Intersection of Artificial Intelligence (AI) and Regenerative Medicine in Musculoskeletal (MSK) Diseases: A Narrative Review

Abstract

Musculoskeletal (MSK) diseases present major health and economic challenges globally. Advancing age, diseases like osteoarthritis (OA), osteoporosis (OP), fracture and other conditions significantly reduce the quality of life (QOL) of these patients. Current pharmaceutical approaches are able to manage symptoms for some of these; however, they do not provide long-term solutions. Surgeries which are usually the final resort, present an added layer of challenges with the risk of post-surgical complications. The last couple of decades have observed an increase in the use of tissue engineering and regenerative medicine (TERM) for bone tissue engineering (BTE) applications. With the advent of artificial intelligence (AI), there will inevitably be an intersection of AI with TERM for MSK conditions. As of 2025, AI is already in use for small-scale applications in BTE including data extraction, image analysis, scaffold design and fabrication using three-dimensional (3D) printing techniques. This review outlines the convergence of these three fields and discusses the potential of their intersection. The author describes the need for this convergence, a brief update of TERM in MSK in the last decade, followed by the potential of AI in MSK-TERM. The review concludes on the challenges and future directions of the emerging field and hopes to encourage bold and ambitious collaborations between industry, academia, hospitals and health-care start-ups to realize the potential of this unique intersection.

1. Introduction

Musculoskeletal (MSK) diseases present significant health and economic challenges globally. According to the World Health Organization (WHO) 2022 factsheet, at least 1.7 billion people were estimated to be impacted by a single or multiple MSK diseases, making MSK diseases one of the largest contributors of disability worldwide [1]. MSK diseases like osteoarthritis (OA), osteoporosis (OP), rheumatoid arthritis (RA) fracture incidence and other groups of diseases significantly limit movement, ability to carry out daily activities and reduce the quality of life (QOL) of an individual. This eventually affects the individual’s ability to socialize, attend group gatherings and contributes to their isolation, ultimately impacting their mental health. Patients with chronic MSK pain have been reported to have significantly higher symptoms of anxiety, depression, fatigue, and insomnia than patients without it, in a 2022 study [2]. MSK health in general, along with its ability to repair is known to deteriorate with advancing age and progress to the commonly known age-related diseases (ARDs) including OA and OP [3]. These have been associated with ageing, accumulation of cellular damage (cellular senescence) and inflammation over time with ageing (inflammaging) [4,5].

Specifically considering the large burden of trauma-based injuries, fractures, OA and OP-associated fractures, approaches that accelerate the rate of tissue repair and regeneration are a necessity in MSK research and innovation. Tissue engineering and regenerative medicine (TERM) is a field that has grown exponentially in the last couple of decades and continues to do so, with the increasing need for innovation for tissue repair [6,7,8]. Particularly for fracture repair and bone regeneration, TERM needs to consider a combination of a suitable mechanical environment, osteogenic scaffolds, precursor or stem cells (usually mesenchymal stem/stromal cells or MSCs), growth factors, vasculature and host factors; also explained by the diamond concept [9,10]. The mechanical environment may be fabricated using different kinds of materials ranging from naturally existing polymers focused towards a ‘green medicine’ approach [11,12,13], to other materials with high biocompatibility and biodegradability, suitable for bone regeneration [14,15,16,17]. However, the choices for these combinations of materials have vast possibilities, and more often than not, the decisions comes down to feasibility and economic viability.

Artificial intelligence (AI) is increasingly being utilized in various applications including healthcare, clinical, pharmaceutical and biomedical research [18,19,20,21]. In orthopedics and clinical settings, AI is already in use for enhanced imaging and analysis for patient stratification and predicting better patient outcomes [22,23,24,25]. With respect to bone TERM, AI has the potential to be used as an additional aid to accelerate the process, every step of the way [26]. Right from the beginning that involves the production of scaffolds, predicting mechanical properties and cellular interactions, in silico modelling to precision medicine approaches tailored for patient care; AI can be programmed to provide us with the help necessary to fabricate suitable scaffolds for bone tissue engineering (BTE) (Figure 1) [27,28,29]. With the mechanical environment usually fabricated in three-dimensional (3D) form for TERM, integration of AI can strengthen prediction of models to better reflect the physiological conditions. This approach automatically reduces the need for animal models aligning research towards the 3R (replacement, reduction, refinement) principle and takes a step towards precision medicine (PM) using platforms like organ-on-chip (OOC), spheroids and patient-derived 3D organoids [30,31,32,33].

