Global Perspectives on AI-Based Digital Twins in Smart Rehabilitation and Physiotherapy: Convergence of IoMT, Multiphysics Modeling, and Wireless Bio-Integrated Sensing

Abstract

Artificial intelligence (AI)-based digital twins (DTs) are emerging as a groundbreaking paradigm in rehabilitation and physiotherapy, enabling the creation of dynamic virtual representations of patients for continuous monitoring, prognostic assessment and personalised therapeutic interventions. This article presents a global, interdisciplinary review of AI-based DT technologies in rehabilitation settings utilising the Internet of Medical Things (IoMT), with particular emphasis on the integration of wearable and implantable sensor systems in next-generation wireless healthcare applications. The article analyses how multimodal wearable sensors, implantable devices and smart wireless communication networks can support the acquisition of real-time biomechanical and physiological data for adaptive rehabilitation. By combining perspectives from biomedical engineering, physiotherapy, computational intelligence and wireless healthcare systems, this article highlights the emerging opportunities and challenges associated with the creation of scalable digital twin ecosystems for precision rehabilitation. The proposed vision contributes to the development of smart, connected and personalized rehabilitation infrastructures, in line with future paradigms of healthcare and wireless communication.

1. Introduction

Artificial intelligence (AI)-based digital twins (DTs) are emerging as a breakthrough paradigm in rehabilitation and physiotherapy, enabling the creation of dynamic virtual representations of patients for continuous monitoring, predictive assessment, and personalized therapeutic intervention [1]. By integrating real-time physiological, biomechanical, and behavioral data, AI-based DTs provide an adaptive framework for optimizing treatment strategies throughout the rehabilitation process [2]. The convergence of the Internet of Medical Things (IoMT), wireless bio-integrated sensor technologies, and multiphysics computational modeling has significantly increased the accuracy and clinical utility of these intelligent digital replicas [3]. Advanced wearable and implantable biosensors continuously collect multimodal health data, enabling DTs to reflect patients’ changing physiological and functional conditions with unprecedented precision [4]. Simultaneously, AI algorithms, including machine learning (ML) and deep learning (DL) techniques, analyze these large-scale datasets to predict recovery pathways, detect early signs of functional decline, and support evidence-based clinical decision-making. Multiphysics modeling further enhances DT platforms by simulating complex interactions between musculoskeletal, neurological, cardiovascular, and biomechanical systems, thereby improving understanding of patient-specific rehabilitation responses (Figure 1) [5].

This integrated ecosystem facilitates personalized therapy planning, remote patient monitoring, and real-time optimization of rehabilitation protocols, while reducing healthcare costs and improving clinical outcomes [6]. Furthermore, AI-based DTs support the transition toward precision rehabilitation by enabling continuous feedback between patients, physicians, and intelligent healthcare systems (Figure 2) [7,8,9].

Despite these promising advances, significant challenges remain regarding data interoperability, cybersecurity, privacy protection, model validation, regulatory compliance, and the ethical implementation of AI-based healthcare technologies [10,11,12]. This article presents a comprehensive, global perspective on AI-based digital twins in smart rehabilitation and physiotherapy, highlighting recent technological advances, new clinical applications, current limitations, and future research directions at the intersection of AI, IoMT, multiphysics modeling, and wireless bio-integrated sensors.

