2023 Review | Advances in Artificial Intelligence Technology Applications in Joint Surgery

🥼 Practice and Clinical
,
1

This article is transcoded by SimpRead, original address www.orthonline.com.cn

2023 Review | Advances in Artificial Intelligence Technology in Joint Orthopedics

Authors: Li Haifeng, Chai Wei

Source: Orthopedics Department of General Hospital of the Chinese People’s Liberation Army

As an emerging science and technology, Artificial Intelligence (AI) generally refers to a revolutionary approach that simulates human intelligence including perception, understanding, reasoning, learning, and decision-making through computer programs and algorithms, to solve various complex problems. In recent years, AI has been widely applied in medical fields such as medical imaging-assisted diagnosis, clinical decision support, intelligent health management, and novel drug development. In orthopedics, AI-based disease diagnosis, personalized treatment, robot-assisted surgery, prognosis prediction, and other aspects are particularly important for improving patient treatment outcomes [1].

AI has also deeply penetrated various stages of joint replacement surgery, ranging from preoperative planning to postoperative care and monitoring. AI not only assists joint orthopedic surgeons in clinical decision-making, resource allocation, and disease intervention but also applies to diagnosis and grading of diseases such as osteoarthritis and femoral head necrosis, predicting implant size and position [2]; it can evaluate the outcomes of joint replacement, risks of prosthetic dislocation and infection, length of hospital stay, and economic costs, and can also monitor the patient’s joint function recovery process.

This article aims to summarize the latest research on AI in the field of joint orthopedics in 2023, covering various aspects of AI including ChatGPT, deep learning, and machine learning, to help orthopedic surgeons, especially joint specialists, gain a broader and more comprehensive understanding of the latest developments in this field.

Figure 1 Application of Artificial Intelligence in Orthopedics

I. Application of ChatGPT in Joint Orthopedics

ChatGPT (OpenAI, San Francisco, USA) was released in November 2022, quickly producing a huge and profound impact across various industries. ChatGPT is increasingly used in healthcare. Jeremy et al. [3] compared the top 10 most frequently asked questions (FAQs) and corresponding top 10 common answers searched by ChatGPT users about “total knee arthroplasty” and “total hip arthroplasty” with those from the most widely used search engine Google Web Search in the United States today. The results showed that querying medical and health questions using ChatGPT has a similar effective value as using Google. The authors believe that ChatGPT can continue to serve as a useful potential resource for patients to enhance the accuracy of self-diagnosis.

Aleksander et al. attempted to determine whether ChatGPT can effectively answer FAQs about total hip arthroplasty (THA). The authors posed ten common questions about THA to ChatGPT and analyzed the accuracy of each response using evidence-based methods. Responses were rated as “excellent replies needing clarification,” “satisfactory replies needing minimal clarification,” “satisfactory replies needing moderate clarification,” or “unsatisfactory replies needing extensive clarification.” The results showed that among ChatGPT’s replies, only one was rated “unsatisfactory,” two required no clarification, and most required minor (4 of 10) or moderate (3 of 10) clarifications. Although some replies needed more detailed explanations, ChatGPT’s responses were generally fair and evidence-supported, even on controversial topics. The authors concluded that ChatGPT can effectively provide evidence-based answers to patients’ frequently asked questions before undergoing THA and present information in a way most patients can understand, serving as a valuable clinical tool for future preoperative patient education in orthopedics.

ChatGPT has powerful natural language and data processing capabilities to assist joint orthopedic surgeons in disease diagnosis and treatment selection. It can also aid physicians in differential diagnosis, treatment recommendations, decision support, development of predictive models, tele-rehabilitation support, patient education, improving communication with rehabilitation teams, guiding postoperative rehabilitation, and assisting joint orthopedic surgeons with literature reviews, idea generation, and data analysis.

II. Application of AI in Musculoskeletal Imaging

With the development and promotion of advanced AI technologies, musculoskeletal disease imaging diagnosis has entered a new era.

