EDP Sciences logo
Submit your paper
Search
Menu
All issues Volume 10 (2024) SICOT-J, 10 (2024) 49 Full HTML
Open Access
Review
Issue
SICOT-J
Volume 10, 2024
Article Number 49
Number of page(s) 8
Section Knee
DOI https://doi.org/10.1051/sicotj/2024044
Published online 21 November 2024

© The Authors, published by EDP Sciences, 2024

Licence Creative CommonsThis is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Introduction

Artificial intelligence (AI) refers to the application of computer algorithms that provide machines with the ability to perform tasks innately characteristic of human intelligence [1]. It encompasses various subfields, including machine learning (ML), deep learning (DL), and natural language processing (NLP) [1]. Grounded in principles of pattern recognition and iterative improvement through training on large datasets, these algorithms can evolve from initially requiring human input to functioning autonomously [2].

Musculoskeletal conditions are the most significant contributor to global disability, with approximately 1.71 billion people afflicted [3]. As such, current service demand for planned orthopaedic surgical procedures is straining healthcare systems globally, exacerbated by the COVID-19 pandemic and the effects of an ageing population [4, 5]. Patients face prolonged waiting times for necessary treatment, resulting in further deterioration and potentially poorer clinical outcomes when they eventually progress to surgery [6].

AI holds promise in transforming planned orthopaedics with wide-ranging applications throughout the perioperative period, encompassing automated diagnostic imaging, clinical decision-making tools, optimisation of implant design, robotic surgery, and remote patient monitoring [1]. Incorporation of these tools into the clinical setting could optimise patient care and streamline system efficiency and thus potentially reduce waiting lists. However, despite the ever-increasing body of orthopaedic research in AI, challenges remain, and current application in clinical practice is limited [7, 8]. This review explores AI’s role in planned orthopaedic care by examining applications across preoperative, intraoperative, and postoperative stages (Figure 1).

thumbnail Figure 1

The applications of artificial intelligence in planned orthopaedic care discussed in this article. AI = artificial intelligence, OA = osteoarthritis, TKA = total knee arthroplasty, THA = total hip arthroplasty, NLP = natural language processing, EPR = electronic patient records, PROM = patient-reported outcome measure,, AR = augmented reality.

Preoperative

AI is poised to play a pivotal role preoperatively in diagnostic imaging, patient selection, clinical outcome prediction, risk stratification, and surgical planning.

Imaging

Traditionally, osteoarthritis (OA) diagnosis and classification has relied upon manual radiographic assessment; however, multiple AI algorithms capable of automated detection and staging have now been developed [9–12]. Xue et al. [9] created a model to diagnose hip OA by training a previously validated convolutional neural network (CNN) for image recognition on 420 hip radiographs. The model demonstrated promising diagnostic aptitude, with a performance comparable to that of a physician with ten years of experience [9]. Beyond diagnosis, AI can further categorise the severity of knee OA using the Kellgren-Lawrence (KL) grading system with good accuracy and improved reproducibility compared to manual evaluation [11, 12]. Uniquely, one study using a Deep Siamese CNN architecture provided probability distributions for each KL grade, which may be of clinical utility in grading ambiguous cases [13]. However, it’s important to note that when an algorithm is trained on radiographs from a single institution, generalisability may be encumbered by different labelling systems, radiation exposure and protocols, thus making it invalid [14].

In the context of total joint replacement (TJR), which constitutes a significant portion of planned orthopaedic procedures, a growing body of research has demonstrated the potential of ML and DL models capable of outperforming traditional logarithmic regression (LR) to predict structural disease progression of OA and identify individuals at high risk of TJR [15–19]. Guan et al. [15] developed a DL programme using baseline knee radiographs to predict medial joint space loss over 48 months. The study compared a traditional model incorporating demographic and radiographic risk factors with a DL model trained on 2,300 radiographs from the Osteoarthritis Initiative. The DL model demonstrated significantly higher accuracy, with an AUC of 0.799 compared to 0.660 of the traditional model [15]. Similarly, Leung et al. [18] found that applying DL to predict total knee arthroplasty (TKA) based solely on knee radiographs outperformed conventional binary outcome models, attaining high sensitivity and specificity, 83% and 87%, respectively [13]. Despite the efficacy of these models, there is a well-documented discrepancy between radiologic severity and clinical symptoms in OA; consequently, the decision for TKA is primarily based on symptom severity rather than radiographic findings alone [13, 20, 21]. Therefore, although ML algorithms that automate OA grading or predict arthroplasty based solely on imaging have limited clinical utility, they provide valuable insights for developing more comprehensive predictive tools. Future studies incorporating additional clinical parameters alongside imaging data may yield more clinically relevant tools to support surgical decision-making [13].

A further application of AI-automated pathology quantification lies in the accurate evaluation of limb alignment, which is crucial for planning corrective surgeries for malalignment and joint replacement. Lee et al. [22] trained and tested a DL-based algorithm to quantify four alignment parameters of the knee; on external validation, not only did the measurement angles strongly correlate with those performed manually, but it was also 3.44 times faster. Similarly, in the context of preoperative planning for TKA, Vidhani and colleagues validated a DL and computer vision-based framework for measuring five hip and knee anatomic parameters from CT scans; the average time taken was 2.58 (±1.92) minutes per patient with no significant difference from manual measurements [23]. Central to both systems is the incorporation of CNNs for automatic image segmentation, precisely identifying the specific regions of interest from which measurements are derived [14].