Despite the rapid development of AI, DTs, and IoMT technologies, their integration into current and future rehabilitation and physiotherapy systems remains fragmented, with limited interdisciplinary frameworks combining computational intelligence, multiphysics modeling, and wireless, bio-integrated sensors within a unified ecosystem [13,14]. Existing research focuses primarily on isolated technological components rather than comprehensive AI-based DTs architectures capable of supporting continuous, adaptive, and personalized rehabilitation in diverse clinical settings [15]. Another significant research gap is the lack of globally accepted standards for interoperability, semantic data exchange, and DTs model validation, which limits the scalability and reproducibility of AI-assisted rehabilitation systems, including for global interoperability [16,17]. Furthermore, there is insufficient harmonization of the regulatory framework governing AI-assisted physiotherapy in hospitals, clinics, and home rehabilitation environments, creating barriers to widespread international implementation [18]. Current research also provides limited evidence for the seamless integration of heterogeneous multimodal data acquired from wearable sensors, implantable devices, electronic health records (EHRs), medical imaging, and biomechanical simulations into dynamic, patient-specific digital twins [19,20]. This paper introduces a new, global perspective by systematically converging AI, IoMT, multiphysics modeling, wireless bio-integrated sensors, and digital twin technologies into a unified conceptual and technological framework for smart rehabilitation and physiotherapy. The main contribution of this work is the identification of key requirements for standardized architectures, interoperable communication protocols, explainable AI (XAI) models, cybersecurity, privacy protection, and trustworthy clinical decision support that enable robust implementation in international healthcare systems. The proposed framework further expands current knowledge by considering the convergence of clinical rehabilitation facilities and AI-assisted home rehabilitation, emphasizing continuity of care through remote monitoring, adaptive therapy optimization, and real-time digital twin synchronization. Furthermore, the article discusses the challenges of globalization, including international data management, cross-border interoperability, ethical AI implementation, regulatory harmonization, and equitable access to smart rehabilitation technologies in countries with diverse healthcare infrastructures. Collectively, this work provides a comprehensive roadmap toward globally standardized, interoperable, safe, and clinically validated AI-based DT ecosystems that support precision rehabilitation, improve patient outcomes, and accelerate the digital transformation of future physiotherapy, both in the clinical setting and at home, including the surge in home AI development beyond 2030.

The aim of this review is to critically examine the global development and application of AI-based DTs in smart rehabilitation and physiotherapy by examining the convergence of IoMT, multiphysics modeling, wearable and wireless bio-integrated sensors, and intelligent data-driven approaches. The study aims to identify current technological advances, interdisciplinary integration strategies, implementation challenges, and future research directions for developing scalable, secure, and personalized digital rehabilitation ecosystems that support precision healthcare.

2. Materials and Methods

This article is a narrative review that synthesises evidence from publications to identify current challenges and limitations in AI-based DTs in Smart Rehabilitation and Physiotherapy. The proposed cloud-edge architecture has been deliberately designed as a direct response to the gaps revealed in this review, including issues relating to latency, scalability, interoperability, privacy and real-time clinical decision support. The article maps the main limitations identified in the literature to the corresponding architectural choices, demonstrating how the proposed platform addresses them. Thus, the results of the literature review justify the proposed framework, and this structure reinforces the logical flow of the manuscript, demonstrating that the architecture is evidence-based rather than an independent add-on.

2.1. Review Design, Search Strategy, and Study Selection

Bibliometric methods were employed to examine scientific publication databases and answer the following research question (RQs):

• RQ1: How are AI-based digital twins integrating with the Internet of Things (IoMT), wearable technologies, and wireless bio-integrated sensor systems to enable real-time intelligent monitoring and personalized rehabilitation in physiotherapy?
• RQ2: What role do multiphysics modeling, ML, and DL play in improving biomechanical analysis, digital biomarker extraction, predictive assessment, and adaptive therapy in rehabilitation systems using digital twins?
• RQ3: What are the main technical, ethical, cybersecurity, interoperability, and data management challenges impacting the implementation, scalability, and global adoption of AI-based digital twin ecosystems in smart rehabilitation?
• RQ4: What emerging research trends, technological opportunities, and future directions could accelerate the development of secure, understandable, scalable, and equitable AI-based digital twin platforms for precision rehabilitation and next-generation wireless healthcare?

The analysis identified several key aspects of the literature, including the current state of knowledge, the emergence and evolution of research themes, contemporary publication patterns (covering institutions, countries, and, where available, funding organizations), as well as the most influential authors and publications based on their scientific impact. This approach provided a comprehensive overview of current research trends and industrial developments in the field of digital transformation. Furthermore, the analysis and interpretation of bibliometric data support ongoing scientific and clinical discussions while establishing a robust foundation for future research and technological innovation in this domain.

