Artificial intelligence in orthopaedics: Clinical decision support, medical imaging, surgical planning, and outcome prediction

Abstract

Artificial intelligence (AI) is increasingly transforming orthopaedic practice by enhancing diagnostic accuracy, clinical decision-making, surgical planning, and postoperative monitoring. Advances in computational modelling, imaging analysis, and predictive analytics now enable clinicians to process complex multimodal datasets derived from radiological imaging, clinical records, biomechanical parameters, and patient-reported outcomes. These technologies are being applied across multiple orthopaedic subspecialties, including trauma, spine surgery, arthroplasty, sports medicine, oncology, infection, and rehabilitation. AI-driven systems have demonstrated strong performance in fracture detection, implant assessment, tumour characterisation, infection diagnosis, and prediction of surgical outcomes, often matching or exceeding conventional analytical approaches. In operative settings, integration with robotics, navigation platforms, and augmented visualisation systems improves procedural precision, reproducibility, and intraoperative decision support. Predictive models also assist clinicians in risk stratification, patient selection, complication forecasting, and personalised rehabilitation planning. Despite these promising developments, widespread clinical implementation remains limited by challenges related to dataset variability, algorithm transparency, regulatory oversight, cost, and concerns regarding bias and data privacy. Ongoing multicentre validation studies, development of explainable computational frameworks, and global collaboration will be essential for safe and equitable adoption. AI is unlikely to replace orthopaedic clinicians but is expected to function as a powerful adjunct that enhances clinical judgement and supports precision-based musculoskeletal care. Continued technological refinement and responsible integration into clinical workflows will determine its long-term impact on patient outcomes and healthcare delivery.

Key Words: Artificial intelligence; Orthopaedics; Machine learning; Deep learning; Clinical decision support systems; Medical imaging; Surgical planning; Predictive analytics; Precision medicine; Musculoskeletal disorders

Core Tip: Artificial intelligence is rapidly advancing orthopaedic care by improving diagnostic accuracy, surgical planning, risk prediction, and personalised rehabilitation through integration of imaging, clinical, and biomechanical data. Emerging technologies such as decision support systems, robotics, and predictive modelling are enhancing precision and efficiency across subspecialties. Despite promising clinical performance, challenges including dataset bias, limited external validation, regulatory uncertainty, and data privacy concerns must be addressed. Artificial intelligence is best viewed as a clinical adjunct that augments surgeon judgement and supports the transition toward data-driven, precision musculoskeletal medicine.

INTRODUCTION

Artificial intelligence (AI) is increasingly influencing the field of orthopaedics, largely in response to the growing availability of digital health data derived from medical imaging, electronic health records, wearable technologies, and surgical analytics. Advances in machine learning (ML) and deep learning (DL) have made it possible to analyse complex and diverse datasets with a level of speed and pattern recognition that was previously difficult to achieve through conventional analytical approaches. These developments are enabling clinicians to identify clinically relevant trends and generate predictive insights that support diagnostic accuracy and treatment planning across several orthopaedic subspecialties[1,2].

In recent years, AI-based applications have shown encouraging results in musculoskeletal imaging, particularly in areas such as fracture identification, implant evaluation, tumour characterisation, and prediction of treatment outcomes. These tools have the potential to enhance diagnostic consistency while simultaneously reducing the burden of repetitive and time-consuming image interpretation tasks for clinicians[3,4]. As orthopaedic surgical procedures continue to evolve in complexity, there has been a parallel increase in the demand for technologies that support precision-driven patient care. In this context, AI-assisted clinical decision support systems (CDSS) are increasingly being explored for their ability to assist in patient selection, preoperative planning, risk stratification, and formulation of personalised rehabilitation protocols[5,6].

Beyond its role in individual patient management, AI is gradually being incorporated into integrated orthopaedic care models. By combining clinical findings with imaging data, biomechanical parameters, and perioperative variables, AI has the potential to support a more individualised and data-driven approach to orthopaedic treatment, aligning with the broader goals of precision medicine[7,8]. However, despite these promising advancements, several challenges continue to limit widespread clinical adoption. Variability in dataset quality, limited external validation across diverse populations, ethical and regulatory concerns, and disparities in healthcare infrastructure remain important considerations. Addressing these challenges through responsible implementation strategies and collaborative global research efforts will be essential for ensuring that AI is safely and effectively integrated into routine orthopaedic practice[9,10].

LITERATURE SEARCH METHODOLOGY

This narrative review was conducted using a structured literature search to identify relevant studies on AI applications in orthopaedics. Electronic databases including PubMed/MEDLINE, Scopus, Web of Science, and Google Scholar were searched for articles published between January 2015 and January 2026. The search strategy incorporated combinations of keywords and Medical Subject Headings terms including “artificial intelligence”, “machine learning”, “deep learning”, “orthopaedics”, “musculoskeletal imaging”, “surgical planning”, “clinical decision support”, “arthroplasty”, “spine surgery”, “sports medicine”, and “orthopaedic oncology”.

Studies were selected based on relevance to clinical applications of AI in orthopaedics, including diagnostic imaging, surgical planning, operative assistance, outcome prediction, and rehabilitation. Priority was given to systematic reviews, meta-analyses, clinical studies, and high-quality narrative reviews published in peer-reviewed journals. Articles focusing primarily on non-clinical experimental models, non-musculoskeletal applications, conference abstracts without full text, or lacking sufficient clinical relevance were excluded.

Reference lists of relevant articles were also manually screened to identify additional pertinent publications. In total, approximately 220 articles were screened, and 180 relevant publications were included in this narrative review based on clinical applicability, scientific quality, and relevance to orthopaedic practice. This review was conducted as a structured narrative synthesis rather than a formal systematic review, with the objective of providing a comprehensive overview of current clinical applications, technological advancements, limitations, and future directions of AI in orthopaedics.

FUNDAMENTALS OF AI FOR ORTHOPAEDIC CLINICIANS

AI refers to computational systems capable of performing tasks traditionally requiring human cognition, including pattern recognition, decision-making, and predictive modelling. ML, a subset of AI, enables algorithms to identify patterns from datasets without explicit programming, whereas DL, a specialised form of ML, utilises multilayer neural networks capable of analysing complex high-dimensional datasets such as medical imaging[11,12]. In orthopaedics, these technologies have transformed diagnostic imaging, surgical planning, and outcome prediction.

AI, ML, and DL

ML algorithms rely on statistical models trained using structured datasets such as electronic health records, biomechanical parameters, and clinical registries. DL models, particularly convolutional neural networks (CNNs), analyse unstructured data including radiographs, computed tomography (CT), and magnetic resonance imaging (MRI) scans by extracting hierarchical imaging features. These systems demonstrate improved classification, detection, and prognostic prediction in multiple orthopaedic conditions[13,14].

Computer vision

Computer vision is one of the most widely applied AI technologies in orthopaedics, enabling automated interpretation of imaging datasets for fracture detection, joint degeneration grading, tumour characterisation, and implant assessment. AI-assisted imaging interpretation demonstrates diagnostic performance comparable to expert clinicians while reducing inter-observer variability and reporting time[13,15].

Radiomics

Radiomics involves quantitative extraction of imaging features such as texture, shape, and signal intensity that are not visually appreciable. When combined with AI, radiomics enables tumour characterisation, infection differentiation, and treatment response prediction, supporting personalised clinical decision-making[13,16].

Natural language processing

Natural language processing (NLP) enables AI systems to analyse unstructured clinical text from operative notes, radiology reports, and electronic health records. NLP facilitates automated data extraction, clinical documentation assistance, and large-scale orthopaedic research analytics[17].

Predictive analytics

Predictive AI models integrate clinical variables, imaging findings, and patient-reported outcomes to estimate complication risk, functional recovery, implant survivorship, and disease progression. These models support personalised treatment planning and perioperative optimisation[6].

Multimodal AI

Modern AI systems increasingly integrate heterogeneous datasets including imaging, clinical records, biomechanical data, and molecular biomarkers. Multimodal AI models demonstrate superior diagnostic accuracy compared with single-modality approaches and represent a major advancement toward precision orthopaedic medicine[14,15].

Foundation models and transfer learning

More recently, large foundation AI models trained on datasets collected from multiple institutions have expanded the scope of transfer learning within orthopaedics. By enabling algorithms to adapt previously learned information to new clinical tasks, these models improve overall reliability while reducing the need to generate dedicated datasets for each individual application[12].