2.1 Application of Artificial Intelligence in Imaging Recognition of Knee Osteoarthritis

Numerous studies have applied AI in the diagnosis, classification, and grading of osteoarthritis (OA), especially in detecting cartilage lesions, OA diagnosis, and OA risk assessment. AI has demonstrated diagnostic performance comparable to human intelligence in imaging diagnosis, including X-rays and MRI. Various AI methods have been employed for fully automated segmentation of knee cartilage and bone tissues, showing higher segmentation accuracy and significantly reduced segmentation time compared to other current methods. Moreover, various AI models analyzing knee osteoarthritis imaging—including X-rays and MRI—have been developed to assess OA risk. These models exhibit high diagnostic performance in predicting OA prognosis, including incidence and progression of radiographic knee OA, onset and progression of joint pain, and whether total knee arthroplasty will be needed in the future.

Maria et al. [4] selected X-rays from 124 knee OA patients and analyzed radiographic features such as Kellgren-Lawrence (K-L) grading, joint space narrowing, subchondral sclerosis, and osteophytes using an AI-based image annotation tool by both senior and junior physicians, comparing intra- and inter-group consistency. The study found that utilizing AI-based imaging analysis software improved senior physicians’ radiographic assessments of knee OA (Figure 2). After AI assistance, consistency and accuracy of junior physicians’ assessments were comparable to seniors. The authors suggested that integrating AI software in radiological evaluation of knee OA can enhance diagnostic accuracy. Neubauer et al. [5] selected DICOM X-ray data from 71 knee OA patients and analyzed it using AI software, finding AI-assisted diagnostic grading correlated better with K-L scores and KOOS scores. The authors concluded that AI-assisted systems can improve knee X-ray grading and show stronger correlation with clinical severity.

Figure 2 Inter-rater agreement among physicians of different seniority levels

2.2 Application of Artificial Intelligence in Imaging Recognition of Avascular Necrosis of the Femoral Head

Deep learning, machine learning, and other AI techniques can distinguish early and late stages of avascular necrosis of the femoral head (AVN), predicting femoral head collapse, thereby better assisting less experienced surgeons, especially valuable in large-scale medical imaging screening and community healthcare scenarios lacking expert consultation.

Hernigou et al. [6] applied AI methods to analyze risk factors for collapse in non-traumatic femoral head osteonecrosis patients. Data from 900 patients with non-traumatic AVN were collected, including 50 variables related to osteonecrosis. Patients were randomly assigned to training (80%) and validation (20%) groups, and machine learning algorithms evaluated selected variables. The results showed the machine learning model predicted femoral head collapse within three years with an accuracy of 81.2%. Accuracy for predicting collapse at 6 months, 12 months, and 24 months was 54.2%, 67.3%, and 71.2%, respectively. This study was the first to demonstrate that machine learning algorithms can predict femoral head collapse (Figure 3).

Klontzas et al. [7] also developed an AI method to distinguish early and late hip AVN to guide treatment decisions. MRI data of 104 AVN patients were analyzed using three convolutional neural networks (CNNs): VGG-16, Inception ResnetV2, and InceptionV3, aiming to differentiate early (ARCO 1-2) from late (ARCO 3-4) stages. The results were compared with diagnoses by two radiologists. Inception-ResnetV2 performed best, with an area under the curve (AUC) of 99.7% (95% CI 99–100%), followed by InceptionV3 and VGG-16 with AUCs of 99.3% (95% CI 98.4–100%) and 97.3% (95% CI 95.5–99.2%) respectively. External validation with data from another country showed VGG-16 had the highest individual AUC of 78.9% (95% CI 51.6–79.6%), and the CNN ensemble model achieved the best external performance with an AUC of 85.5% (95% CI 72.2–93.9%). No significant difference was found between the CNN ensemble model and expert radiologists (p = 0.22 and 0.092). The authors concluded that an externally validated CNN ensemble model can accurately distinguish early and late-stage AVN and performs comparably to expert radiologists.