2.2. Bibliometric and Thematic Analysis

For this study, four major bibliographic databases were analyzed: Web of Science (WoS), Scopus, PubMed, and dblp. These databases were selected due to their broad coverage and comprehensive citation data, which enable robust bibliometric analyses within the selected research domain (Figure 2). To ensure relevance, search filters were applied to restrict results to original research and review articles published in English. After this filtering stage, all retrieved records were manually screened according to predefined inclusion criteria, yielding the final dataset used for analysis. The review process involved three independent reviewers, who evaluated each article for inclusion or exclusion. Any discrepancies in assessment were resolved through discussion and consensus among at least two of the three reviewers. Subsequently, the dataset was examined to identify its key characteristics, including the most influential authors, research groups and institutions, countries of origin, thematic clusters, and emerging research trends. This analysis facilitated the mapping of the evolution of core concepts and the identification of major scientific contributions within the investigated field. Where applicable, temporal analyses were performed to explore the progression of research trends over time, and publications were classified into thematic clusters to reveal relationships and interconnections among different research areas. This methodological approach facilitated the identification of the principal research themes and subdisciplines characterizing the analyzed research domain.

The study selection process followed the PRISMA 2020 framework (ten core items) [21]. After duplicate records were removed, all retrieved publications were screened for relevance. Titles and abstracts were evaluated according to predefined inclusion and exclusion criteria, followed by a full-text assessment of studies deemed potentially eligible. Only publications that met all eligibility criteria were included in the final review and subsequent bibliometric analysis (Figure 3).

The study was conducted in accordance with selected components of the PRISMA 2020 guidelines [21] for bibliographic reviews, and the corresponding PRISMA 2020 Checklist is provided in the Supplementary Materials. The methodological framework incorporated the following PRISMA items: rationale (Item 3), objectives (Item 4), eligibility criteria (Item 5), information sources (Item 6), search strategy (Item 7), selection process (Item 8), data collection process (Item 9), synthesis methods (Item 13a), synthesis results (Item 20b), and discussion (Item 23a).

The bibliometric analysis was conducted using the analytical tools integrated into the Web of Science (WoS), Scopus, PubMed, and DBLP databases. The adopted review methodology enabled the systematic classification of publications according to research concepts (keywords), research areas, authors and research groups, institutional affiliations, countries of origin, document types, and publication sources.

In the Web of Science (WoS) database, searches were performed using the Subject field, which includes titles, abstracts, author keywords, and Keywords Plus. In Scopus, the search strategy targeted article titles, abstracts, and keywords, whereas in PubMed and DBLP, records were retrieved using manually defined keyword combinations. Across all databases, optimized keyword-based search strategies were employed to ensure comprehensive and relevant retrieval of publications: “artificial intelligence” AND “digital twin” AND (rehabilitation OR physiotherapy OR “physiotherapy”) (Figure 4).

The selected publication set underwent an additional verification stage involving manual re-screening, during which irrelevant records and duplicate entries were removed, resulting in the final sample size (Figure 5). To ensure the reliability and validity of the included studies, a combination of quality assessment methods was applied.

The PRISMA 2020 flow diagram presented above provides a transparent, step-by-step overview of the processes involved in identifying, screening, assessing eligibility, and ultimately including studies in the review, thereby ensuring methodological rigor and reproducibility. The process begins with the identification phase, during which records are retrieved from databases and other sources, capturing the full set of potentially relevant literature prior to duplicate removal. The subsequent screening phase involves evaluating titles and abstracts against predefined eligibility criteria, excluding studies not directly related to wearable devices, machine learning, or digital transformation of eHealth technologies. During the eligibility phase, the full texts of the remaining articles are assessed in detail with regard to methodological quality, adequacy of data collection and security, and relevance to topics. Finally, the inclusion phase reports the number of studies retained after the exclusion process, forming the evidence base used for subsequent synthesis and analysis. This interpretation of the PRISMA diagram enhances the transparency of the review process by not only illustrating the scope of the analyzed literature but also clearly outlining the rationale for exclusions at each stage. Consequently, it strengthens the credibility and validity of the study’s conclusions.

In this study, basic frequency analyses relating to authors, countries and institutions were deliberately employed to provide a concise, descriptive overview of the current state of research, rather than to conduct a comprehensive bibliometric analysis. The main objective was to summarize the scientific evidence regarding AI-based DTs in smart rehabilitation and physiotherapy, with bibliometric indicators serving merely as supplementary contextual information. Advanced analyses, such as keyword co-occurrence networks, citation burst detection and thematic clustering, require a dedicated bibliometric research project that goes beyond the scope and objectives of this review. Consequently, the approach adopted ensured that the focus remained on a qualitative synthesis of new technologies and their applications, rather than on a detailed mapping of the intellectual structure of this field.