Explainable AI

Explainable AI (XAI) frameworks have been developed to improve transparency by highlighting the imaging features and clinical variables that influence algorithmic predictions. By providing greater clarity regarding how decisions are generated, these models help strengthen clinician confidence, support regulatory evaluation, and promote the ethical integration of AI technologies into orthopaedic clinical practice[18]. Nevertheless, despite rapid technological advancement, several challenges continue to limit widespread implementation. Variations in dataset quality, difficulties in model interpretability, and challenges related to integration into routine clinical workflows remain significant barriers to broader adoption[19]. Interpretability is important for clinical use of AI, as clinicians must understand how predictions are generated. Current XAI methods can be grouped into four main types: Gradient-based, perturbation-based, attention-based, and concept-based approaches.

Gradient-based methods, such as saliency maps and Grad-CAM, highlight image regions influencing predictions but may vary with small input changes. Perturbation-based methods, such as occlusion testing and local interpretable model-agnostic explanations, assess how input modifications affect outputs, although results may be inconsistent. Attention-based methods assign importance weights to input features but do not always reflect true causal relationships. Concept-based approaches link predictions to clinically meaningful features but can be difficult to standardise. These methods improve transparency but may show variability and do not always fully represent model reasoning. Further improvements in interpretability will be essential for reliable clinical implementation. The principal AI methodologies and their clinical orthopaedic applications are summarised in Table 1.

Table 1 Core artificial intelligence techniques and their orthopaedic clinical applications.

AI technique	Description	Orthopaedic application
Machine learning	Algorithms that learn patterns from structured data	Outcome prediction, complication risk analysis
Deep learning	Multilayer neural networks analysing complex datasets	Fracture detection, tumour classification
Computer vision	Automated interpretation of medical images	Imaging diagnosis, surgical navigation
Radiomics	Quantitative extraction of imaging features	Tumour grading, infection differentiation
Natural language processing	Analysis of unstructured clinical text	EMR analysis, research data extraction
Predictive analytics	Statistical modelling for outcome estimation	Implant survivorship prediction
Multimodal AI	Integration of heterogeneous datasets	Precision orthopaedic decision-making
Explainable AI	Transparent model reasoning systems	Clinical trust and regulatory validation

AI: Artificial intelligence; EMR: Electronic medical record.

DATA ECOSYSTEM IN ORTHOPAEDICS

The effectiveness of AI in orthopaedics depends on high-quality multimodal datasets. The orthopaedic data ecosystem includes imaging, clinical records, surgical data, biomechanical monitoring, molecular datasets, and large-scale registries that collectively enable development of predictive and diagnostic AI models[12,15].

Medical imaging data

Radiographs, CT scans, and MRI form the foundation of orthopaedic AI development. Imaging datasets allow automated detection, classification, and prognostic evaluation of fractures, degenerative disorders, tumours, and implant complications. AI-based imaging analysis improves diagnostic consistency and workflow efficiency[13,15].

Electronic medical records and patient-reported outcomes

Electronic medical records provide structured demographic, laboratory, and clinical datasets, while patient-reported outcome measures capture functional recovery and quality-of-life metrics. Integration of these datasets supports development of predictive models for postoperative outcomes, complications, and rehabilitation strategies[6].

Arthroscopy and surgical video data

Video-based AI analysis is increasingly utilised in arthroscopic procedures for anatomical landmark recognition, automated surgical phase identification, and objective surgical skill evaluation. These applications support intraoperative decision-making, training, and quality improvement initiatives[20].

Surgical navigation and intraoperative sensor data

Computer-assisted navigation systems generate real-time intraoperative datasets including alignment parameters, instrument trajectories, and biomechanical metrics. AI analysis of these datasets supports precision surgery and complication prediction[14].

Wearables and gait analysis

Wearable sensors enable continuous monitoring of gait patterns, joint biomechanics, and rehabilitation progress. AI-driven wearable data analysis supports early detection of functional deterioration and personalised rehabilitation planning[17].

Genomics and biomarker data

Integration of molecular biomarkers, genetic data, and inflammatory markers enables development of precision orthopaedic AI models supporting disease susceptibility prediction, infection risk stratification, and tumour behaviour assessment[15].

Orthopaedic registries and big data repositories

National arthroplasty registries and trauma databases provide longitudinal real-world datasets essential for AI model development and validation. Registry-based AI research enables prediction of implant survivorship, complication risk, and functional outcomes across diverse patient populations[5,6].

Internet of Medical Things and smart orthopaedic devices

The Internet of Medical Things integrates wearable sensors, smart implants, and remote monitoring devices that generate continuous patient-specific data. AI analysis of these datasets supports postoperative monitoring, early complication detection, and personalised rehabilitation[15].

Federated learning and privacy-preserving AI

Federated learning enables collaborative AI development across institutions without direct patient data sharing, improving dataset diversity and addressing privacy concerns[21].

Digital twin data integration

Digital twin platforms combine multimodal datasets to create virtual biomechanical patient models, enabling simulation of surgical interventions, postoperative outcome prediction, and personalised treatment optimisation[15]. The multimodal data ecosystem supporting AI applications in orthopaedics is illustrated in Figure 1.

Figure 1 Artificial intelligence-enabled orthopaedic care workflow. Artificial intelligence supports clinical assessment, diagnosis, treatment planning, artificial intelligence-guided surgery, rehabilitation monitoring, and outcome surveillance, enabling improved decision-making and patient outcome optimisation across the orthopaedic care continuum. AI: Artificial intelligence.

AI IN MUSCULOSKELETAL IMAGING

AI, particularly ML and DL, has emerged as a transformative technology in medical imaging. In musculoskeletal radiology, AI applications include fracture detection, osteoarthritis grading, osteoporosis screening, tumour characterisation, infection differentiation, and implant assessment. By leveraging large imaging datasets and computational modelling, AI improves diagnostic accuracy, reduces inter-observer variability, and enhances workflow efficiency. These developments are particularly relevant in orthopaedics, where imaging is fundamental to diagnosis, treatment planning, and postoperative monitoring[22,23].

AI techniques in musculoskeletal imaging

Most AI applications in musculoskeletal imaging utilise CNNs, which automatically extract hierarchical imaging features from radiographs, CT, and MRI, eliminating the need for manual feature engineering[24]. Radiomics combined with AI enables quantitative evaluation of bone and soft-tissue characteristics, supporting improved diagnostic precision and prognostic prediction[25].

Fracture detection: Fracture detection is among the most validated AI applications in musculoskeletal imaging. DL algorithms have demonstrated diagnostic performance comparable to expert radiologists on plain radiographs[26,27]. AI systems are particularly effective in identifying subtle fractures of the hip, wrist, ankle, and proximal humerus, which are frequently missed in emergency settings[28]. AI-assisted decision-support systems reduce reporting time and diagnostic errors, although variability across institutions and imaging protocols necessitates further external validation[29].

Osteoarthritis: AI algorithms enable automated assessment of joint space narrowing, osteophytes, and subchondral sclerosis, facilitating reproducible grading using established classification systems such as the Kellgren-Lawrence scale[30]. Improved grading consistency allows reliable longitudinal monitoring and supports personalised disease management strategies[31].

Osteoporosis: AI models can estimate bone mineral density and fracture risk using routine CT and radiographic imaging, enabling opportunistic osteoporosis screening without additional imaging or radiation exposure[32]. Early detection using AI-based assessment may reduce fragility fracture burden, particularly in underdiagnosed populations[33].

Bone tumours: Radiomics-based AI systems assist in differentiating benign from malignant bone tumours and predicting tumour grade and aggressiveness by analysing complex imaging features[34]. AI also improves tumour margin delineation and treatment response evaluation. Clinical translation remains limited due to dataset heterogeneity and lack of standardised imaging protocols[35].

Infection (tubercular vs pyogenic): AI is increasingly being investigated for detecting musculoskeletal infections and inflammatory changes on MRI and CT. Early models demonstrate potential in recognising infection patterns and disease extent, although differentiation between tubercular and pyogenic infections requires further validation[36]. This application is particularly relevant in regions with high tuberculosis prevalence, where delayed diagnosis contributes to significant morbidity.

Implant loosening and postoperative assessment: AI facilitates automated evaluation of implant alignment, positioning, and early loosening. Detection of radiolucent lines and subtle migration patterns may allow earlier identification of implant failure and improve long-term outcomes[37]. Integration with arthroplasty registry data may further enhance implant survivorship prediction.