This review aims to discuss AI at the intersection of TERM in MSK conditions, providing a framework where these three very distinct areas converge. The article begins with an update in TERM for MSK diseases in the last decade followed by the potential of AI within MSK health and disease, focusing on TERM. Next several considerations are outlined followed by discussions around shared challenges in the field. Ultimately, with this review the author hopes to bring together the potential of inter-disciplinary research and future funding landscape in areas that would benefit from collaborations across expertise, in all healthcare settings including start-ups, academia, industry and medical institutions.

2. Methodology

To begin with, the keywords of ‘musculoskeletal’, ‘tissue engineering’, ‘regenerative medicine’ and ‘artificial intelligence’ were used as inputs for the literature search. Both PubMed and Google Scholar were used as platforms for the database search, and the literature was narrowed down to the last decade. Review articles, research articles, and editorials primarily focusing on MSK diseases were included. Publications primarily discussing diseases of other systems and discussing MSK diseases as secondary or tertiary focus were excluded. Publications that did not have full texts were also excluded from this manuscript preparation. For specific sections of cells, materials and strategies (Section 3.1, Section 3.2 and Section 3.3, respectively), these words were further added to the original keywords for the literature search, and similar inclusion/exclusion criteria were applied for finalizing the literature search.

The author acknowledges that, even with the aim of conducting a comprehensive literature search, the risk of bias cannot be completely ignored. It is possible that strategy utilized for this manuscript and the choices made using the keywords, inclusion/exclusion criteria and the interpretation of relevance that shaped this article, may have overlooked some of the studies in the field. Thus, this narrative review does not claim to be exhaustive but a convergence of the judgement used by the author with the resources available to them. During the preparation of this manuscript the author acknowledges the use of Copilot, Version 2601, for the purposes of creating two components of Figure 2 (central capsule panel and vascular panel on the top-right). The author reviewed and edited the output and takes full responsibility for the content of this publication.

3. Tissue Engineering and Regenerative Medicine (TERM) in MSK: The Last Decade

3.1. Cells

The cells being used for TERM in MSK and in BTE need to demonstrate compatibility and viability with the material(s) being used; they must be able to attach to these surfaces and maintain their functionality throughout the application stage of the scaffold. The most common type of cell that has been used for BTE is the bone marrow (BM) MSC followed by adipose tissue-derived MSCs (ADMSCs), which are multipotent cells and precursors of bone, fat and cartilage. These cells release growth factors and have ‘homing’ properties, meaning that they can migrate to injury sites based on chemo-tactical signals and can adhere to surfaces [34,35,36]. All of these properties make MSCs an ideal choice for regeneration of MSK tissues, especially the bone. MSCs can also be acquired and isolated from other sources including dental pulp (DPMSCs), periosteum, muscle, tendons and umbilical cord [37], and have been reported to demonstrate several properties including anti-inflammatory, anti-oxidant and anti-apoptotic properties, useful in diseases like inflammatory bowel disease (IBD) models [38].

While MSCs are the cells of primary choice for MSK TERM, combating cellular senescence in these cells is a challenge. Senescent MSCs and MSCs derived from older-aged donors often have lower proliferative, regenerative and differentiation capacities. These cells also tend to produce inflammatory cytokines and other harmful growth factors known as senescence-associated secretory phenotype or SASP [39,40,41]. Thus, strategies to combat MSC senescence including the use of seno-therapeutic drugs [42] and biologics like serum and platelet-rich plasma (PRP) are becoming critical for MSK and bone regeneration [43,44].

Embryonic stem cells (ESCs) are cells obtained from the inner mass of a blastocyst, prior to the implantation stage up to 3–4 days after fertilization, enabling these cells to be highly pluripotent and may be differentiated into any cell type in the body [45]. They are empowered with unlimited self-renewal capacity, making them a strong candidate for TERM [46,47]. However, ESCs have several challenges in relation to access and isolation of cells from embryos, potential tumour formation, managing differentiation and ethical, legal, as well as regulatory restrictions [48,49].