As the number of total hip arthroplasty (THA) procedures increases, revisions are anticipated to rise by between 43% to 70% by 2030 [24]. Preoperative planning involves correctly identifying the implant type; this can be difficult and time-consuming, typically relying on radiographs, medical records, and implant labels. Although numerous AI-powered software programs have been developed to identify implants automatically, they often face limitations such as small training datasets, training exclusively on AP radiographs, and lack of acetabular classification [25]. Addressing these challenges, Rouzrokh et al. [25] developed a DL tool for THA automatic implant detection, trained on over 240,00 radiographs. It can identify common designs of 20 femoral and 8 acetabular components from various views (AP, lateral, or oblique), achieving an internal accuracy of 98.9% for both components and 97.0% accuracy in external validation for femoral components [25]. To explain model performance, they leveraged gradient mapping to highlight the radiographic features affecting network decisions. A unique aspect of their model is the incorporation of uncertainty quantification (UQ) to gauge the confidence in the model’s output, along with outlier detection to identify data that may compromise results. These features enhance the model’s trustworthiness, supporting its potential for clinical deployment [25].

Patient selection

Prolonged waiting times adversely impact patients’ preoperative well-being and may detract from the potential gains of arthroplasty surgery [6]. Automated stratification of patients based on routine EPR data through NLP could be employed to guide clinical pathways and streamline the referral process for operative intervention, thus improving system efficiency. NLP enables computers to understand and analyse human language by translating it into structured data, overcoming the limitations of manual processing in terms of timeliness, reproducibility, and scalability [1, 26].

Patients seeking treatment through the UK’s National Health Service (NHS) rely on referrals from non-specialist general practitioners (GPs). Farrow et al. [4] are exploring the use of AI to assist GPs in correctly and rapidly identifying suitable candidates for arthroplasty surgery as part of the AI to Revolutionise the Patient Care Pathway in Hip and Knee Arthroplasty (ARCHERY) project. The study hinges on employing NLP to synthesise patient information, such as imaging reports, clinical letters, patient health records, and patient-reported outcome measures (PROMs), to predict if a patient would be suitable for surgery. Preliminary results discussed at the British Hip Society meeting in February 2024, focusing on their THA model, indicated that NLP fails to perform accurate clinical inference for identifying patients appropriate for THA, with performance deteriorating further on external validation [27]. This highlights the challenge of applying NLP across different healthcare settings or globally, where heterogeneity of clinical terminology, free text information, and criteria for surgical candidacy exist [28].

Clinical outcome prediction

PROMs serve as a quality benchmark in arthroplasty. As many as 20–30% of patients who undergo TJR report no improvement or dissatisfaction with their surgical results [29, 30]. Identifying patients unlikely to achieve long-term, meaningful gains from surgery is crucial for setting patient expectations and informing presurgical decision-making, which could be advantageous in reducing costs to healthcare systems by preventing patients unlikely to benefit from receiving unnecessary treatment [2, 13].

Multiple ML models have been developed to predict patient dissatisfaction, offering potentially valuable adjuncts to support shared decision-making [13, 31–34]. When high-risk patients are recognised, interventions such as preoperative education and health optimisation may be implemented, while others may choose to seek non-surgical options [34, 35]. Liu et al. [33] trained three ML models (image-only, clinical-data-only, and multimodal) on a cohort of 5,720 knee OA patients to predict postoperative dissatisfaction at two years. Dissatisfaction was defined as failing to achieve the minimal clinically important difference (MCID) – the minimum change considered beneficial or meaningful – corresponding to three PROM scoring systems, including the Oxford Knee Score (OKS). The clinical-data-only and multimodal models demonstrated excellent performance, with AUCs of 0.806 and 0.816, respectively. Preoperative functional scores and mental health status were among the most influential features correlating with dissatisfaction [33].

Using PROMs as a metric for patient dissatisfaction or attempting to quantify a subjective measure for AI prediction has inherent limitations. Challenges lie in their inability to capture the nuanced outcomes of surgery, compounded by the ceiling effect whereby patients who are relatively well before surgery may show modest measurable improvement postoperatively, as the potential for significant change is constrained [31]. Hunter and colleagues demonstrated this in their study, where an unsupervised ML programme identified clusters of patients with predictable PROM changes [31]. Those with the worst preoperative scores experienced the greatest improvements, conversely, patients who were relatively well, or whose quality of life was less affected preoperatively, did not show as marked a change in PROM despite their clinical outcomes being adequate. Suggesting that while AI models can be useful in predicting dissatisfaction, they may be less effective in predicting outcomes for those with less severe baseline symptoms. Therefore, relying solely on PROM scores to generate AI models could oversimplify the complexities of patient experiences, underscoring the need for more comprehensive and personalised tools in assessing surgical outcomes.

Risk stratification

AI-based algorithms that predict transfusion risk, periprosthetic joint infection (PJI), length of stay (LOS), and unplanned readmission could be used to forecast patient outcomes following orthopaedic interventions [36–44]. These models can assist in preoperative optimisation, resource management, and treatment planning, improving patient care and healthcare efficiency.