Workflow optimisation

AI also improves imaging workflow through automated triage, protocol selection, and structured reporting, reducing radiologist workload and improving turnaround times while supporting clinical decision-making[23,38].

AI in soft tissue imaging

AI is increasingly applied to musculoskeletal soft-tissue imaging, including ligament, tendon, muscle, and soft-tissue tumour evaluation. DL improves analysis of complex MRI and CT datasets, enhancing detection of sports-related injuries. Multimodal AI integrates imaging with biomechanical and physiological data, enabling injury risk prediction and personalised rehabilitation planning. Radiomics-driven AI further improves tumour characterisation and treatment response assessment. Foundation and multimodal models enhance segmentation accuracy, reduce acquisition time, and improve workflow efficiency. However, anatomical complexity, motion artefacts, dataset heterogeneity, and algorithm transparency remain barriers to large-scale clinical adoption[12,39].

Multimodal AI integrates imaging with clinical and demographic data, enabling comprehensive disease modelling beyond single-modality systems. Advanced architectures, including transformer-based and graph neural network models, improve disease characterisation and clinical generalisability, facilitating translation of AI research into precision orthopaedic practice[40].

Limitations and future directions

Despite significant progress, several challenges limit widespread clinical implementation of AI in musculoskeletal imaging. Many current models lack robust external validation and demonstrate reduced generalisability across diverse imaging protocols and patient populations. Dataset heterogeneity, inconsistent annotation standards, and algorithmic bias remain major technical limitations. Additionally, medicolegal concerns, data privacy issues, and accountability for AI-assisted decision-making continue to pose regulatory challenges.

Future research is expected to focus on large multicentre prospective validation studies, development of XAI frameworks, and integration of multimodal data combining imaging, clinical, biomechanical, and genomic information. Emerging technologies such as foundation models and continuous learning systems may enhance model adaptability and precision. Successful translation of AI into routine orthopaedic practice will depend on standardised validation frameworks, ethical governance, and seamless integration into clinical workflows[18].

AI has shown significant promise in musculoskeletal imaging and orthopaedic practice. While it is unlikely to replace clinicians, AI can enhance diagnostic accuracy, efficiency, and consistency. With continued technological progress and responsible implementation, AI is expected to become an important component of musculoskeletal radiology and precision orthopaedic care.

AI IN TRAUMA AND EMERGENCY ORTHOPAEDICS

Triage

AI enhances trauma triage by rapidly analysing vital signs, imaging, and clinical parameters, often outperforming traditional scoring systems such as the Emergency Severity Index. AI systems integrated with picture archiving and communication systems enable automated prioritisation of urgent orthopaedic cases and improve workflow efficiency[5,9,41]. Prehospital AI models have demonstrated high accuracy in predicting injury severity using physiological parameters, facilitating early resource allocation and optimised emergency department preparedness[42]. Implementation of AI-assisted triage has been associated with improved efficiency and reduced diagnostic errors, particularly in high-volume trauma centres[43].

Missed injury detection

AI demonstrates strong performance in detecting subtle fractures that may be overlooked by trainees or non-specialists. CNNs have achieved diagnostic accuracy of 90%-98% on radiographs, often exceeding resident performance[44]. Clinical studies have reported approximately 10% reduction in missed fractures with AI-assisted interpretation, helping bridge the diagnostic gap between junior clinicians and expert radiologists[45].

Polytrauma prioritisation

AI models integrate injury severity scores, imaging findings, and physiological data to stratify polytrauma patients and optimise treatment sequencing[46]. ML algorithms have been successfully applied to predict haemorrhagic shock, coagulopathy, transfusion requirements, need for emergency surgical intervention, intensive care admission, and timing of fracture fixation using real-time analysis of clinical and laboratory data[47,48]. AI-based trauma simulations also provide advanced training platforms for resident education and clinical decision-making[49].

Surgical urgency prediction

ML models are increasingly used to determine optimal timing of fracture reconstruction in patients with concurrent systemic injuries, guiding decisions between urgent and delayed fixation. Predictive analytics can estimate risks of infection, implant failure, and mortality associated with delayed surgical intervention[50]. AI has also demonstrated superior performance compared with surgeons in classifying surgical urgency in hip fractures and predicting mortality risk. AI-driven decision-support systems further assist in operating room prioritisation and surgical team allocation[44,51].

Fracture healing prediction

AI models analysing sequential radiographs and clinical variables are increasingly being used to predict fracture healing progression, identify risk of delayed union or nonunion, and support personalised postoperative rehabilitation planning[26,52].

Prehospital and remote trauma AI

AI-assisted prehospital trauma systems integrating wearable monitoring, telemedicine imaging, and predictive analytics are emerging as tools to improve early injury recognition and optimise transport decisions and resource allocation, particularly in remote and mass casualty settings[42].

Autonomous imaging triage

Autonomous AI-based imaging triage platforms integrated within radiology workflows enable automatic prioritisation of critical musculoskeletal injuries, improving reporting efficiency and reducing time to definitive treatment[42].

Overall perspective

AI is rapidly advancing trauma and emergency orthopaedics by enabling automated data integration, early risk stratification, and improved clinical decision support across the trauma care continuum. These technologies facilitate proactive trauma management and support personalised, data-driven treatment strategies.

AI IN SPINE SURGERY

Diagnosis

AI has been widely applied to imaging-based diagnosis in spine surgery, particularly for degenerative disorders and spinal deformities[53-55]. CNNs and related DL models automatically identify vertebral levels, quantify spinal canal stenosis, measure Cobb angles and spinopelvic parameters, and detect fractures or implant-related complications on radiographs, CT, and MRI with improved consistency and efficiency[56-58]. However, most diagnostic algorithms are developed using single-centre or homogeneous datasets, raising concerns regarding generalisability across populations and imaging protocols[59,60].

Surgical planning

AI has emerged as a valuable tool in preoperative planning, particularly in spinal deformity correction and complex reconstructive procedures. ML models can recommend osteotomy levels and fusion extent required to achieve optimal spinopelvic balance. AI-based three-dimensional templating platforms simulate various surgical strategies, optimise implant placement and screw trajectories, and predict their impact on alignment and implant loading[61-63].

Alignment optimisation

AI-driven models incorporate preoperative spinopelvic parameters, patient age, body mass index, bone quality, and radiographic patterns to recommend individualised alignment targets based on spinal morphotypes[60]. These models predict postoperative alignment across different osteotomy techniques and help prevent under- or over-correction[64]. Intraoperative AI-assisted tracking of anatomical landmarks and instrumentation can help maintain planned alignment in real time[62]. Preliminary evidence suggests AI-guided alignment strategies may reduce mechanical complications such as proximal junctional kyphosis and rod fracture, although current evidence remains largely retrospective with limited long-term validation[65,66].

Outcome prediction

ML models have been developed to predict patient-reported outcomes, including Oswestry Disability Index scores, complication risks, revision surgery, pseudoarthrosis, transfusion requirements, and hospital length of stay following spinal deformity and degenerative procedures. Multimodal and multitask learning frameworks integrating wearable activity monitoring and electronic health record data have demonstrated potential in predicting postoperative pain interference and functional recovery[67-71].

Overall perspective

Recent literature suggests that AI has progressed from theoretical applications to early clinical integration across the spine surgery continuum, encompassing diagnosis, surgical planning, alignment optimisation, and outcome prediction. However, most available studies remain retrospective and single-centre with variable validation quality, highlighting the need for standardised reporting, multicentre datasets, and prospective clinical evaluation. At present, AI should be considered an adjunctive tool that enhances diagnostic precision, surgical planning, and risk stratification, while final clinical judgment and ethical responsibility remain with the surgeon.

AI IN ARTHROPLASTY

Implant selection

AI has transformed preoperative templating by enabling automated identification of patient-specific anatomy and prediction of optimal implant type and size[72,73]. Conventional two-dimensional templating is limited by magnification errors and inter-observer variability. DL models have demonstrated greater than 90% accuracy in predicting component size in total knee arthroplasty and total hip arthroplasty, outperforming manual templating methods[74-76]. AI-driven implant identification tools further assist preoperative planning by accurately recognising implant models, manufacturers, and sizes, particularly in revision cases where prior documentation may be unavailable[77,78].