Figure 3 Collapse risk surface plot based on osteonecrosis volume percentage and hip pain importance

2.3 Application of Artificial Intelligence in Prosthesis Identification and Measurement

Karnuta et al. [8] trained, validated, and externally tested an AI method using 3.5 million X-ray films to automatically identify total knee arthroplasty (TKA) prostheses. After 1000 training epochs in a deep learning system, the model automatically distinguished all nine types of prostheses on a test dataset of 744 anteroposterior X-rays, achieving an AUC of 0.989, accuracy of 97.4%, sensitivity of 89.2%, and specificity of 99.0%. The average identification time per image was 0.02 seconds. Similarly, other scholars [9] used 2 million X-rays to train, validate, and test AI models to automatically identify total hip arthroplasty (THA) prostheses. After 1000 training epochs, the model identified eight types of prostheses on a test set of 588 anteroposterior X-rays with an AUC of 0.991, accuracy of 97.9%, sensitivity of 88.6%, and specificity of 98.9%. Average recognition time per image was 0.02 seconds. The authors believe that AI software for identifying knee and hip replacement prostheses performs outstandingly, demonstrating clinical applicability and significant potential for assisting preoperative planning of knee and hip revision surgeries (Figures 4, 5).

Murphy et al. [10] designed an AI program to automatically correct pelvic tilt and determine acetabular cup anteversion. The AI model was trained, validated, and tested on X-rays from 2,945 patients. For some patients, CT scans and 3D reconstructions were performed to measure cup orientation. The AI measurement time on X-rays averaged 0.22±0.03 seconds. Compared with manual measurements, AI measurements were closer to CT results (P < .001). AI predicted acetabular cup anteversion confirmed by 17 X-rays with 100.0% accuracy (n=45). The authors concluded that AI algorithms measuring acetabular cup orientation on X-rays can correct pelvic tilt and outperform manual measurements.

Figure 4 Sample heatmap of knee prostheses demonstrating AI model’s focus areas

Figure 5 Heatmap images showing AI algorithms’ prosthesis features of interest, purposefully enhancing the anteroposterior X-ray images into imperfect renderings

Prosthesis types: DePuy SROM (top left), DePuy Corail (top right), Stryker Accolade (bottom left), and Stryker Restoration Modular (bottom right)

III. Application of Artificial Intelligence in Preoperative Evaluation, Planning, and Postoperative Blood Transfusion

Houserman et al. [11] selected 8,301 X-ray images of three views (anteroposterior, lateral, and axial) from 2,767 patients along with data on whether patients underwent joint replacement surgery (including UKA or TKA) to train (70%) and validate (30%) AI models using transfer learning for computer vision models. The AI model achieved an AUC of 0.97 for TKA, 0.96 for UKA, and 0.98 for non-surgical cases. Accuracy for predicting surgical vs. non-surgical intervention was 93.8%, and for predicting TKA vs. non-TKA was 88%. The authors believe AI models can predict which patients are suitable for UKA, TKA, or non-surgical intervention.

Salman et al. [12] reviewed AI accuracy in predicting implant size during total knee arthroplasty (TKA), including four papers with a total of 34,547 patients. AI predicted femoral implant size with accuracy ranging from 88.3% to 99.7% ± 1 size, and tibial implant size with accuracy ranging from 90% to 99.9% ± 1 size. This study confirms AI’s great potential in planning TKA with satisfactory performance in implant size prediction.

Burge et al. [13] used CT images from 98 subjects and a series of machine learning methods including classification, object detection, and image segmentation models to create 3D predictive models of the femur and tibia. Customized implant models were generated by computer-aided design and virtually fitted to the “real” 3D bone models of each test subject. AI assistance yielded highly accurate designs unaffected by subjects’ gender, height, age, or side. The authors concluded that a powerful, accurate, automated CT-based workflow for customized total knee arthroplasty is feasible and provides significant time and cost advantages compared to traditional methods (Figures 6, 7, 8).