Jo et al. [38] conducted the first study leveraging ML to predict blood transfusion after TKA using six preoperative variables (haemoglobin, platelet count, type of surgery, tranexamic acid use, age, and body weight). The model demonstrated good performance with an AUC of 0.880 on external validation [38]. Conversely, in their study, Harris et al. [37] found that while models performed with good accuracy at discriminating the likelihood of patients experiencing renal or cardiac complications and death within 30 days of arthroplasty, on external validation, they were no longer robust in predicting renal complications, demonstrating the need for further development prior to clinical adoption. They postulated that incorporating NLP analysis of clinical progress notes and radiographs could be used to improve predictive accuracy [37].

PJIs represent a severe complication of TJA, leading to increased morbidity, readmissions, and substantial healthcare costs [45]. Predicting PJI accurately is challenging due to the numerous patient and surgical risk factors [46]. AI could be deployed clinically to improve PJI prevention and management by assessing risk and assisting in diagnosis, antibiotic selection, and prognosis [47–50]. Yeo et al. [48] developed an artificial neuronal network to predict superficial surgical site infections and PJIs following TKA. The model was trained on 10,021 patients with an average follow-up of 2.8 years and found predictors included obesity, Charlson Comorbidity Index, and smoking, with obesity emerging as a stronger predictor of infection than previously thought, possibly due to the increased accuracy of data analysis provided by ML algorithms in handling complex datasets [48]. Furthermore, an ML system outperformed the widely used International Consensus Meeting (ICM) criteria for diagnosing PJI [49, 51]. Unlike the fixed ICM criteria, which relies on factors that may be infrequent or expensive to assess, ML can construct adaptive diagnostic models based on available data, incorporating explainability features such as “if-then” decision paths to increase transparency and clinician trust [49].

LOS is the principal cost driver in the episode of care, and reliable predictions could improve resource allocation and care equity, particularly when integrated into payment models that account for patient-specific risks rather than using a “one-size-fits-all” approach [39, 52]. Ramkumar et al. [39] devised an algorithm to predict payment based on LOS; however, as patient complexity increased, the error margin in payment prediction also grew, reaching 32% for patients with extreme comorbidities, underscoring the difficulties of managing complex cases in payment models [39]. Nevertheless, a key benefit of ML models is their ability to capture nonlinear relationships that traditional analytical methods may overlook, offering insight into the factors influencing outcomes. For example, in their LOS predictive model, Chen et al. [41] identified age, transfusion, BMI, operation time, and preoperative blood results (haematocrit, platelet count, and white blood cell count) as the most important predictors of LOS. In another study, NSAID use, systemic corticosteroids, and poor social support were significant predictors of readmission after TJR [43]. Based on such findings, clinicians may be able to allocate additional resources to target high-risk individuals more effectively.

However, using such predictive methods engenders concerns about the potential marginalisation of patients deemed too high risk who may be labelled as inoperable. Therefore, careful consideration is needed to ensure these models are applied equitably and do not exclude vulnerable individuals.

Surgical planning

Preoperative planning AI software can inform implant design, positioning, and 3D templating [53–56]. For example, Cilla et al. [53] applied ML to optimise short stem implant design to reduce stress shielding, while Jang and colleagues used DL to estimate patient-specific adjustments for optimal landmark usage in TKA planning [54].

In their retrospective review, Ding et al. evaluated the accuracy of an AI-based 3D preoperative planning software (AIHIP; Beijing Changmugu Medical Technology Co., Ltd., Beijing, China) for THA [55]. They found that it offered a more accurate surgical planning tool than traditional 2D templating by enabling precise visualisation of anatomical structures, facilitating reliable predictions of component size and implant position, surpassing the accuracy of manual techniques [55]. This leaves surgeons better prepared for surgery, helping to avoid unforeseen intraoperative challenges and reducing the time required for preoperative planning. That said, applying this technology incurs additional costs and radiation exposure, as it requires preoperative CT scans. Future AI applications may mitigate these drawbacks; for instance, Fernandes et al. [57] have developed an innovative AI program that can convert 2D radiographs into 3D bone models with high accuracy, reliability, and repeatability compared to CT scans.

Intraoperative

The integration of AI intraoperatively is rapidly expanding, with its incorporation into tools such as augmented reality (AR)-based navigation systems, AI-assisted arthroscopic procedures, and robotic surgery [36]. These innovations aim to support decision-making, minimise the risk of complications, and reduce inter-surgeon variability to deliver consistent and high-quality care between institutions [36, 58]. For instance, a navigation-assisted anchor screw insertion tool for shoulder arthroscopy has shown high repeatability and reproducibility in cadaveric studies, with both novice and expert operators attaining insertion of the anchor at an angle close to the predetermined target angle [58].

Robotic surgery, known for its minimally invasive approach coupled with greater precision, has been demonstrated to reduce blood loss, pain, and recovery time [59, 60]. AI’s role in this field is expanding with the scope for the integration of ML algorithms capable of refining surgical planning through data analysis from previous surgeries [36]. Additionally, real-time feedback during procedures involving bone cutting or soft tissue resection can enhance precision and reduce the likelihood of errors, thereby ensuring consistent and reproducible results [36]. Nonetheless, the widespread adoption of robotics into clinical practice faces various challenges. These include the substantial financial investment to acquire and maintain the equipment, the consumables used during procedures, system compatibility with only specific implant types, and the need for surgeon and staff training to ensure safe use [36].