Alignment

Accurate implant alignment is critical for implant longevity and functional outcomes. AI-integrated robotic arthroplasty and computer vision-based navigation systems provide real-time kinematic feedback through three-dimensional anatomical reconstruction, enabling sub-millimetre precision during bone preparation[79-81]. AI-assisted three-dimensional planning has demonstrated improved hip-knee-ankle alignment and reduced alignment outliers compared with conventional techniques, contributing to improved postoperative functional outcomes[81].

Infection risk prediction

Periprosthetic joint infection remains a major complication following arthroplasty. ML models analyse complex datasets including patient comorbidities, nutritional status, operative variables, and biomechanical factors to generate personalised infection risk predictions that support perioperative optimisation strategies. Generative AI models have also been evaluated for patient-specific infection risk assessment, demonstrating moderate predictive accuracy and potential for real-time clinical decision support[77,82,83].

Revision prediction

Predicting implant failure due to aseptic loosening or mechanical complications is essential for long-term implant survivorship. ML models utilise registry-based data, including demographic variables, implant characteristics, prior surgical history, radiographic progression, and patient-reported outcome measures, to predict early revision risk and guide surveillance protocols[84-86].

Overall perspective

AI has rapidly evolved from a theoretical concept to a clinically relevant tool in modern arthroplasty. By integrating ML, DL, and computer vision technologies, AI supports personalised surgical planning, optimises implant positioning, and improves complication prediction, facilitating the transition from standardised surgical approaches to precision orthopaedic care.

AI IN SPORTS MEDICINE AND ARTHROSCOPY

Anterior cruciate ligament

CNNs demonstrate high diagnostic accuracy in detecting complete and partial anterior cruciate ligament tears on MRI, with pooled sensitivity of 90%-96% and specificity of 92%-97%, frequently outperforming experienced surgeons[87-90]. AI-driven image-free navigation systems utilise three-dimensional bone morphing to optimise femoral and tibial tunnel placement and reduce graft malposition[91]. AI models also enable intraoperative multidirectional laxity assessment and graft tension optimisation, while ML models incorporating biomechanical parameters can predict reconstruction failure risk[92,93].

Meniscal injuries

AI algorithms accurately detect and localise meniscal tears on MRI, with sensitivity of 87% and specificity of 89%, comparable to expert radiologists while improving interpretation efficiency. Multimodal DL models integrating MRI and arthroscopic findings further enhance diagnostic accuracy and surgical planning[94-97].

Rotator cuff pathology

Three-dimensional CNNs detect and classify rotator cuff tears based on size and anatomical location on MRI, achieving diagnostic accuracy of 90%-94%. Systematic reviews confirm reliable differentiation of rotator cuff tears from other shoulder pathologies using AI-based imaging analysis[98-100].

Muscle injury prediction

ML models integrate physiological stress markers and external workload parameters to predict muscle injury risk with high accuracy[101]. Support vector machine algorithms have demonstrated approximately 93% accuracy in classifying muscle atrophy severity using biochemical and morphometric data[102,103].

Return-to-sport prediction

ML models predict return-to-sport outcomes following anterior cruciate ligament reconstruction using early postoperative functional assessments. Random forest algorithms outperform conventional regression models by capturing complex interactions between biomechanical, fatigue-related, and psychological recovery variables[104,105].

Arthroscopy video analysis

AI is increasingly applied to arthroscopic video analysis for automated tissue identification, tear classification, instrument tracking, and surgical workflow recognition, enabling real-time intraoperative decision support and objective surgical performance assessment[106].

Wearable rehabilitation monitoring

AI-integrated wearable sensors enable continuous monitoring of biomechanics, rehabilitation adherence, and functional recovery, allowing early detection of reinjury risk and personalised rehabilitation optimisation[107].

XAI

XAI frameworks improve transparency and clinical interpretability of sports injury prediction and diagnostic models, enhancing clinician trust and supporting ethical integration of AI-assisted decision-making[108].

Overall perspective

AI is reshaping sports medicine and arthroscopy by improving diagnostic precision, surgical optimisation, injury prediction, and personalised rehabilitation through integration of biomechanical, imaging, and clinical datasets.

AI IN ORTHOPAEDIC ONCOLOGY

Orthopaedic oncology includes primary bone tumours, soft-tissue sarcomas, and metastatic lesions, where early diagnosis, accurate grading, and optimal surgical planning are essential for survival and limb preservation. AI, including ML and DL, enhances diagnostic accuracy and prognostic assessment by integrating imaging, clinical, and pathological data[22,38].

Tumour detection

AI algorithms demonstrate strong performance in detecting bone and soft-tissue tumours on radiographs, CT, and MRI. DL models identify imaging features such as cortical destruction, marrow infiltration, periosteal reaction, and soft-tissue extension, improving early tumour recognition and reducing diagnostic delays, particularly in rare malignancies[18,34,109,110].

Grade prediction

Radiomics-based AI models enable non-invasive tumour grading by analysing imaging texture, morphology, and signal intensity. These models reliably differentiate low-grade from high-grade sarcomas, supporting preoperative planning and therapy selection while reducing biopsy-related sampling errors[111-114].

Margin mapping and surgical planning

AI-assisted segmentation allows precise tumour boundary delineation and three-dimensional visualisation, improving margin mapping and limb-salvage surgical planning. Integration with navigation systems and three-dimensional printing enhances resection accuracy and may reduce local recurrence[25,115,116].

Metastasis risk prediction

AI models combining imaging, clinical, and pathological data predict metastatic risk and disease progression. Radiomics-based biomarkers enable risk stratification, personalised surveillance, and identification of patients who may benefit from systemic therapy[117-119].

Clinical integration

AI functions as a decision-support tool that enhances multidisciplinary tumour board decision-making, reduces inter-observer variability, and improves consistency in oncologic management[120].

Limitations and future directions

Clinical implementation remains limited by small datasets, tumour heterogeneity, lack of external validation, and data privacy concerns. Future research should focus on multicentre validation and multimodal integration of imaging, histopathology, and genomic data, supported by XAI frameworks[121,122]. AI is transforming orthopaedic oncology into a precision oncology framework by integrating imaging, clinical, and pathological data, while radiomics-based AI enables non-invasive tumour characterisation through quantitative imaging biomarkers that reflect tumour heterogeneity and aggressiveness; furthermore, multimodal AI models combining radiological, pathological, and clinical variables have demonstrated superior accuracy in predicting metastasis and disease progression compared with traditional prognostic systems. AI improves tumour detection, grading, surgical planning, and metastasis prediction in orthopaedic oncology. With continued validation and ethical integration, AI is expected to become a key adjunct in personalised oncologic care.

AI IN ORTHOPAEDIC INFECTION

AI is increasingly being applied in orthopaedic infections to improve early diagnosis, risk prediction, pathogen identification, and treatment optimisation. By integrating clinical data, laboratory parameters, microbiological profiles, and imaging findings, AI models enhance diagnostic accuracy and support personalised antimicrobial and surgical decision-making. These applications are particularly relevant in complex infections such as periprosthetic joint infection, osteomyelitis, and implant-related infections, where delayed diagnosis significantly worsens outcomes[4,123,124].

Diagnosis and early detection

AI-driven models have demonstrated strong performance in diagnosing orthopaedic infections by analysing multimodal datasets including inflammatory markers, microbiological results, and imaging findings. ML algorithms have shown improved diagnostic accuracy compared with traditional criteria in identifying periprosthetic joint infection and differentiating aseptic loosening from infection[83,124]. Radiomics-based AI systems analysing magnetic resonance and nuclear imaging demonstrate potential for detecting early osteomyelitis and implant-related infections before radiographic changes become evident[125].

Pathogen identification and antibiotic optimisation

AI models have been utilised to predict causative microorganisms using clinical, laboratory, and demographic variables, enabling targeted antimicrobial therapy before culture results become available. ML algorithms demonstrate high accuracy in predicting antimicrobial resistance patterns and guiding empiric antibiotic selection, thereby reducing inappropriate antimicrobial use and improving infection control outcomes[4,126].

Risk stratification and outcome prediction

Predictive AI models incorporating patient comorbidities, surgical variables, implant characteristics, and perioperative laboratory parameters have demonstrated effectiveness in identifying patients at increased risk of surgical site infections and periprosthetic joint infections. These models facilitate perioperative optimisation, early surveillance, and individualised treatment planning. AI-based outcome prediction systems have also been developed to estimate infection recurrence, treatment failure, and implant survivorship following revision surgery[127-129].

