint J . 2022 ; 104-B ( 8 ): 911 – 914 . Crossref PubMed Google Scholar
77. Vigdorchik JM , Jang SJ , Taunton MJ , Haddad FS . Deep learning in orthopaedic research: weighing idealism against realism . Bone Joint J . 2022 ; 104-B ( 8 ): 909 – 910 . Crossref PubMed Google Scholar
78. Polisetty TS , Jain S , Pang M , et al. Concerns surrounding application of artificial intelligence in hip and knee arthroplasty . Bone Joint J . 2022 ; 104-B ( 12 ): 1292 – 1303 . Crossref PubMed Google Scholar
Author contributions
A. B. Lisacek-Kiosoglous: Data curation, Investigation, Methodology, Project administration, Resources, Validation, Visualization, Writing – original draft, Writing – review & editing.
A. S. Powling: Data curation, Investigation, Methodology, Project administration, Resources, Validation, Visualization, Writing – original draft, Writing – review & editing.
A. Fontalis: Conceptualization, Data curation, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing.
A. Gabr: Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing.
E. Mazomenos: Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing.
F. S. Haddad: Conceptualization, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing.
Funding statement
The authors received no financial or material support for the research, authorship, and/or publication of this article.
ICMJE COI statement
E. Mazomenos reports institutional grants from Wellcome/EPSRC Centre for Interventional and Surgical Sciences (WEISS) (Awards WT:203145Z/16/Z, EPSRC: NS/A000050/1), related to this study. F. S. Haddad is Editor-in-Chief of The Bone & Joint Journal.
Data sharing
The data that support the findings for this study are available to other researchers from the corresponding author upon reasonable request.
Acknowledgements
Andreas Fontalis would like to acknowledge and thank the Onassis Foundation for the financial support of his PhD studies - Scholarship ID: F ZR 065-1/2021-2022.
© 2023 Author(s) et al. This is an open-access article distributed under the terms of the Creative Commons Attribution Non-Commercial No Derivatives (CC BY-NC-ND 4.0) licence, which permits the copying and redistribution of the work only, and provided the original author and source are credited. See https://creativecommons.org/licenses/by-nc-nd/4.0/