This challenge gave rise to another type of stem cell population known as induced pluripotent stem cells (iPSCs), which provided a promising strategy for TERM. These cells can be produced by de-differentiated adult cells using transcriptional factors [46,47]. Peripheral blood mononuclear cells (MNCs) have been shown to be induced into osteoblasts for bone regeneration on collagen scaffolds [50], into MSCs [51] and have potential applications for BTE [52,53]. Unfortunately, iPSCs also have their set of challenges. These include genetic abnormalities from reprogramming, tumour formation and inconsistencies in reprogramming efficiencies [54,55].

3.2. Materials, Fabrication and Techniques for TERM

A vast number of materials from different sources may be used to fabricate 3D scaffolds for TERM using various techniques. The materials may simply be classified as natural or synthetic and maybe used individually or in combination, to allow complementary properties for scaffold fabrication [56]. Natural materials involved those that can be directly sourced from nature (plants, animals and marine sources). Cellulose from plants [57,58], chitosan from crustacean and mushroom sources [13,59] and shell nacre cement (SNC) from marine sources [12,60] are some of the examples of materials from natural sources that have been used for BTE. Synthetic materials involve polymers like poly e-caprolactone (PCL), poly(lactic acid) (PLA), poly(glycolide) (PGA) and poly(lactic-co-glycolic acid) (PLGA) [56,61,62]. Metals like Titanium (Ti) and its alloys [63,64] and Strontium-based biomaterials (Sr-BMs) [65] are also used effectively in BTE applications. Another popular material has been the use of synthetic ceramic or beta-tricalcium phosphate (β-TCP) due to its osteo-conductivity and resemblance to cancellous bone [66,67]. Each of these materials have their own set of properties that provides them with their unique features and application in scaffolds for BTE, to be used individually or in combination with other materials.

Using any of these materials for fabrication of scaffolds must be optimized for several parameters to ensure appropriate size, location and intended application of the scaffold. These parameters include mechanical strength, dissolution rate, pore size or porosity, fiber diameter in fiber-based scaffold, bio-compatibility, cell adhesion and cell viability [68]. For example, β-TCP is known for its osteo-inductive properties but often needs to be used with other polymers like PCL/PLGA to enhance its mechanical strength, biocompatibility and resorption rate [69,70,71]. For porosity, pores within the compact and cancellous bones vary broadly in size depending on their location and their specific function. The average pore diameter within bones also depends on multiple factors, e.g., bone function, age, and patient condition. Patient variation is likely, yet it is generally observed that the range spans between 10 and 50 μm and 300–600 μm, for cortical and trabecular regions, respectively [72]. In vitro and in vivo examinations have also indicated that porosity has the ability to influence osteogenic potential, bone growth and osteo-integration [73,74], thus making it an important factor to optimize based on the application [75,76].

Another important consideration for BTE and TERM in general is the method used for fabrication of the scaffolds. A variety of methods are available including electrospinning for fiber-based scaffolds [77,78,79], freeze-drying [80,81,82], solvent-casting [83,84,85], hydrogel fabrication [86,87,88], and the more recently developed 3D printing techniques [89,90,91,92], amongst many others. The decision of the method to be used for fabrication will depend on the final desired scaffold design and application. Each of these methods have distinct advantages and disadvantages outlined below in

Table 1. Outline of examples of different methods of scaffold fabrication. 

Method	Advantages	Disadvantages	Decade Introduced
Freeze-Drying	Highly porous; good for biomolecules	Weak mechanical properties; long processing time	1970s
Solvent Casting and Particulate Leaching	Simple, low-cost; control over pore size	Limited mechanical strength; residual solvent risk	1980s
Sintering	High strength; good for ceramics	High temperature; brittle	1980s
Electrospinning	ECM-like nanofibers; high surface area	Limited pore size; difficult for thick constructs	1990s
Thermally Induced Phase Separation (TIPS)	Nanofibrous pores; ECM-like	Requires solvents; limited scalability	1990s
Gas Foaming	Solvent-free; interconnected pores	Poor pore control; low strength	1990s
3D Printing	High precision; patient-specific	Limited biomaterial printability; high cost	2000s
Self-Assembly	Biomimetic; bioactive	Slow; low scalability	2000s
Extrusion Bioprinting	Cell-laden constructs; multi-material	Low resolution; weak hydrogels	2010s
Melt Electrospinning Writing (MEW)	High-resolution fibres; good mechanics	Limited to thermoplastics	2010s

3D: Three-dimensional, ECM: extracellular matrix.