Imaging and surgical planning

AI-assisted imaging analysis improves differentiation between infection, inflammation, and aseptic loosening by detecting subtle imaging patterns not readily appreciable through conventional interpretation. These systems assist surgical planning by identifying infection extent, guiding debridement strategies, and predicting bone and soft-tissue involvement[125,130,131].

Monitoring treatment response

AI models analysing longitudinal laboratory parameters, imaging progression, and patient-reported outcomes demonstrate potential in monitoring treatment response and predicting infection resolution. Continuous data integration enables dynamic risk assessment and supports personalised antimicrobial duration and follow-up strategies[128].

Biofilm prediction and detection

AI is increasingly being investigated for predicting biofilm formation and persistence on orthopaedic implants by analysing microbial, biomaterial, and host-related factors. Biofilm formation plays a central role in chronic osteomyelitis and implant-associated infections by protecting microorganisms from host immune responses and antimicrobial therapy. AI-driven models integrating microbial characteristics and host factors may improve surgical decision-making, including implant retention or staged revision strategies[132].

Genomic and metagenomic infection detection

AI-assisted analysis of microbial genomic and metagenomic sequencing data is emerging as a powerful tool for rapid pathogen identification and antimicrobial resistance prediction, particularly in culture-negative orthopaedic infections. Metagenomic next-generation sequencing demonstrates superior sensitivity compared with conventional microbial cultures and enables detection of polymicrobial infections, facilitating earlier targeted therapy and improved antimicrobial stewardship[53,133].

AI-driven antimicrobial stewardship and personalised therapy

ML-driven antimicrobial stewardship models optimise antibiotic selection, dosing, and treatment duration by integrating patient-specific risk factors, microbial resistance patterns, and infection severity. AI-based predictive systems demonstrate potential in guiding empiric antimicrobial therapy, reducing inappropriate broad-spectrum antibiotic use, and improving treatment outcomes in periprosthetic joint and musculoskeletal infections[134,135].

Limitations and future directions

Despite promising results, clinical adoption of AI in orthopaedic infections remains limited by dataset heterogeneity, variability in diagnostic definitions, and lack of multicentre prospective validation. Additional challenges include algorithm interpretability, data privacy concerns, and integration into routine clinical workflows. Future research is expected to focus on multimodal AI models combining microbiological genomics, imaging, and clinical datasets to enhance precision infection management.

AI is emerging as a valuable adjunct in orthopaedic infection management by improving diagnostic accuracy, pathogen prediction, risk stratification, surgical planning, and treatment monitoring. With continued validation and integration into multidisciplinary infection care pathways, AI has the potential to significantly improve patient outcomes and antimicrobial stewardship in orthopaedic practice.

AI IN OPERATIVE ORTHOPAEDICS

AI has emerged as an important adjunct in operative orthopaedics through robotics, computer-assisted navigation, intra-operative decision support, fluoroscopic image interpretation, and augmented reality-based visualisation. These technologies enhance surgical precision, reproducibility, and safety through real-time integration of imaging, sensor feedback, and predictive analytics. AI-enabled intra-operative tools have demonstrated improved implant alignment, reduced technical outliers, optimised radiation usage, and enhanced visualisation of anatomical structures[136-138]. However, challenges including cost, learning curve, data generalisability, and regulatory considerations remain barriers to widespread adoption.

Robotics in orthopaedic surgery

AI-enabled robotic systems are increasingly utilised in joint arthroplasty and are expanding into spine and trauma procedures. These platforms integrate preoperative imaging with intra-operative haptic feedback and robotic motion control to optimise bone preparation and implant positioning. Robotic-assisted total knee arthroplasty has demonstrated improved component alignment and reduced mechanical outliers compared with conventional instrumentation[136,137]. In spine surgery, robotic platforms assist pedicle screw trajectory planning and placement, improving accuracy and reducing revision rates in selected patient cohorts[138]. Emerging semi-autonomous robotic platforms are capable of executing predefined surgical steps under surgeon supervision, integrating real-time imaging and sensor feedback to enhance procedural precision and intra-operative safety[139,140].

Computer-assisted navigation

Navigation systems combine optical trackers, intra-operative sensors, and imaging datasets to provide real-time alignment feedback. AI enhances these platforms by predicting optimal implant positioning and dynamically adjusting alignment parameters during surgery. Studies have demonstrated improved coronal and sagittal alignment accuracy in knee arthroplasty using navigation systems, although long-term functional superiority remains under investigation[100]. In spinal instrumentation, navigation technologies have been associated with reduced pedicle screw misplacement and decreased radiation exposure compared with freehand techniques[138,141].

Intra-operative decision support systems

AI-driven decision support tools analyse patient-specific anatomical data, soft-tissue balancing metrics, and real-time intra-operative sensor feedback to assist surgical decision-making. These systems help guide implant sizing, ligament balancing, and implant positioning, reducing technical variability and cognitive workload. Emerging evidence suggests improved consistency in procedures such as arthroplasty and complex deformity correction, while preserving surgeon autonomy[4,137].

Fluoroscopy interpretation and radiation optimisation

Computer vision-based AI models support intra-operative fluoroscopy interpretation by helping surgeons better assess fracture reduction, improve implant positioning, and more accurately identify anatomical landmarks. Automated image analysis reduces the need for repeated imaging, thereby lowering cumulative radiation exposure to surgical teams. In trauma and spine procedures, AI-assisted real-time trajectory guidance improves operative efficiency and surgical confidence[27].

Augmented reality in orthopaedic surgery

Augmented reality platforms enhance intra-operative visualisation through head-mounted displays or projection-based interfaces that overlay digital anatomical reconstructions and alignment guidance. AI enhances augmented reality systems by enabling real-time anatomical differentiation and dynamic adjustment of overlays during surgical motion. Early clinical studies in spine and arthroplasty procedures suggest improved anatomical visualisation and reduction in alignment and fracture reduction errors, highlighting augmented reality as a promising interface for immersive surgical guidance[142].

Smart implants and sensor-enabled prostheses

Smart orthopaedic implants embedded with biosensors enable continuous monitoring of implant loading, joint kinematics, and early indicators of implant loosening or infection. AI-based analysis of implant-derived data facilitates early detection of mechanical failure and supports personalised postoperative surveillance and rehabilitation strategies[143,144].

Digital twin technology in orthopaedic surgery

Digital twin models utilise patient-specific anatomical, biomechanical, and physiological data to create virtual simulations of surgical interventions. AI-driven digital twins enable preoperative optimisation of implant positioning, prediction of postoperative biomechanics, and personalised surgical planning, representing a major advancement in precision orthopaedic surgery[145,146].

AI-driven surgical workflow optimisation

AI models analysing intra-operative workflow data can identify procedural inefficiencies, predict operative duration, optimise instrument usage, and enhance operating room resource allocation. These systems contribute to improved surgical efficiency, reduced operative variability, and enhanced patient safety[147,148].

Limitations

While AI technologies in operative orthopaedics continue to advance, their integration into routine clinical practice remains gradual. Practical concerns such as the cost of infrastructure, temporary disruption of established surgical workflows during early adoption, and the time required for surgeons to become familiar with these systems continue to pose challenges. Medico-legal uncertainties related to algorithm-supported decision-making also remain an important consideration. In addition, concerns regarding potential algorithmic bias, protection of patient data, and regulatory requirements highlight the need for careful validation and responsible oversight. For these reasons, AI-based technologies are more appropriately viewed as supportive tools that enhance surgical precision and assist clinical judgement, rather than substitutes for surgical expertise[4,137].

AI technologies such as robotics, navigation platforms, intra-operative decision support, advanced fluoroscopic guidance, augmented reality, smart implants, digital twin modelling, and workflow optimisation are gradually reshaping operative orthopaedics. By incorporating real-time imaging, sensor-derived data, and predictive analytical tools, these innovations help surgeons achieve greater precision and procedural consistency, while ensuring that clinical decision-making remains under direct surgeon supervision. Continued prospective validation, cost-effectiveness evaluation, and ethical implementation will determine the long-term impact of AI on orthopaedic surgical practice. The spectrum of AI applications across major orthopaedic subspecialties is outlined in Table 2.

Table 2 Applications of artificial intelligence across orthopaedic subspecialties.