Figure 6 Workflow for CT-3D surface model prediction process

Figure 7 AI automatic customization process for femoral prosthesis (top row) and tibial baseplate (bottom row)

Figure 8 Maximum OUH calculation for femoral component (top row) and tibial plate (bottom row)

Garcia et al. [14] evaluated the current research status of AI-based three-dimensional (3D) templating in preoperative planning of total joint arthroplasty. Nine studies involving primary or revision arthroplasty using AI-based 3D templates for surgical planning were included. Compared with traditional X-ray templating, AI-based 3D templating systems reduced planning time and improved accuracy of implant size/position and imaging feature estimation. The authors believe that AI-based 3D templating systems can provide more accurate and personalized preoperative planning, potentially improving patients’ functional outcomes.

Cohen-Levy et al. [15] used AI models to predict postoperative blood transfusion rates after primary total hip arthroplasty (THA). Four machine learning algorithms were developed to assess 7,265 consecutively treated patients undergoing primary THA. Model discrimination (AUC > 0.78), calibration, and decision curve analysis showed excellent performance, useful for predicting postoperative transfusion rates in primary THA patients. Data from large national databases on 101,266 primary THA and 8,594 revision THA patients were used by Zhang et al. (PMID: 37315632) to train and validate five AI algorithms predicting transfusion risk after primary and revision THA. All ML models showed excellent discrimination (AUC > 0.8) in both primary and revision patients, successfully validating the previously developed AI algorithms for this purpose.

IV. Application of Artificial Intelligence in Predicting Outcomes After Artificial Joint Replacement Surgery

4.1 Predicting Postoperative Risk and Readmission

Kunze et al. [16] developed AI algorithms based on clinical registry data from 616 primary THA patients to predict all-cause complications at least 2 years postoperatively, performing internal validation. The observed complication rate was 16.6%. The random gradient boosting model achieved the best predictive performance with an AUC of 0.88, calibration intercept of 0.1, calibration slope of 1.22, and Brier score of 0.09. The authors concluded that gradient boosting showed good discrimination in identifying patients at high risk of postoperative complications. Zhang et al. (PMID: 35933638) developed and validated machine learning models to predict unplanned 90-day readmissions after total knee arthroplasty (TKA). Out of 10,021 TKA patients, 644 (6.4%) were readmitted within 90 days. Four ML algorithms (artificial neural network, support vector machine, k-nearest neighbors classification, and elastic net logistic regression) were developed for prediction. All models demonstrated excellent performance in discrimination (AUC > 0.82), calibration, and decision curve analysis. The authors suggested all four models have outstanding prediction abilities for unplanned readmissions.

4.2 Predicting Postoperative Thrombotic Risk

Wang et al. [17] analyzed data from 6,897 patients who underwent primary knee/hip arthroplasty followed by bilateral lower limb venous ultrasound between January 2017 and December 2021. Patient features were extracted from electronic health records (EHR) and six AI models—including Extreme Gradient Boosting, Random Forest, Support Vector Machine, Logistic Regression, Ensemble, and Backpropagation Neural Network—were trained on 80% and tested on 20% of the dataset. Among them, 1,161 cases (16.8%) were positive for deep vein thrombosis (DVT), and 5,736 (83.2%) were negative. The Ensemble model achieved the highest AUC of 0.9206 (0.8956–0.9364), sensitivity of 0.8027, specificity of 0.9059, positive predictive value of 0.6100, negative predictive value of 0.9573, and F1 score of 0.7003. The authors concluded that EHR-based machine learning models can help predict risk of deep venous thrombosis after knee/hip arthroplasty.

4.3 Predicting Postoperative Prosthetic Dislocation and Loosening

Hernigou et al. [18] reviewed 75 articles on primary hip arthroplasty, collecting data on 1,069,565 prostheses and 26,488 dislocation cases. Using deep learning models, they evaluated different surgical approaches—including hemiarthroplasty, standard total hip arthroplasty, dual-mobility cups, constrained liners—and predicted postoperative dislocation risk. Resulting dislocation risks varied with prosthesis type: dislocation rates of 0% to 3.9% (mean 0.31%) for 3,045 dual-mobility hemiarthroplasties; 0.2% to 1.2% (overall 0.91%) for 457 constrained liners; 1.76% to 4.2% (mean 2.1%) for 895,734 conventional total hip arthroplasties; and 0.76% to 12.2% (mean 4.5%) for 170,329 hemiarthroplasties. The AI model predicted postoperative dislocation with 95% accuracy.