. Considerations for the method to be used enable functional integration by optimizing cell–material interaction, helps in determining vascularization potential, can influence scalability and thus the translational ability, and can aid in the integration of different technologies [93,94,95,96].

There has also been an increase in the use of novel and innovative materials for TERM that have not been traditionally used in the field. While not all of these materials have been optimized or utilized yet for BTE in particular, the materials/approach will potentially be applied for bone regeneration as well. These include origami-based methods [97,98], egg shells [99,100,101] and rattan wood [102,103] and wood from cane and pine trees [104,105]. The reason the wood-based approach has been gaining momentum in the last couple of years is due to the similarity that the cross-section of wood shares with that of bone. The wood is the skeleton of the tree, supporting its functions, and acts as storage for minerals such as calcium and phosphate, as well as in the maintenance of homeostasis, just like in bones [104]. Structurally, cellulose myofibrils (3–5 nm) and the tropocollagen (1.5–2 nm) from inside the bone can be traced down from the bark of wood and the outside of the bone, respectively. This comparison has been shown beautifully by Nefjodovs et al., recently [104]. The other reason for these unconventional approaches is a sustainable approach towards ‘green medicine’ in TERM [11,106,107].

3.3. Drug and Growth Factor Delivery, Antimicrobial Approaches and Anti-Ageing Strategies

Drug delivery in BTE is used for several functions, ranging from delivery of growth factors like bone morphogenic proteins 2 and 7 (BMP-2, BMP-7) to enhance bone formation and osteogenesis [108,109,110]; to the delivery of drugs like those for cancer treatment or antibiotics for antimicrobial activity for prevention of post-surgical infections [111,112]. Drug delivery in BTE may also be used to control the rate at which the growth factors and drugs are being released (Figure 2). Often drug molecules and growth factors are engineered at the nanoscale to optimize delivery rates and to ensure drug encapsulation within scaffolds [113], and may even use a dual drug delivery platform to deliver more than one active ingredient [114]. These included different nano-scale formulations including nanoparticles (NPs) [115,116,117], nanofibers (NFs) [118,119,120], graphene-based nano-sheets (NS) [121,122,123], carbon/quantum dots [124,125,126] and others [108,127,128] that may be incorporated within the scaffolds for various functions.

Controlled and sustained delivery of antibiotics and antimicrobials has increased in the past couple of decades to prevent and reduce the risk of infections within the bone, especially after surgeries, to combat post-surgical infections. Some of the examples include using chitosan and cerium oxide NPs [129], antimicrobial peptides [130], dual delivery mesoporous NPs [131], use of hybrid scaffolds for better control over the drug release rate [132] and stimuli (pH/temperature)-responsive scaffolds for antimicrobial drug delivery [133]. Due to the rising global risk of antimicrobial resistance (AMR) [134,135], more recently, several scientists are working towards antimicrobial materials with the same antimicrobial effect, but without the use of antibiotics. This is being done by the use of antimicrobial polymeric biomaterials, nano and micro-scale formulations, peptides, naturally occurring antimicrobial medicinal plants and metals with antimicrobial activity [136,137,138,139,140]. Investigations in this area will continue to grow due to the risk that AMR poses across the world.

Another development in the area of BTE has been the fabrication of scaffolds for anti-ageing strategies [4,141] that focus on materials or other approaches to reduce cellular senescence and bring healthy MSK ageing and longevity within reach. As outlined in Section 3.1, MSC senescence and ageing can be a source of several challenges in the BM [142], thus; using materials or approaches that can combat MSC ageing and senescence are becoming popular strategies. Marine-derived SNCs that are osteogenic indicated a high proliferation rate in MSCs from older donors, which was found to be comparable to that of MSCs from younger donors [12]. PRP has recently been indicated to have anti-ageing potential [43,143] and is already in use for bone regeneration in OA [144,145,146]. Thus, scaffolds fabricated in combination with PRP as one of their ingredients that are aimed at bone regeneration [147,148,149,150], may potentially hold anti-senescence, anti-inflammageing and anti-ageing properties too [151]. Research and innovation in this area have enormous potential, especially considering the aging population increasing across the globe [152].