Subspecialty	AI Application	Clinical benefit
Trauma	Fracture detection	Reduced missed injuries
Spine	Surgical planning	Improved alignment accuracy
Arthroplasty	Implant sizing	Enhanced implant longevity
Sports medicine	Injury prediction	Optimised return-to-sport timing
Oncology	Tumour grading	Personalised treatment planning
Infection	Pathogen prediction	Targeted antimicrobial therapy
Paediatric orthopaedics	Growth modelling	Early deformity detection
Rehabilitation	Wearable monitoring	Personalised recovery plans

AI: Artificial intelligence.

AI-DRIVEN ORTHOPAEDIC CARE PATHWAY

AI is gradually becoming part of everyday orthopaedic practice, mainly by helping clinicians make more informed decisions throughout patient care. Earlier AI applications were often limited to specific tasks such as image interpretation or surgical assistance. However, newer systems are beginning to use information from several sources at the same time, including radiological imaging, clinical findings, biomechanical data, and surgical records, which allows a more comprehensive assessment of patient management from the first consultation to follow-up. This shift reflects the growing emphasis on tailoring treatment to individual patient characteristics while also aiming to improve overall healthcare efficiency and outcomes[6,9].

Patient triage

AI is increasingly being explored as a support tool in early orthopaedic triage, particularly for identifying patients who may require urgent care. By analysing clinical findings, vital parameters, imaging results, and electronic medical records together, AI systems may assist clinicians in prioritising high-risk cases. Several studies suggest that ML models can help estimate injury severity, predict the likelihood of hospital admission, and identify patients who may require early surgical intervention, sometimes performing comparably to traditional triage scoring systems[149,150]. In prehospital settings, AI-enabled wearable monitoring and telemedicine platforms may support earlier recognition of significant injuries and help optimise the use of emergency resources, especially in high-volume trauma environments. Similarly, automated imaging triage tools integrated within radiology workflows may help reduce delays by flagging potentially serious musculoskeletal injuries for prompt review[9].

Imaging and diagnosis

Computer vision and radiomics-based AI systems facilitate automated interpretation of radiographs, CT scans, and MRI, improving detection of fractures, degenerative disorders, tumours, infections, and implant complications. DL models provide diagnostic performance comparable to specialist interpretation while reducing inter-observer variability[16,151]. Multimodal AI models combining imaging, laboratory parameters, and clinical datasets further enhance diagnostic precision and support early disease detection and risk stratification[152].

Surgical decision-making

Predictive AI models support surgical planning by integrating patient-specific anatomical, functional, and comorbidity profiles. These systems assist in determining operative indications, implant selection, alignment targets, and procedural strategies. AI-based patient selection and risk prediction models have demonstrated improved decision-making accuracy and perioperative optimisation. Digital twin simulations allow virtual modelling of surgical interventions and biomechanical outcomes, enabling personalised surgical planning and optimisation of implant positioning and joint biomechanics[6,149].

Operative execution

Intraoperative AI technologies including robotic platforms, computer-assisted navigation, augmented reality visualisation, and decision support systems enhance surgical precision and reproducibility. These systems integrate real-time imaging, sensor feedback, and biomechanical analysis to optimise implant positioning, soft-tissue balancing, and surgical workflow efficiency[152,153]. Computer vision algorithms also support fluoroscopic interpretation and instrument tracking, reducing radiation exposure and improving intraoperative safety[16].

Rehabilitation and functional recovery

AI-driven rehabilitation programmes utilise wearable sensors, gait analysis systems, and patient-reported outcome monitoring to provide continuous functional assessment. ML models predict recovery trajectories, reinjury risk, and return-to-activity timelines, enabling personalised rehabilitation strategies. Integration of biomechanical monitoring and tele-rehabilitation platforms improves adherence and facilitates early intervention for functional deterioration[139].

Long-term surveillance and outcome monitoring

AI enables longitudinal patient monitoring through integration of orthopaedic registries, smart implants, wearable devices, and electronic health records. Predictive models can detect early implant failure, infection recurrence, or functional decline. NLP systems further enable automated outcome surveillance through analysis of clinical documentation and patient-reported outcomes[154]. Continuous surveillance allows personalised follow-up protocols and supports value-based orthopaedic care by improving long-term outcomes and reducing revision surgery rates[6].

Future directions of the AI-integrated orthopaedic pathway

Emerging technologies such as multimodal AI, federated learning, and Internet of Medical Things platforms are expected to further integrate orthopaedic data ecosystems. These systems may enable continuous real-world monitoring, collaborative model development across institutions, and improved generalisability of predictive algorithms. As AI technologies continue to develop, orthopaedic care is expected to gradually move toward more integrated and personalised treatment pathways that combine clinical experience with data-supported decision-making[5,9]. The integration of AI across the orthopaedic clinical pathway is illustrated in Figure 2.

Figure 2 Multimodal data integration framework for artificial intelligence-driven clinical decision support in orthopaedics. Artificial intelligence integrates diverse data sources including imaging, electronic medical records, wearable sensors, surgical navigation systems, registries, genomics, arthroscopy video, and smart implants to support clinical decision-making and enable precision orthopaedic care. EMR: Electronic medical record.

CDSS are increasingly being used to assist orthopaedic surgeons by combining AI with patient-specific clinical details, imaging findings, and biomechanical information. Instead of replacing clinical judgement, these systems aim to support decision-making by analysing large and varied datasets to highlight patterns that may help guide treatment planning. They can offer additional insight when considering treatment strategies, implant selection, surgical approach, and likely clinical outcomes. AI-driven CDSS improves clinical consistency, reduces subjective variability, and facilitates personalised treatment strategies across multiple orthopaedic subspecialties[4,155,156].

Identifying patients requiring surgical intervention

Determining which patients benefit from surgical treatment remains one of the most complex clinical decisions in orthopaedics. ML models integrating clinical examination findings, imaging parameters, patient-reported outcome measures, and comorbidity profiles have demonstrated strong predictive performance in identifying patients likely to benefit from operative intervention[4,157]. AI-based predictive tools have been developed for degenerative joint disease, spinal pathology, sports injuries, and fracture management, enabling early surgical referral and improved patient selection. These systems may also reduce unnecessary surgical procedures by identifying patients who are likely to respond favourably to conservative management[50].

Implant selection and personalised surgical planning

AI-driven CDSS assists surgeons in selecting optimal implant design, size, fixation method, and alignment strategy based on patient-specific anatomical and biomechanical characteristics. DL algorithms analysing radiographic and cross-sectional imaging data have demonstrated high accuracy in predicting implant sizing and alignment targets in arthroplasty procedures. Integration of AI with digital templating and biomechanical modelling enables personalised implant selection, potentially improving implant longevity, functional outcomes, and patient satisfaction[72,158].

Selection of surgical approach and technique

AI models analysing anatomical morphology, deformity characteristics, and surgical risk factors assist in determining optimal surgical approaches and operative strategies. In spine surgery, AI-based planning systems predict osteotomy requirements, fusion levels, and alignment correction targets[62]. In trauma surgery, AI algorithms guide fixation strategy selection and fracture reduction planning. These tools support surgical decision-making by providing objective recommendations while maintaining surgeon autonomy[159].

Prediction of failure of conservative management

AI-driven CDSS can identify patients at high risk of failure of non-operative treatment by analysing demographic variables, imaging features, biomechanical parameters, and baseline functional status. Predictive models have demonstrated effectiveness in identifying patients with osteoarthritis, rotator cuff disease, spinal disorders, and ligament injuries who are unlikely to respond to conservative therapy. Early identification of treatment failure risk facilitates timely surgical intervention and reduces prolonged morbidity associated with ineffective conservative management[50,157].

Outcome prediction and risk stratification

Beyond treatment selection, AI-based CDSS provides predictive modelling of postoperative outcomes including complication risk, implant survival, functional recovery, and revision surgery probability. These models assist in patient counselling, shared decision-making, and perioperative risk optimisation, supporting personalised orthopaedic care[4,72,160].

Limitations and future perspectives

Despite significant advances, widespread implementation of AI-driven CDSS remains limited by dataset variability, algorithm interpretability challenges, and integration difficulties within clinical workflows. Future developments are expected to focus on multimodal data integration, real-time intraoperative decision support, and incorporation of wearable monitoring and registry-based datasets to improve predictive accuracy and clinical applicability[4,159].

AI-powered CDSS represent a major advancement in orthopaedic care by improving surgical decision-making, implant selection, and treatment outcome prediction. When integrated with clinical expertise, CDSS has the potential to enhance personalised orthopaedic care and improve patient outcomes while supporting evidence-based surgical practice[156]. The multilevel architecture through which AI integrates patient data to generate clinical decision support is illustrated in Figure 3.