As these technologies become increasingly integrated into modern practice, there are concerns regarding automation bias resulting in overreliance on technology. This could detract from surgeons’ clinical skills and, if technology fails, lead to severe consequences. Drawing parallels with the airline industry’s use of autopilot systems, it forces speculation as to whether surgeons may be primarily only called upon in emergencies should things go wrong. Moreover, it raises the question of whether surgical training should evolve to include knowledge and operation of these machines.

Postoperative

Critical to the success of surgery is the postoperative recovery phase, rehabilitation, and monitoring for complications. Through wearable devices and automated analysis of post-operative imaging, AI could be used to enhance this.

Remote patient monitoring

Internet of Things and mHealth refers to incorporating AI into wearable devices, sensors, and smartphone apps that can track various parameters and collect large volumes of real-time data, which clinicians can monitor to assess rehabilitation progress and intervene if milestones are unmet [1, 36]. Such innovations have shown value in reducing opioid use and improving therapeutic adherence with anticoagulation for venous thromboembolism prophylaxis [61, 62].

Ramkumar et al. [63] validated a motion-based ML software kit to assess arcs of shoulder motion. They applied this technology to devise a system for remote patient monitoring of mobility, home exercise compliance, and range of motion following TKA [64]. The system comprises a knee sleeve containing two Bluetooth sensors that transmit three-dimensional spatial orientation changes to a smartphone app. The app processes the data using the ML algorithm, while also collecting additional information entered by the patients, such as weekly PROMs and opioid consumption. Overall, it showed promise, with patients finding it motivating and engaging, and allowed mobility and compliance with physiotherapy to be evaluated remotely, potentially alerting clinicians to early complications [64].

Image analysis to screen for complications

The utility of AI in image analysis extends beyond the preoperative period; applications have also been tested postoperatively to detect early signs of implant loosening and determine the risk of dislocation with greater accuracy and efficiency than traditional methods [65, 66]. Through early detection, prompt intervention may be sought.

Prosthetic loosening is the leading cause of arthroplasty failure, accounting for 30% and 55% of revision surgeries following TKA and THA, respectively [67, 68]. Borjali et al. [65] developed a fully automated CNN algorithm to detect mechanical loosening of THA implants from plain film radiographs, reporting a lower specificity of 75.0% but a higher sensitivity of 91.6% compared to manual assessment by an expert orthopaedic surgeon, illustrating its potential role in supplementing the decision-making process, freeing time for clinicians to engage in more patient-centric activities.

Similarly, dislocation following THA presents a considerable challenge, resulting in severe pain, limb dysfunction, readmission, revision, and substantial financial implications, with treatment costs estimated to be three times higher than those with uncomplicated THA [69]. Identifying patients at risk of dislocation is crucial for tailoring postoperative rehabilitation protocols [70]. A CNN, trained on over 92,000 radiographs, has been developed to classify THA patients for dislocation risk based on single postoperative AP pelvis radiographs, achieving 89.0% sensitivity [66]. Given the burden of follow-up visits and limitations of hip precautions, those identified by CNN as low risk may be managed with fewer postoperative restrictions.

Conclusion

Incorporating AI-powered technologies across the perioperative period offers an opportunity to improve system efficiency and enhance patient care as a means of addressing the growing strain on planned orthopaedic services. Nevertheless, key challenges such as accuracy, generalisability, and transparency must be addressed. Few models discussed in this review have undergone external validation, and a high AUC does not necessarily equate to strong clinical performance. Furthermore, robust regulatory and quality frameworks must be concurrently created as metrics from which these models can be assessed to ensure safe and effective implementation.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Conflicts of interest

The authors declare that they have no relevant financial or non-financial interests to report.

Data availability statement

This article has no associated data generated.

Author contribution statement

Author 1: Conceptualisation, Writing an original draft.

Author 2: Conceptualisation, Writing, Reviewing, and Editing.

Author 3: Conceptualisation, Writing, Reviewing, and Editing.

Author 4: Conceptualisation, Supervision, Writing, Reviewing, and Editing.

Ethics approval

Ethical approval was not required for this review article.

Informed consent

This article does not contain any studies with human participants or animals performed by any authors.