Kim et al. [19] included five retrospective studies with 2,013 patient data sets, totaling 3,236 images including 2,442 (75.5%) THA and 794 (24.5%) TKA. The most frequently used and best performing ML algorithm was DenseNet. The pooled sensitivity was 0.92, pooled specificity was 0.95, pooled diagnostic odds ratio was 194.09, and AUC was 0.9853. The authors believe AI shows good accuracy, sensitivity, and specificity in predicting prosthetic loosening in THA and TKA via X-rays and recommend integrating AI approaches into prosthetic loosening screening programs.

4.4 Predicting Postoperative Infection

AI methods aid in preoperative planning for periprosthetic joint infection (PJI), early diagnosis of infection, early administration of appropriate antibiotics, and predicting clinical outcomes after joint replacement. Wu et al. [20] demonstrated that AI can be used to automatically detect complex surgical site infections. Albano et al. [21] investigated whether AI can differentiate the likelihood of purulent infection in planned revision total hip arthroplasty (THA) based on preoperative MRI features. They included 173 THA revision patients (98 females, age 67 ± 12 years) with preoperative pelvic MRI data. Patients were divided into training, validation, internal test group (n=117), and an independent external test group (n=56). MRI features were used to train, validate, and test an SVM-based AI algorithm. Results showed that in the primary group, AI had sensitivity of 92%, specificity of 62%, and AUC of 81% in predicting THA infection. In the external test group, AI sensitivity to detect THA infection remained 92%, specificity increased to 79%, and AUC to 89%. The authors concluded that using SVM-based AI to interpret MRI features is a promising method for predicting postoperative THA infection and may help radiologists identify infection after THA (Figure 9).Klemt et al. [22] reviewed a database of 1,432 patients who underwent aseptic revision total knee arthroplasty (TKA), among whom 208 underwent revision surgery again due to periprosthetic joint infection (PJI) (14.5%). Three machine learning algorithms (artificial neural networks, support vector machines, k-nearest neighbors) were used to predict this outcome, and these models were evaluated using discrimination, calibration, and decision curve analysis, with the neural network model showing the best performance in discrimination (AUC=0.78), calibration, and decision curve analysis. The study found that using machine learning to predict PJI following aseptic revision TKA has excellent performance. Validated machine learning models can help surgeons stratify patient-specific risks and assist in preoperative counseling and clinical decision-making for patients undergoing aseptic revision TKA. It is evident that artificial intelligence methods have broad application prospects in monitoring PJI.

Figure 9 Pelvic MRI of a 77-year-old female with infected total hip arthroplasty (THA)

Coronal STIR (A) shows periprosthetic acetabular bone edema (white arrow) and extra-capsular edema (hollow arrow). No bone destruction is detected on coronal T1-weighted image (B). Axial T2-weighted images (C and D) demonstrate inflammation and synovitis.

5. Conclusion

The application of artificial intelligence in the field of hip and knee arthroplasty has developed rapidly. In preoperative assessment, artificial intelligence has demonstrated several notable advantages, such as accurately predicting length of hospital stay, hospitalization costs, and discharge disposition. Artificial intelligence can also be applied to image recognition and classification of knee osteoarthritis, with accuracy comparable to that of trained joint replacement surgeons. Additionally, artificial intelligence can accurately predict prosthesis size, lower limb alignment, and prosthesis angle, with accuracy reaching up to 95%, significantly higher than traditional methods. Moreover, AI applications in preoperative planning for total joint arthroplasty (TJA) are beginning to show strong practical utility. Compared to radiology templates, AI can more accurately predict prosthesis sizes. In postoperative care after TJA, AI’s utility is equally important. Furthermore, image-based machine learning models have shown exceptional accuracy in predicting postoperative complications, reflecting clinical imaging characteristics. In summary, artificial intelligence is paving the way for innovation in contemporary joint surgery and will fundamentally transform the diagnostic and treatment paradigms of traditional joint surgery.