Figure 3 Artificial intelligence-driven clinical decision support framework in orthopaedics. Patient clinical inputs and multimodal data are integrated and processed using artificial intelligence techniques, including machine learning and deep learning, to support diagnosis, risk stratification, and treatment planning, ultimately improving clinical accuracy, efficiency, safety, and personalised patient care. NLP: Natural language processing.

AI IN ORTHOPAEDIC RESEARCH AND SCIENTIFIC WRITING

AI is increasingly influencing orthopaedic research and scientific writing by improving efficiency across literature screening, data synthesis, manuscript drafting, language editing, and reference management. These technologies have accelerated research productivity and improved accessibility to scientific publishing[161,162]. However, AI-generated academic content introduces important ethical concerns, including fabricated citations, inaccurate data interpretation, algorithmic bias, and challenges related to authorship accountability and research transparency[163,164].

Applications and benefits

AI tools are now being used more frequently during literature reviews, where they help researchers quickly sort through large numbers of publications and identify studies that are most relevant. Some platforms can also assist with extracting key information from selected articles, which may help reduce the time required for data collection. In addition, generative AI programs are increasingly being used to support manuscript preparation by helping with drafting, language improvement, and formatting. This can make academic writing more approachable, particularly for early-career researchers and authors who may not be native English speakers[161,162]. Overall, these developments are helping improve research productivity and may encourage wider participation in orthopaedic academic work across different regions.

Risks of AI-generated content

Despite its potential benefits, the use of AI in academic writing also raises several concerns. Generative AI platforms can occasionally produce incorrect references or invalid digital identifiers, a problem often described as citation hallucination[163,165]. In addition, AI-generated summaries may sometimes misrepresent source material if the content is not carefully reviewed by the author. Overdependence on these tools may also increase the risk of unintended plagiarism and reduce the level of critical scholarly oversight that is essential in academic research[164].

Ghost citations and scientific integrity

Another important concern in academic writing is the generation of ghost references, where AI tools produce citations that appear authentic but do not actually exist. Such errors undermine scientific credibility and may propagate misinformation, particularly in clinical research where inaccurate evidence can influence treatment decisions[163,165]. Independent verification of all references remains essential.

Authorship accountability and transparency

AI systems cannot be recognised as academic authors because they lack intellectual responsibility and accountability. Researchers remain fully responsible for verifying AI-generated content. Increasingly, journals recommend disclosure of AI assistance during manuscript preparation to maintain transparency and editorial integrity[162].

Bias, reproducibility, and editorial use of AI

AI-generated research summaries may reinforce publication bias or underrepresent research from developing regions. Variability in AI-generated outputs may also affect reproducibility of scientific writing[2,164]. AI is additionally being explored in peer-review and editorial workflows for plagiarism detection and manuscript screening, although continued human oversight remains essential[165].

Future perspectives

Emerging AI tools capable of automated source verification and reference validation may reduce risks associated with fabricated citations. Responsible integration of AI into orthopaedic research has the potential to improve productivity while preserving academic accountability and research integrity[161,162].

ETHICAL, LEGAL AND BIAS ISSUES

The expanding integration of AI in orthopaedics introduces important ethical, legal, and governance challenges that must be addressed to ensure safe and equitable clinical implementation. While AI is showing encouraging benefits in improving diagnosis, surgical planning, and treatment decisions, its wider use in everyday clinical practice is still approached with caution. Concerns related to potential algorithmic bias, questions around legal responsibility, challenges in obtaining meaningful informed consent, and the absence of well-defined regulatory frameworks continue to slow its routine adoption in orthopaedics[166,167].

Dataset bias and algorithmic fairness

AI performance is strongly influenced by training dataset quality and diversity. Algorithms trained on limited demographic or geographic populations may demonstrate reduced generalisability and contribute to healthcare disparities. Bias within medical AI systems may perpetuate inequities in clinical decision-making if not appropriately addressed. Multi-institutional datasets, external validation, and continuous monitoring are essential to minimise algorithmic bias and maintain reliability[168,169].

Legal liability and clinical accountability

AI tools are generally used to support clinical judgement rather than replace it. Even when AI provides diagnostic or treatment recommendations, the treating clinician remains responsible for interpreting the information and making final management decisions. However, the question of how legal responsibility should be shared between clinicians, healthcare institutions, and AI developers is still not clearly defined. This ongoing uncertainty highlights the need for clearer medico-legal guidance and governance frameworks as AI becomes more integrated into clinical practice[170,171].

Responsibility in AI-assisted clinical decisions

AI-generated recommendations may occasionally differ from clinical judgement, which can create challenging situations during treatment planning. For this reason, continued human oversight remains essential. Clinicians must critically review algorithm outputs and ensure that decisions are guided by both clinical experience and technological input. Many experts now view human-AI collaboration as a practical approach that may improve diagnostic accuracy while maintaining professional accountability[172].

Patient consent and data privacy

The development of AI systems relies heavily on access to large volumes of patient data, which naturally raises concerns regarding privacy and informed consent. Ethical use of AI requires transparency regarding how patient information is collected, stored, and utilised. Strong cybersecurity safeguards and adherence to national and institutional data protection policies are equally important. There is also growing recognition that patients should be informed when AI contributes to their diagnostic assessment or treatment planning, which helps maintain trust and supports shared decision-making[173,174].

Regulatory approval and clinical validation

Regulatory oversight of AI-based healthcare technologies is continuing to evolve. Many regulatory bodies now require thorough validation before these technologies are introduced into routine clinical practice. Current frameworks emphasise transparency of algorithm design, validation across diverse patient populations, and ongoing safety monitoring after implementation. Continued evaluation is particularly important because AI performance may change over time as new datasets and healthcare environments influence algorithm behaviour[167,175].

Clinical translation challenges: Technical, validation, and operational barriers

Although AI shows strong potential in orthopaedics, several challenges limit its routine clinical use. These challenges can be grouped into technical, validation, and operational barriers. Technical barriers include differences in imaging protocols, variable data quality, and the time and expertise required for accurate data annotation. Models trained on one dataset may perform less reliably in different clinical settings due to differences in patient populations or imaging systems.

Validation barriers remain a major concern. Many studies rely on single-centre retrospective datasets, and external validation using independent, multi-centre data is still limited. Prospective clinical validation studies are also relatively uncommon. Operational barriers include difficulties in integrating AI into clinical workflows, regulatory requirements, and concerns related to user trust and medico-legal responsibility. Addressing these challenges will be important for safe and effective clinical implementation.

Future perspectives

The long-term integration of AI into orthopaedics will likely depend on close collaboration between clinicians, engineers, data scientists, and regulatory authorities. Transparent algorithm development, reliable validation processes, and clearly defined accountability structures will be essential to ensure patient safety while maintaining confidence in these emerging technologies[166,167].

AI AND GLOBAL ORTHOPAEDIC INEQUALITY

AI has the potential to significantly improve orthopaedic diagnosis, treatment planning, and access to specialist care. However, its practical use across different parts of the world remains uneven. Many currently available AI systems are developed using datasets that largely represent populations from high-income countries. As a result, their performance may not always translate reliably to low- and middle-income settings, which can unintentionally widen existing gaps in healthcare delivery[4,156]. Differences in disease patterns, nutritional status, environmental influences, and variations in healthcare resources can all affect how well these technologies function in different populations. Recent reports have also drawn attention to the practical difficulties of applying AI systems designed in resource-rich environments to healthcare systems with limited infrastructure or access to advanced technologies[155,176].

Consequently, AI models may demonstrate reduced accuracy in detecting region-specific conditions such as musculoskeletal tuberculosis, neglected deformities, metabolic bone disease, and malnutrition-associated skeletal disorders, which remain more prevalent in developing regions[177,178]. Additionally, unequal access to digital healthcare infrastructure, advanced imaging technologies, and AI-assisted clinical tools may further widen existing global disparities by disproportionately benefiting healthcare systems with greater technological resources and data availability[50,176]. Addressing these limitations requires development of globally representative datasets, multinational research collaboration, and implementation of cost-effective and scalable AI platforms to ensure equitable integration of AI into orthopaedic care worldwide[4,155,160]. A concise overview of the key advantages and current limitations of AI in orthopaedics is presented in Table 3.

Table 3 Advantages and current limitations of artificial intelligence in orthopaedics.