References

  1. Myers TG, Ramkumar PN, Ricciardi BF, Urish KL, Kipper J, Ketonis C (2020) Artificial intelligence and orthopaedics: an introduction for clinicians. J Bone Joint Surg Am 102, 830–840. [CrossRef] [PubMed] [Google Scholar]
  2. Farhadi F, Barnes MR, Sugito HR, Sin JM, Henderson ER, Levy JJ (2022) Applications of artificial intelligence in orthopaedic surgery. Front Med Technol 4, 995526. [CrossRef] [PubMed] [Google Scholar]
  3. Cieza A, Causey K, Kamenov K, Hanson SW, Chatterji S, Vos T (2021) Global estimates of the need for rehabilitation based on the Global Burden of Disease study 2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 396, 2006–2017. [Google Scholar]
  4. Farrow L, Ashcroft GP, Zhong M, Anderson L (2022) Using artificial intelligence to revolutionise the patient care pathway in hip and knee arthroplasty (ARCHERY): protocol for the development of a clinical prediction model. JMIR Res Protoc 11, e37092. [CrossRef] [PubMed] [Google Scholar]
  5. Yapp LZ, Clarke JV, Moran M, Simpson A, Scott CEH (2021) National operating volume for primary hip and knee arthroplasty in the COVID-19 era: a study utilizing the Scottish arthroplasty project dataset. Bone Jt Open 2, 203–210. [CrossRef] [PubMed] [Google Scholar]
  6. Nikolova S, Harrison M, Sutton M (2016) The impact of waiting time on health gains from surgery: evidence from a National Patient-reported outcome dataset. Health Econ 25, 955–968. [CrossRef] [PubMed] [Google Scholar]
  7. He J, Baxter SL, Xu J, Xu J, Zhou X, Zhang K (2019) The practical implementation of artificial intelligence technologies in medicine. Nat Med 25, 30–36. [CrossRef] [PubMed] [Google Scholar]
  8. Cabitza F, Locoro A, Banfi G (2018) Machine Learning in Orthopedics: A Literature Review. Front Bioeng Biotechnol 6, 75. [CrossRef] [PubMed] [Google Scholar]
  9. Xue Y, Zhang R, Deng Y, Chen K, Jiang T (2017) A preliminary examination of the diagnostic value of deep learning in hip osteoarthritis. PLoS One 12, e0178992. [CrossRef] [PubMed] [Google Scholar]
  10. Üreten K, Arslan T, Gültekin KE, Demir AND, Özer HF, Bilgili Y (2020) Detection of hip osteoarthritis by using plain pelvic radiographs with deep learning methods. Skeletal Radiol 49, 1369–1374. [CrossRef] [PubMed] [Google Scholar]
  11. Swiecicki A, Li N, O’Donnell J, Said N, Yang J, Mather RC, Jiranek WA, Mazurowski MA (2021) Deep learning-based algorithm for assessment of knee osteoarthritis severity in radiographs matches performance of radiologists. Comput Biol Med 133, 104334. [CrossRef] [PubMed] [Google Scholar]
  12. Tiulpin A, Thevenot J, Rahtu E, Lehenkari P, Saarakkala S (2018) Automatic knee osteoarthritis diagnosis from plain radiographs: a deep learning-based approach. Scientific Rep 8, 1727. [CrossRef] [Google Scholar]
  13. Lee LS, Chan PK, Wen C, Fung WC, Cheung A, Chan VWK, Cheung MH, Fu H, Yan CH, Chiu KY (2022) Artificial intelligence in diagnosis of knee osteoarthritis and prediction of arthroplasty outcomes: a review. Arthroplasty 4, 16. [CrossRef] [PubMed] [Google Scholar]
  14. Sharma S (2023) Artificial intelligence for fracture diagnosis in orthopedic X-rays: current developments and future potential. SICOT J 9, 21. [CrossRef] [EDP Sciences] [PubMed] [Google Scholar]
  15. Guan B, Liu F, Haj-Mirzaian A, Demehri S, Samsonov A, Neogi T, Guermazi A, Kijowski R (2020) Deep learning risk assessment models for predicting progression of radiographic medial joint space loss over a 48-month follow-up period. Osteoarthritis Cartilage 28, 428–437. [CrossRef] [PubMed] [Google Scholar]
  16. Arbeeva L, Minnig MC, Yates KA, Nelson AE (2023) Machine learning approaches to the prediction of osteoarthritis phenotypes and outcomes. Curr Rheumatol Rep 25, 213–225. [CrossRef] [PubMed] [Google Scholar]
  17. Schiratti JB, Dubois R, Herent P, Cahane D, Dachary J, Clozel T, Wainrib G, Keime-Guibert F, Lalande A, Pueyo M, Guillier R, Gabarroca C, Moingeon P (2021) A deep learning method for predicting knee osteoarthritis radiographic progression from MRI. Arthritis Res Ther 23, 262. [CrossRef] [PubMed] [Google Scholar]
  18. Leung K, Zhang B, Tan J, Shen Y, Geras KJ, Babb JS, Cho K, Chang G, Deniz CM (2020) Prediction of total knee replacement and diagnosis of osteoarthritis by using deep learning on knee radiographs: data from the osteoarthritis initiative. Radiology 296, 584–593. [CrossRef] [PubMed] [Google Scholar]
  19. Heisinger S, Hitzl W, Hobusch GM, Windhager R, Cotofana S (2020) Predicting total knee replacement from symptomology and radiographic structural change using artificial neural networks-data from the Osteoarthritis Initiative (OAI). J Clin Med 9, 1298. [CrossRef] [PubMed] [Google Scholar]