Author Profiles

Wei Chai, Director of Joint Surgery, Department of Orthopedics, Chinese People’s Liberation Army General Hospital; Chief Physician, Professor.

Academic positions: Member of the International Hip Society, Committee Member and Youth Group Leader of the Joint Surgery Working Committee under the Orthopedic Physicians Branch of the Chinese Medical Doctor Association, Member of the Joint Surgery Group of the Youth Committee of the Chinese Orthopaedic Association, Vice President of the Bone and Joint Branch of the Chinese Geriatrics Society, Deputy Director of the Youth Committee and Group Leader of the Joint Surgery Group under the Orthopedic Branch of the Beijing Medical Association.

Haifeng Li, Associate Chief Physician, Department of Orthopedics, Chinese People’s Liberation Army General Hospital.

Dedicated to basic research and clinical treatment of adult shoulder, elbow, knee, and ankle joint reconstruction surgery. Skilled in peri-knee osteotomy, minimally invasive unicompartmental replacement, total knee arthroplasty, navigation or robot-assisted intelligent techniques to preserve the knee and series of stepwise comprehensive treatments; deformity correction osteotomy for congenital knee deformities and traumatic fracture sequelae; joint reconstruction after knee bone tumor resection; robot-assisted partial and total knee arthroplasty.

Member of the Youth Committee of the Joint Surgery Group and the Bone Infection Group of the Orthopedic Branch of the Beijing Medical Association. Recipient of one second-class and two third-class Military Scientific and Technological Progress Awards, holder of 16 national patents, authored over 60 papers.

References

  1. Garvey KV, Thomas Craig KJ, Russell R, Novak LL, Moore D, Miller BM. Considering clinician competencies for the implementation of artificial intelligence-based tools in health care: findings from a scoping review. JMIR Med Inform. 2022 Nov 16;10(11):e37478.

  2. Groot OQ, Ogink PT, Lans A, Twining PK, Kapoor ND, DiGiovanni W, et al. Machine learning prediction models in orthopedic surgery: a systematic review in transparent reporting. J Orthop Research® 2022;40:475-83.

  3. Jeremy A Dubin, Sandeep S Bains, Zhongming Chen, et al. Using a Google Web Search Analysis to Assess the Utility of ChatGPT in Total Joint Arthroplasty. J Arthroplasty. 2023 Jul;38(7):1195-1202.

  4. Maria Anna Smolle, Christoph Goetz, Dietmar Maurer, et al. Artificial intelligence-based computer-aided system for knee osteoarthritis assessment increases experienced orthopaedic surgeons’ agreement rate and accuracy. Knee Surg Sports Traumatol Arthrosc. 2023 Mar;31(3):1053-62.

  5. Markus Neubauer, Lukas Moser, Johannes Neugebauer, et al. Artificial-Intelligence-Aided Radiographic Diagnostic of Knee Osteoarthritis Leads to a Higher Association of Clinical Findings with Diagnostic Ratings. J Clin Med. 2023 Jan 17;12(3):744.

  6. Philippe Hernigou. Revisiting prediction of collapse in hip osteonecrosis with artificial intelligence and machine learning: a new approach for quantifying and ranking the contribution and association of factors for collapse. Int Orthop. 2023 Mar;47(3):677-89.

  7. Michail E Klontzas, Evangelia E Vassalou, Konstantinos Spanakis, et al. Deep learning enables the differentiation between early and late stages of hip avascular necrosis. Eur Radiol. 2023 Aug 15:1-8.

  8. Karnuta JM, Shaikh HJF, Murphy MP, et al. Artificial Intelligence for Automated Implant Identification in Knee Arthroplasty: A Multicenter External Validation Study Exceeding 3.5 Million Plain Radiographs. J Arthroplasty. 2023 Oct;38(10):2004-2008.