Advantages	Limitations
Improved diagnostic accuracy	Dataset bias
Reduced inter-observer variability	Limited external validation
Personalised treatment planning	High infrastructure cost
Faster workflow efficiency	Regulatory uncertainty
Predictive outcome modelling	Algorithm interpretability issues
Early complication detection	Data privacy concerns
Enhanced surgical precision	Learning curve for clinicians
Decision support	Risk of overreliance

CROSS-STUDY TRENDS, LIMITATIONS, AND METHODOLOGICAL PATTERNS

A review of the current literature reveals several recurring patterns in the application of AI across orthopaedic subspecialties. CNNs remain the most frequently used models, particularly for image-based tasks such as fracture detection, tumour evaluation, and implant assessment, due to their ability to identify spatial relationships within radiological data. More recently, there has been growing interest in transformer-based and multimodal approaches that combine imaging with clinical and biomechanical information. These integrated models show promise in improving predictive accuracy and supporting more individualised clinical decision-making.

Despite encouraging performance, many studies share similar methodological limitations. A large proportion rely on retrospective datasets collected from single institutions, which may limit the generalisability of findings to other clinical settings. External validation using independent datasets is still relatively uncommon, and prospective clinical evaluation remains limited. In addition, dataset imbalance, particularly underrepresentation of uncommon conditions or specific patient populations, may affect algorithm reliability and introduce unintended bias.

Another important challenge relates to interpretability. Although techniques such as heatmaps and feature attribution methods are increasingly used to provide insight into model decision-making, many AI systems still lack sufficient transparency. This can make clinical adoption more cautious, especially in high-risk decision-making environments. Variability in dataset quality, annotation standards, and evaluation methods also makes direct comparison between studies difficult.

Overall, while AI has demonstrated consistent potential to improve diagnostic accuracy, surgical planning, and outcome prediction in orthopaedics, further progress will depend on multicentre validation, improved dataset diversity, and development of more transparent and clinically interpretable models. Strengthening these aspects will be essential to support safe and reliable integration into routine orthopaedic practice. These methodological characteristics, validation practices, and common limitations observed across major orthopaedic subspecialties are summarised in Table 4.

Table 4 Summary of artificial intelligence methodologies, validation status, and limitations in orthopaedic applications.

Orthopaedic domain	Common AI models	Dataset source	External validation	Clinical readiness	Common limitations
Fracture detection	CNN	Radiographs from hospital databases	Limited	Early clinical use in radiology workflows	Mostly single-centre studies, limited prospective validation
Arthroplasty planning	CNN, machine learning models	Imaging data and arthroplasty registries	Moderate	Integrated with robotic and templating systems	Variability in implant systems and dataset heterogeneity
Spine surgery	CNN, machine learning models	Radiographs, CT, MRI, and clinical records	Limited	Early clinical adoption	Predominantly retrospective datasets
Sports medicine	CNN, deep learning models	MRI datasets	Limited	Early clinical use	Small datasets, variability in imaging protocols
Orthopaedic oncology	Radiomics and CNN	MRI and CT imaging datasets	Limited	Experimental stage	Small sample sizes due to rare tumours
Infection prediction	Machine learning models	Clinical and laboratory datasets	Limited	Early decision-support use	Inconsistent diagnostic criteria and dataset imbalance
Surgical robotics and navigation	Computer vision and AI-assisted navigation	Intraoperative imaging and sensor data	Moderate	Used in robotic arthroplasty and spine surgery	High cost and limited widespread availability
Rehabilitation monitoring	Machine learning and wearable-based AI	Wearable sensor and gait data	Limited	Emerging clinical use	Lack of standardisation and long-term validation

CNN: Convolutional neural networks; AI: Artificial intelligence; CT: Computed tomography; MRI: Magnetic resonance imaging.

CRITICAL APPRAISAL OF EVALUATION METRICS AND VALIDATION PRACTICES

Most AI studies in orthopaedics report performance using common classification metrics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve. These metrics are useful for measuring how well a model distinguishes between conditions, but they do not fully reflect clinical reliability. In particular, calibration, which indicates how closely predicted risks match actual outcomes, is not consistently evaluated. Only a limited number of studies report calibration methods such as calibration plots or Brier scores, which are important for assessing real-world applicability.

External validation is also relatively uncommon. Many studies rely on internal validation methods, such as train-test splits or cross-validation within the same dataset. While these approaches are useful during model development, they may overestimate performance because the testing data often shares similar characteristics with the training data. Validation using independent datasets from different institutions is essential to ensure that models perform reliably across diverse clinical settings, but this remains limited in the current literature.

Another important limitation is the inconsistent reporting of statistical uncertainty. Confidence intervals, which help indicate the reliability and precision of performance estimates, are not always provided. In addition, most studies are retrospective, and prospective clinical validation studies remain relatively rare. These findings highlight the need for more rigorous and standardised evaluation practices. Future research should prioritise external validation, calibration assessment, and clearer reporting of statistical uncertainty to ensure that AI models are reliable, generalisable, and clinically useful.

Quantitative trends in AI research in orthopaedics

A review of the included studies shows several consistent quantitative trends. Most studies use CNNs, particularly for imaging-based applications such as fracture detection and tumour classification. Transformer-based and multimodal models are increasingly reported, although they remain less common than conventional convolutional architectures.

External validation remains limited, with the majority of studies relying on internal validation methods using retrospective datasets. Only a smaller proportion of studies evaluate performance using independent datasets from different institutions. Similarly, while performance metrics such as accuracy and area under the receiver operating characteristic curve are widely reported, fewer studies provide calibration assessment or confidence intervals. These observations suggest that while AI research in orthopaedics is expanding rapidly, improvements in external validation, statistical reporting, and prospective evaluation are needed to strengthen clinical reliability and support broader implementation.

FUTURE DIRECTIONS

Future research should prioritise clinically reliable AI systems through improved uncertainty quantification, calibration-aware prediction models, and robust validation across diverse clinical environments. Digital twin models integrating anatomical, biomechanical, and clinical data show promise for improving surgical planning and outcome prediction; however, further work is needed to validate their accuracy, reliability, and clinical utility across diverse patient populations[145,179]. Multimodal AI systems combining imaging, clinical, and biomechanical data may improve risk prediction and patient selection, but future studies should emphasise external validation, calibration assessment, and prospective clinical evaluation to ensure reliability[139]. AI-assisted implant design and biomechanical modelling offer opportunities for personalised treatment; however, standardised benchmarking and comparative clinical studies are needed to establish their long-term effectiveness and safety[180]. Continuous-learning systems integrating wearable and clinical data may improve postoperative monitoring, but ensuring data quality, model stability, and regulatory compliance will be essential for safe implementation[9]. Future priorities should also include federated validation across multiple institutions, adaptation of foundation models to orthopaedic datasets, and development of standardised evaluation frameworks to improve generalisability and clinical adoption[137,139]. Emerging technologies expected to shape the future of AI in orthopaedic practice are summarised in Table 5.

Table 5 Emerging technologies shaping the future of artificial intelligence in orthopaedics.

Technology	Potential clinical impact
Digital twin modelling	Virtual surgical simulation
Multimodal AI systems	Comprehensive disease prediction
Smart implants	Real-time implant monitoring
Federated learning	Secure multi-institution collaboration
Wearable integration	Continuous rehabilitation tracking
Robotic-AI hybrids	Ultra-precise surgical execution
Generative AI	Automated surgical planning
Continuous-learning models	Adaptive real-time decision support

AI: Artificial intelligence.

CONCLUSION

AI is rapidly reshaping orthopaedic practice by enhancing diagnostic accuracy, surgical planning, operative precision, risk prediction, and personalised rehabilitation across multiple subspecialties. AI-based technologies are slowly helping bring more consistency to orthopaedic care and are helping doctors make decisions using available patient data. However, their regular use in daily practice is still limited. Problems such as differences in data quality, lack of testing across different patient groups, difficulty understanding how some AI systems reach their conclusions, unclear regulations, and differences in healthcare facilities continue to slow wider use. There are also important concerns about patient privacy, possible bias in AI results, and questions about who is responsible for decisions made with AI support.

In the future, orthopaedic care may increasingly use AI to help predict patient outcomes, design implants suited to individual patients, create digital models of joints and bones, and monitor recovery over time. AI is unlikely to replace doctors but can help improve accuracy and support clinical decisions. With careful development, proper regulation, and teamwork between doctors, engineers, and researchers, AI has the potential to improve patient care and support more personalised orthopaedic treatment.