  20. Özden F, Nadiye Karaman Ö, Tuğay N, Yalın Kilinç C, Mihriban Kilinç R, Umut Tuğay B (2020) The relationship of radiographic findings with pain, function, and quality of life in patients with knee osteoarthritis. J Clin Orthop Trauma 11, S512–S517. [CrossRef] [PubMed] [Google Scholar]
  21. Bedson J, Croft PR (2008) The discordance between clinical and radiographic knee osteoarthritis: a systematic search and summary of the literature. BMC Musculoskelet Disord 9, 116. [CrossRef] [PubMed] [Google Scholar]
  22. Lee HS, Hwang S, Kim SH, Joon NB, Kim H, Hong YS, Kim S (2024) Automated analysis of knee joint alignment using detailed angular values in long leg radiographs based on deep learning. Sci Rep 14, 7226. [CrossRef] [PubMed] [Google Scholar]
  23. Vidhani FR, Woo JJ, Zhang YB, Olsen RJ, Ramkumar PN (2024) Automating linear and angular measurements for the hip and knee after computed tomography: validation of a three-stage deep learning and computer vision-based pipeline for pathoanatomic assessment. Arthroplasty Today 27, 101394. [CrossRef] [PubMed] [Google Scholar]
  24. Schwartz AM, Farley KX, Guild GN, Bradbury TL (2020) Projections and epidemiology of revision hip and knee arthroplasty in the United States to 2030. J Arthroplasty 35, S79–S85. [CrossRef] [PubMed] [Google Scholar]
  25. Rouzrokh P, Mickley JP, Khosravi B, Faghani S, Moassefi M, Schulz WR, Erickson BJ, Taunton MJ, Wyles CC (2023) THA-AID: Deep learning tool for total hip arthroplasty automatic implant detection with uncertainty and outlier quantification. J Arthroplasty 39, 966–973. [Google Scholar]
  26. Thirukumaran CP, Zaman A, Rubery PT, Calabria C, Li Y, Ricciardi BF, Bakhsh WR, Kautz H (2019) Natural language processing for the identification of surgical site infections in orthopaedics. J Bone Joint Surg Am 101, 2167–2174. [CrossRef] [PubMed] [Google Scholar]
  27. Farrow L, Zhong M, Anderson L (2024) Use of natural language processing techniques to predict patient selection for total hip arthroplasty: results from the AI to revolutionise the patient care pathway in hip and knee arthroplasty (ARCHERY) project. Orthop Proc 106-B, 34–34. [CrossRef] [Google Scholar]
  28. Farrow L, Zhong M, Anderson L (2024) Use of natural language processing techniques to predict patient selection for total hip and knee arthroplasty from radiology reports. Bone Joint J 106-B, 688–695. [CrossRef] [PubMed] [Google Scholar]
  29. Halawi MJ, Jongbloed W, Baron S, Savoy L, Williams VJ, Cote MP (2019) Patient dissatisfaction after primary total joint arthroplasty: the patient perspective. J Arthroplasty 34, 1093–1096. [CrossRef] [PubMed] [Google Scholar]
  30. Noble PC, Conditt MA, Cook KF, Mathis KB (2006) The John Insall Award: Patient expectations affect satisfaction with total knee arthroplasty. Clin Orthop Relat Res 452, 35–43. [CrossRef] [PubMed] [Google Scholar]
  31. Hunter J, Soleymani F, Viktor H, Michalowski W, Poitras S, Beaulé PE (2024) Using unsupervised machine learning to predict quality of life after total knee arthroplasty. J Arthroplasty 39, 677–682. [CrossRef] [PubMed] [Google Scholar]
  32. Kunze KN, Karhade AV, Sadauskas AJ, Schwab JH, Levine BR (2020) Development of machine learning algorithms to predict clinically meaningful improvement for the patient-reported health state after total hip arthroplasty. J Arthroplasty 35, 2119–2123. [CrossRef] [PubMed] [Google Scholar]
  33. Liu X, Liu Y, Lee ML, Hsu W, Liow MHL (2024) Identifying who are unlikely to benefit from total knee arthroplasty using machine learning models. npj Digit Med 7, 266. [CrossRef] [PubMed] [Google Scholar]
  34. Nam HS, Ho JPY, Park SY, Cho JH, Lee YS (2023) The development of machine learning algorithms that can predict patients satisfaction using baseline characteristics, and preoperative and operative factors of total knee arthroplasty. Knee 44, 253–261. [CrossRef] [PubMed] [Google Scholar]
  35. Tolk JJ, Janssen RPA, Haanstra TM, Van der Steen MC, Bierma-Zeinstra SMA, Reijman M (2021) The influence of expectation modification in knee arthroplasty on satisfaction of patients: a randomized controlled trial. Bone Joint J 103-b, 619–626. [CrossRef] [PubMed] [Google Scholar]
  36. Batailler C, Shatrov J, Sappey-Marinier E, Servien E, Parratte S, Lustig S (2022) Artificial intelligence in knee arthroplasty: current concept of the available clinical applications. Arthroplasty 4, 17. [CrossRef] [PubMed] [Google Scholar]
  37. Harris AHS, Kuo AC, Weng Y, Trickey AW, Bowe T, Giori NJ (2019) 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 477, 452–460. [CrossRef] [PubMed] [Google Scholar]
  38. Jo C, Ko S, Shin WC, Han HS, Lee MC, Ko T, Ro DH (2020) Transfusion after total knee arthroplasty can be predicted using the machine learning algorithm. Knee Surg Sports Traumatol Arthrosc 28, 1757–1764. [CrossRef] [PubMed] [Google Scholar]