4. AI, Its Subsets and Their Application in Bone Tissue Engineering (BTE)

TERM and BTE have several considerations as described in Section 3; thus, it is anticipated that BTE will need to strategize decision making at different stages from all the information available, for successful bone regeneration. These include decisions about scaffold design, biomaterial optimization, and predictive modelling of tissue regeneration, often clouded by biological complexity and considerations for translational success. AI can offer transformative potential by enabling data-driven insights across molecular, cellular, medical and biomechanical domains. Machine learning (ML) algorithms can accelerate biomaterial discovery, optimize scaffold architecture, and predict osteogenic outcomes with greater precision, enhancing the physiological relevance of the fabricated scaffold, improving patients’ QOL. Moreover, AI-integrated platforms and ML algorithms can aid clinical decision making with precision medicine strategies, bridging the gap between bench and bedside in regenerative orthopedics.

4.1. AI, ML and Deep Learning (DL): Definitions and Differences

ML algorithms can have various applications in the field of TERM. Before discussing these algorithms, we briefly explore below AI, ML and deep learning (DL); their definitions; how they are different from each other and broadly, their potential applications in BTE. AI refers to the field of computer science that utilizes human intelligence to create different systems. It may also be explained as the integration of human intelligence to machines for complex tasks. Rules are added to these computer systems that are often known as algorithms, and in AI, machines complete tasks based on the set of rules and algorithms fed to the machines. It is the umbrella term that encompasses ML and DL [153].

ML on the other hand is a subset of AI that exhibits experiential learning associated with human intelligence, and it includes approaches that allow machines to learn from data sets. ML aims to train machines for decision making, based on the data inputs. ML is known to be dynamic and thus adapts as and when new data is used as input for optimized decision making [153,154]. ML has also been used to detect and track senescent cells by imaging the nuclei of cells, which may be applied to the elimination of these cells for application on scaffolds for the best possible outcome for bone regeneration.

Finally, DL is further a subset of ML that includes computational models and algorithms that imitate the configuration of the biological neural networks in the brain, also referred to as artificial neural networks (ANNs) [153]. Just as the brain compares new information to existing knowledge, DL refers to making these ‘neuronal’ connections on multiple hierarchical data levels for interpretation of information and decision making. Overall, DL is a subset of ML, which in turn is a subset of AI indicated in the figure in Section 4.2.2 below [153,154].

4.2. ML and DL in TERM and BTE

4.2.1. Machine Learning Algorithms in BTE

Logistic Regression (LR) is a model that is commonly used for classification, based on probability principles and is known to achieve its optimal performance when working with data that can be separated linearly [29]. It is a simple and interpretable model but may overfit, i.e., learn to recognize specific patterns but may fail to generalize information, when faced with multi-modal data or different types of data sets, as it automatically assumes ‘linearity’ with any large data set [155]. This may be used in linear data sets like patient stratification or stem cell differentiation prediction for BTE applications.

Support vector machine (SVM) is applicable to classification, regression, and various tasks, known to construct hyperplanes in high-dimensional space in different categories [156]. SVMs, employing kernels like polynomial, linear, and radial basis functions, excel in high-dimensional spaces [29]. In simple words, SVMs learn to sort different elements into groups, by optimizing the best possible separation or segregation between them. When investigating methods to identify senescent cells from large data, SVM provided 99% accuracy, in comparison to 80% accuracy by other algorithms [157]. However, it faces challenges in data sets with disturbance/noise in the background, and its efficiency reduces with increasing classes. In the field of BTE, it may be employed for predictive biomarkers, stem cell subtype classification and genetic/proteomics analysis [155].

K-Means clustering is an unsupervised clustering algorithm aimed at grouping based on shared traits. It is efficient but sensitive to outliers and thus has limited biological interpretability [26,29]. It may be used for revealing stem cell heterogeneity in single-cell RNA sequencing done on bone 3D models for BTE. K-Nearest neighbours (K-NNs), also addressed as a non-generalizing or instance-based “lazy learning” algorithm, stores training data points within an n-dimensional space, and it categorizes new data instances by majority vote/most dominant class from the KNNs [29,156]. It is a simple, noise-resistant and intuitive algorithm, but it is highly data-dependent, and its performance drops with poor-quality data. For BTE, it may be used for classifying tissue imaging data and predicting stem cell differentiation outcomes [26].