  9. Karnuta JM, Murphy MP, Luu BC, et al. Artificial Intelligence for Automated Implant Identification in Total Hip Arthroplasty: A Multicenter External Validation Study Exceeding Two Million Plain Radiographs. J Arthroplasty. 2023 Oct;38(10):1998-2003.

  10. Michael P Murphy, Cameron J Killen, Sara R Winfrey, et al. Artificial Intelligence Autonomously Measures Cup Orientation, Corrects for Pelvis Orientation, and Identifies Retroversion From Antero-Posterior Pelvis Radiographs. J Arthroplasty. 2023 Jul;38(7 Suppl 2):S319-S323.

  11. David J Houserman, Keith R Berend, Adolph V Lombardi Jr, et al. The Viability of an Artificial Intelligence/Machine Learning Prediction Model to Determine Candidates for Knee Arthroplasty. J Arthroplasty. 2023 Oct;38(10):2075-2080.

  12. Loay A Salman, Harman Khatkar, Abdallah Al-Ani, et al. Reliability of artificial intelligence in predicting total knee arthroplasty component sizes: a systematic review. Eur J Orthop Surg Traumatol. 2023 Nov 27:47-58.

  13. Thomas A Burge, Jonathan R T Jeffers, Connor W Myant. Applying machine learning methods to enable automatic customization of knee replacement implants from CT data. Sci Rep. 2023 Feb 27;13(1):3317.

  14. Ausberto Velasquez Garcia, Lainey G Bukowiec, Linjun Yang, et al. Artificial intelligence-based three-dimensional templating for total joint arthroplasty planning: a scoping review. Int Orthop. 2024 Jan 15:23-35.

  15. Wayne Brian Cohen-Levy, Christian Klemt, Venkatsaiakhil Tirumala, et al. Artificial neural networks for the prediction of transfusion rates in primary total hip arthroplasty. Arch Orthop Trauma Surg. 2023 Mar;143(3):1643-1650.

  16. Kyle N Kunze, Aditya V Karhade, Evan M Polce, et al. Development and internal validation of machine learning algorithms for predicting complications after primary total hip arthroplasty. Arch Orthop Trauma, 2023 Apr;143(4):2181-2188.

  17. Xinguang Wang, Hanxu Xi, Xiao Geng, et al. Artificial Intelligence-Based Prediction of Lower Extremity Deep Vein Thrombosis Risk After Knee/Hip Arthroplasty. Sur Clin Appl Thromb Hemost. 2023 Jan-Dec:29:1-9.

  18. Philippe Hernigou, Olivier Barbier, Philippe Chenaie. Hip arthroplasty dislocation risk calculator: evaluation of one million primary implants and twenty-five thousand dislocations with deep learning artificial intelligence in a systematic review of reviews. Int Orthop. 2023 Feb;47(2):557-571.

  19. Man-Soo Kim, Jae-Jung Kim, Ki-Ho Kang, et al. Detection of Prosthetic Loosening in Hip and Knee Arthroplasty Using Machine Learning: A Systematic Review and Meta-Analysis. Medicina (Kaunas) 2023 Apr 17;59(4):782-800.

  20. Guosong Wu, Cheligeer Cheligeer, Danielle A Southern, et al. Development of machine learning models for the detection of surgical site infections following total hip and knee arthroplasty: a multicenter cohort study. Antimicrob Resist Infect Control. 2023 Sep 2;12(1):88-97.

  21. Domenico Albano, Salvatore Gitto, Carmelo Messina, et al. MRI-based artificial intelligence to predict infection following total hip arthroplasty failure. Radiol Med. 2023 Mar;128(3):340-346.

  22. Christian Klemt, Ingwon Yeo, Michael Harvey, et al. The Use of Artificial Intelligence for the Prediction of Periprosthetic Joint Infection Following Aseptic Revision Total Knee Arthroplasty. J Knee Surg. 2024 Jan;37(2):158-166.