  39. Ramkumar PN, Navarro SM, Haeberle HS, Karnuta JM, Mont MA, Iannotti JP, Patterson BM, Krebs VE (2019) Development and validation of a machine learning algorithm after primary total hip arthroplasty: applications to length of stay and payment models. J Arthroplasty 34, 632–637. [CrossRef] [PubMed] [Google Scholar]
  40. Arvind V, London DA, Cirino C, Keswani A, Cagle PJ (2021) Comparison of machine learning techniques to predict unplanned readmission following total shoulder arthroplasty. J Shoulder Elbow Surg 30, e50–e59. [CrossRef] [PubMed] [Google Scholar]
  41. Chen TL, Buddhiraju A, Costales TG, Subih MA, Seo HH, Kwon YM (2023) Machine learning models based on a national-scale cohort identify patients at high risk for prolonged lengths of stay following primary total hip arthroplasty. J Arthroplasty 38, 1967–1972. [CrossRef] [PubMed] [Google Scholar]
  42. Mohammadi R, Jain S, Namin AT, Scholem Heller M, Palacholla R, Kamarthi S, Wallace B (2020) Predicting unplanned readmissions following a hip or knee arthroplasty: retrospective observational study. JMIR Med Inform 8, e19761. [CrossRef] [PubMed] [Google Scholar]
  43. Park J, Zhong X, Miley EN, Rutledge RS, Kakalecik J, Johnson MC, Gray CF (2024) Machine learning-based predictive models for 90-day readmission of total joint arthroplasty using comprehensive electronic health records and patient-reported outcome measures. Arthroplasty Today 25, 101308. [CrossRef] [PubMed] [Google Scholar]
  44. Klemt C, Tirumala V, Habibi Y, Buddhiraju A, Chen TL, Kwon YM (2023) The utilization of artificial neural networks for the prediction of 90-day unplanned readmissions following total knee arthroplasty. Arch Orthop Trauma Surg 143, 3279–3289. [Google Scholar]
  45. Zmistowski B, Karam JA, Durinka JB, Casper DS, Parvizi J (2013) Periprosthetic joint infection increases the risk of one-year mortality. J Bone Joint Surg Am 95, 2177–2184. [CrossRef] [PubMed] [Google Scholar]
  46. Eka A, Chen AF (2015) Patient-related medical risk factors for periprosthetic joint infection of the hip and knee. Ann Transl Med 3, 233. [PubMed] [Google Scholar]
  47. Chong YY, Chan PK, Chan VWK, Cheung A, Luk MH, Cheung MH, Fu H, Chiu KY (2023) Application of machine learning in the prevention of periprosthetic joint infection following total knee arthroplasty: a systematic review. Arthroplasty 5, 38. [CrossRef] [PubMed] [Google Scholar]
  48. Yeo I, Klemt C, Robinson MG, Esposito JG, Uzosike AC, Kwon YM (2023) The use of artificial neural networks for the prediction of surgical site infection following TKA. J Knee Surg 36, 637–643. [CrossRef] [PubMed] [Google Scholar]
  49. Kuo FC, Hu WH, Hu YJ (2022) Periprosthetic joint infection prediction via machine learning: comprehensible personalized decision support for diagnosis. J Arthroplasty 37, 132–141. [CrossRef] [PubMed] [Google Scholar]
  50. Lüftinger L, Ferreira I, Frank BJH, Beisken S, Weinberger J, von Haeseler A, Rattei T, Hofstaetter JG, Posch AE, Materna A (2021) Predictive antibiotic susceptibility testing by next-generation sequencing for periprosthetic joint infections: potential and limitations. Biomedicines 9, 910. [CrossRef] [PubMed] [Google Scholar]
  51. Parvizi J, Tan TL, Goswami K, Higuera C, Della Valle C, Chen AF, Shohat N (2018) The 2018 definition of periprosthetic hip and knee infection: an evidence-based and validated criteria. J Arthroplasty 33, 1309–1314.e1302. [CrossRef] [PubMed] [Google Scholar]
  52. Burn E, Edwards CJ, Murray DW, Silman A, Cooper C, Arden NK, Pinedo-Villanueva R, Prieto-Alhambra D (2018) Trends and determinants of length of stay and hospital reimbursement following knee and hip replacement: evidence from linked primary care and NHS hospital records from 1997 to 2014. BMJ Open 8, e019146. [CrossRef] [PubMed] [Google Scholar]
  53. Cilla M, Borgiani E, Martinez J, Duda GN, Checa S (2017) Machine learning techniques for the optimization of joint replacements: application to a short-stem hip implant. PLoS One 12, e0183755. [CrossRef] [PubMed] [Google Scholar]
  54. Jang SJ, Kunze KN, Brilliant ZR, Henson M, Mayman DJ, Jerabek SA, Vigdorchik JM, Sculco PK (2022) Comparison of tibial alignment parameters based on clinically relevant anatomical landmarks: a deep learning radiological analysis. Bone Jt Open 3, 767–776. [CrossRef] [PubMed] [Google Scholar]
  55. Ding X, Zhang B, Li W, Huo J, Liu S, Wu T, Han Y (2021) Value of preoperative three-dimensional planning software (AI-HIP) in primary total hip arthroplasty: a retrospective study. J Int Med Res 49, 1–21. [Google Scholar]