Decision trees, random forests and gradient boosting (DT, RF and GB)-DTs are hierarchical structures classifying instances based on feature values, each node representing a classification feature. RFs enhance this by using multiple DTs for a robust ensemble output. GB machines, akin to RFs, combine weak classifiers, predominantly DTs, standing out as highly successful algorithms. Thus, DTs are interpretable, and RF/boosting improve the accuracy and feature selection and thus make decisions by dividing data into smaller parts [26,29,156]. However, the method may overfit, and ensemble methods are less interpretable using this method. For BTE, this may be utilized for clinical outcome prediction (e.g., graft rejection), ranking genes/proteins for tissue repair [26].

4.2.2. Deep Learning (DL) Algorithms in BTE

Reinforcement learning (RL) works on a reward or penalty-based system for complex data sets. Simply put, RL is all about choosing actions that maximize wins or rewards and minimize losses or penalties. This is based on the interaction between an agent and its environmental factors. RL enables independent assessment for optimal behaviour to enhance efficiency [26,29]. This is an adaptable algorithm that optimizes decision making from feedback and may be used for the application of medicine dosage automation, scaffold fabrication process optimization, bioreactor control and adaptive regenerative therapies for BTE [26,155].

Artificial neural networks (ANNs) and DL belong to a broader category of ML approaches, grounded in ANNs and are characterized by representation learning. It offers a computational framework that integrates various processing layers—input, output, and hidden layers—to acquire insights from data [29]. This is also one of the most widely used fields in tissue engineering as it is scalable and powerful for image and OMICS-related data. However, it works best with large data sets, and interpretability still remains a challenge. It is frequently used for automated histology, 3D scaffolds imaging, and complex relationships for biomedical data analysis in BTE [156]. Examples of the hierarchies of ML and DL in AI are indicated below in Figure 3.

Convolutional neural networks (CNNs) are models that are aimed at enhanced visual and image-based data analysis and are multi-layered. Extraction is performed by convolutional layers, pooling layers reduce the dimensions of the data and fully connected layers combine all this information and integrate them for data prediction [26]. Thus, CNN’s hierarchical approach makes it suitable for vision-based tasks like object detection, identification and recognition of visual cues. Paek and colleagues used AI-based DL algorithms to perform image analysis for the complex platform and used the CNN for classifying the images by groups and analyzing the information to ultimately predict drug efficacy [158]. A deep CNN (DCNN) was used to examine if analysis of computed tomography (CT) scanned images of the spine could be optimized to predict OP [159]. The authors concluded that, with robust construction and training of CNNs, it should be possible to potentially predict, diagnose and prevent a significant number of fractures due to OP.

Another application of neural networks in AI for images is the generative adversarial networks (GAN). GANs are DNNs that can generate or transform images, create images faster that are more realistic. MSK images generated by GANs are driving biomedical research and hold immense potential within MSK radiology. A GAN consists of two NNs: a generator and a discriminator. The generator tries to generate a high-dimensional image, which is fed into the discriminator. The discriminator then tries to distinguish between the generated and real images [160]. Specifically in terms of BTE, GANs may be used for 3D bone image synthesis, synthesis of scaffold designing and optimization (Figure 4) [26,161,162].

A simulation method that has been applied for the mechanical calculations of bones is Finite Element analysis (FEA), which is a physics-based simulation. While it is not a part of AI, ML or DL, it is a numerical technique for simulations and can be used to assess the deformation and mechanical properties of scaffolds underload in BTE [156]. In 1981, a very significant correlation was indicated between bone structure and mechanical stress with FEA. Soon after it was also observed that cells respond to mechanical cues based on the mechanical load [163]. Gryko et al. have also demonstrated the use of FEA to understand the influence of porosity and pore geometry on the mechanical properties of orthopedic scaffolds [164].

5. Challenges

While the intersection of these three fields, MSK, AI and TERM, has immense potential, there are several challenges that must be considered. With the vast variety of materials, fabrication methods and factors to be considered, information available on a specific material or method may appear fragmented, especially for use in AI for generating large data sets [26,165]. There will most likely be a lack of unified protocol as most laboratories and research group across the globe, perform these experiments on small ‘n’ numbers for different application sites. This will be a barrier towards data generalizability due to the heterogeneity in terms of the variations in data collected. Thus, application bias will be inevitable, and training for methods for efficient and relevant data extraction will be crucial [28]. Once this has been addressed, the next challenges to address will include training ML and DL algorithms against data bias, integrating systems for best data sharing practices, respecting privacy, consent and avoiding any data breach [165].

Specifically for some of the approaches discussed above, there are challenges and limitations that must be considered. For LR, it automatically assumes that all data set is linear, which may not always be the case. DT, RF and GB are prone to overfitting, and, thus, even if they learn patterns from the same data, they will struggle to adapt to new data sets. NNs are growing in their application; however, NNs need very large data sets and are often in need of substantial computational resources to provide results. RL on the other hand has limited potential when exposed to complex environments, and SVMs have overall limited ability to match up to large-scale data sets.

Considering the fact that this dynamic area of research will bring together different types of data sets ranging from multi-dimensional data sets to cell ingression videos, handling, filtering, collating and re-distributing information as needed by researchers will also be a challenge—also commonly addressed as the AI “black-box” challenge [155,166]. Due to the differences in sample sizes, materials and other factors, application of AI in BTE will face data scarcity for realizing the full potential of this intersection. This often also involves the lack of explanation for complex biological processes involved in BTE, even in the most sophisticated AI models, which leads to a lack of trust within the medical, ethical and regulatory communities [167]. Standardized protocols, transparent data-related rules and an adaptability-based approach to accommodate novel challenges must be recognised when considering regulatory and ethical aspects at the intersection of AI, MSK and TERM.

Finally, another challenge that needs addressing includes the barriers to translation of information to the human body and physiologically relevant environments [28]. This will potentially lead to translational bottlenecks, further widening the gap between the computational predictions and biological reality. AI and ML-based models often struggle to provide a picture that truly mimics the human body in vivo. Differences between patients, even for the same given diseases like immunity, inherent strength, ageing, genetic differences, and socio-economic factors, often get ignored by AI-based programmes [155]. For BTE, integrating patient-based, cellular, medical and TERM data in an effective and efficient way will require rigorous and collaborative training from experts from all associated arenas.

6. Conclusions and Future Directions

Despite the challenges, the intersection of AI and ML in BTE will reform how regenerative strategies are conceptualized and optimized. As multimodal data sets from various platforms like advanced imaging, biomechanical testing, 3D in vitro models and clinical outcomes continue to expand, AI-driven models will enable multiscale dissection of bone biology. These tools will be able to point towards complex and intricate relationships between molecular signalling, cellular behaviour, tissue-level mechanics, and patient outcomes, offering predictive insights with robust analytical approaches.

ML algorithms will potentially accelerate biomaterial choices and designs by predicting optimal biomaterial combinations for the fabrication of different scaffold architectures and applications that modulate osteogenesis, angiogenesis, and immune responses to better reflect human bone physiology. Generative design frameworks will support the creation of patient-specific scaffolds that integrate bone’s niche geometry, load-bearing aspects, and biological cues, enabling optimized constructs for complex and large bone defects. In parallel, AI-enhanced image analysis will transform real-time monitoring of cell dynamics, mineralization, and scaffold degradation in vitro and in vivo, and biocompatibility, creating quantitative feedback loops that refine experimental design and reduce reliance on animal models, upholding the ‘3R’ principle.

In conclusion, the convergence of AI, MSK and TERM offers a powerful route to model MSK ageing, diseases like OA, OP, and inflammatory microenvironments in physiologically relevant platforms. As clinical data sets continue to grow, AI-based decisions may guide precision medicine approaches by predicting scaffold performance, complication risks, and long-term functional outcomes. Realizing these opportunities will require rigorous attention to data quality, model interpretability, regulatory frameworks, and ethical considerations surrounding algorithmic bias, patient consent and privacy. Ultimately, the maturation of AI and ML in BTE promises a shift toward predictive, adaptive, and precision-engineered regenerative therapies, transforming fundamental research, applied sciences as well as clinical translation.