  56. Velasquez Garcia A, Bukowiec LG, Yang L, Nishikawa H, Fitzsimmons JS, Larson AN, Taunton MJ, Sanchez-Sotelo J, O’Driscoll SW, Wyles CC (2024) Artificial intelligence-based three-dimensional templating for total joint arthroplasty planning: a scoping review. Int Orthop 48, 997–1010. [CrossRef] [PubMed] [Google Scholar]
  57. Fernandes LR, Arce C, Martinho G, Campos JP, Meneghini RM (2023) Accuracy, reliability, and repeatability of a novel artificial intelligence algorithm converting two-dimensional radiographs to three-dimensional bone models for total knee arthroplasty. J Arthroplasty 38, 2032–2036. [CrossRef] [PubMed] [Google Scholar]
  58. Jung K, Kim H, Kholinne E, Park D, Choi H, Lee S, Shin M-J, Kim D-M, Hong J, Koh KH, Jeon I-H (2020) Navigation-assisted anchor insertion in shoulder arthroscopy: a validity study. BMC Musculoskelet Disord 21, 812. [CrossRef] [PubMed] [Google Scholar]
  59. Hu W, Tang J, Wu X, Zhang L, Ke B (2016) Minimally invasive versus open transforaminal lumbar fusion: a systematic review of complications. Int Orthop 40, 1883–1890. [CrossRef] [PubMed] [Google Scholar]
  60. Fan M, Liu Y, He D, Han X, Zhao J, Duan F, Liu B, Tian W (2020) Improved accuracy of cervical spinal surgery with robot-assisted screw insertion: a prospective, randomized, controlled study. Spine 45, 285–291. [CrossRef] [PubMed] [Google Scholar]
  61. Anthony CA, Rojas EO, Keffala V, Glass NA, Shah AS, Miller BJ, Hogue M, Willey MC, Karam M, Marsh JL (2020) Acceptance and commitment therapy delivered via a mobile phone messaging robot to decrease postoperative opioid use in patients with orthopedic trauma: randomized controlled trial. J Med Internet Res 22, e17750. [CrossRef] [PubMed] [Google Scholar]
  62. Wittig-Wells D, Higgins M, Carter J, Davis E, Holmes E, Jacob A, Samms-McPherson J, Simms S (2019) Impact of a preset daily cell phone alarm on medication adherence for aspirin as antithrombotic therapy. Orthop Nurs 38, 311–316. [CrossRef] [PubMed] [Google Scholar]
  63. Ramkumar PN, Haeberle HS, Navarro SM, Sultan AA, Mont MA, Ricchetti ET, Schickendantz MS, Iannotti JP (2018) Mobile technology and telemedicine for shoulder range of motion: validation of a motion-based machine-learning software development kit. J Shoulder Elbow Surg 27, 1198–1204. [CrossRef] [PubMed] [Google Scholar]
  64. Ramkumar PN, Haeberle HS, Ramanathan D, Cantrell WA, Navarro SM, Mont MA, Bloomfield M, Patterson BM (2019) Remote patient monitoring using mobile health for total knee arthroplasty: validation of a wearable and machine learning-based surveillance platform. J Arthroplasty 34, 2253–2259. [CrossRef] [PubMed] [Google Scholar]
  65. Borjali A, Chen A, Muratoglu O, Morid M, Varadarajan K (2019) Detecting mechanical loosening of total hip replacement implant from plain radiograph using deep convolutional neural network. arXiv:1912.00943. [Google Scholar]
  66. Rouzrokh P, Ramazanian T, Wyles CC, Philbrick KA, Cai JC, Taunton MJ, Maradit Kremers H, Lewallen DG, Erickson BJ (2021) Deep learning artificial intelligence model for assessment of hip dislocation risk following primary total hip arthroplasty from postoperative radiographs. J Arthroplasty 36, 2197–2203. [CrossRef] [PubMed] [Google Scholar]
  67. Sadoghi P, Liebensteiner M, Agreiter M, Leithner A, Böhler N, Labek G (2013) Revision surgery after total joint arthroplasty: a complication-based analysis using worldwide arthroplasty registers. J Arthroplasty 28, 1329–1332. [CrossRef] [PubMed] [Google Scholar]
  68. Schroer WC, Berend KR, Lombardi AV, Barnes CL, Bolognesi MP, Berend ME, Ritter MA, Nunley RM (2013) Why are total knees failing today? Etiology of total knee revision in 2010 and 2011 J Arthroplasty 28, 116–119. [CrossRef] [PubMed] [Google Scholar]
  69. Bozic KJ, Kurtz SM, Lau E, Ong K, Vail TP, Berry DJ (2009) The epidemiology of revision total hip arthroplasty in the United States. J Bone Joint Surg Am 91, 128–133. [CrossRef] [PubMed] [Google Scholar]
  70. Kunutsor SK, Barrett MC, Beswick AD, Judge A, Blom AW, Wylde V, Whitehouse MR (2019) Risk factors for dislocation after primary total hip replacement: meta-analysis of 125 studies involving approximately five million hip replacements. Lancet Rheumatol 1, e111–e121. [CrossRef] [Google Scholar]

Cite this article as: Georgiakakis ECT, Khan AM, Logishetty K & Sarraf KM (2024) Artificial intelligence in planned orthopaedic care. SICOT-J 10, 49. https://doi.org/10.1051/sicotj/2024044.

All Figures

thumbnail Figure 1

The applications of artificial intelligence in planned orthopaedic care discussed in this article. AI = artificial intelligence, OA = osteoarthritis, TKA = total knee arthroplasty, THA = total hip arthroplasty, NLP = natural language processing, EPR = electronic patient records, PROM = patient-reported outcome measure,, AR = augmented reality.
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.