3. Bibliometric Results

Table 1. Overview of results obtained from bibliometric analyses across Web of Science (WoS), Scopus, PubMed, and DBLP databases.

Parameter/Feature	Value
Dominant publication types	Review (41.20%), Article (24.70%), Conference paper (17.60%)
Dominant science areas	Computer Science (25.90%), Medicine (18.40%), Engineering (17.20%)
Dominant countries/territories	China (17), USA (12), India (9), Poland (8), Italy (7), Canada (4), Switzerland (4), UK (4)
Dominant scientists	Mikołajewska E. (6), Masiak J. (5), Mikołajewski D. (5), Chen J. (3), Panos E. (3), Yi C. (3)
Dominant affiliations	Nicolaus Copernicus University (6), Medical University of Lublin (5), Kazimierz Wielki University (5)
Dominant funders (where information available)	European Commission (6), National Natural Science Foundation of China (6)
Dominant SDGs	Good Health and Wellbeing (8), Industry Innovation and Infrastructure (7), Responsible Consumption and Production (6), Quality Education (1), Sustainable Cities and Communities (1), Partnership for the Goals (1)

and Figure 6, Figure 7, Figure 8, Figure 9, Figure 10, Figure 11 and Figure 12 provide a general overview of the bibliographic analysis results. The review included 97 articles published between 2020 and 2026. Studies published prior to 2020 were excluded, as the topic represents an emerging research area. Furthermore, the integration of artificial intelligence and advanced wireless technologies within the digital transformation of healthcare has only become widely established in recent years. Earlier studies may therefore reflect outdated methods or technologies that do not adequately represent current capabilities or research trends.

3.1. Publication Trends

Accordingly, restricting the review to the 2020–2026 period ensures a focus on the most relevant, innovative, and technically applicable approaches. The omission of literature published prior to 2020 from the narrative review, in accordance with part of the PRISMA 2020 guidelines, is justified, as this review focuses specifically on the latest developments in the field of AI-based digital twins used in smart rehabilitation and physiotherapy—a field that has undergone rapid development over the last few years. Earlier research focused primarily on the emerging trends in manufacturing rather than the convergence of AI, digital twins, the Internet of Medical Things (IoMT), multiphysics modelling and wireless sensors integrated into the body. Including these publications would divert the review from its main objective and introduce concepts that are only indirectly related to contemporary DTs models. Furthermore, fundamental discussions on early concepts of digital replicas in healthcare fall outside the defined scope of this focused review, which adheres to the PRISMA guidelines and emphasizes current AI-based implementations and global development trends. Limiting the review to studies published after 2020 therefore ensures methodological consistency, thematic relevance and a focused synthesis of the most significant technological advances in the field of intelligent rehabilitation systems.

In the databases WoS, Scopus, PubMed, and dblp, the most common publication type was review articles (41.2%), followed by research articles (24.7%) and conference proceedings (17.6%), indicating a strong emphasis on evidence synthesis and dissemination of scientific work. The dominant research areas were computer science (25.9%), medicine (18.4%), and engineering (17.2%), highlighting the interdisciplinary nature of the literature.

3.2. Geographic and Institutional Distribution

The leading contributing countries were China (17 publications), the United States (12), India (9), Poland (8), and Italy (7), reflecting the broad international participation. The most prolific authors were E. Mikołajewska (6 publications), J. Masiak (5), and D. Mikołajewski (5). The most productive institutions were Nicolaus Copernicus University (6 publications), Medical University of Lublin (5), and Casimir the Great University (5), while the leading funding organizations were the European Commission and the National Natural Science Foundation of China, each supporting six studies for which funding information was available.

3.3. Thematic Focus

The field is in its early stages of development and emphasizes developing intelligent healthcare systems, fostering technological innovation, and promoting sustainable, patient-centered rehabilitation through interdisciplinary and internationally funded research. Research is strongly interdisciplinary, with computer science leading the way, closely followed by medicine and engineering, reflecting the convergence of AI, digital technologies, biomedical engineering, and clinical rehabilitation. The quantitative dominance of the USA and China, with their authors absent from the top 5, shows a high degree of research dispersion (i.e., research clusters are still in the process of formation, and consolidation of research centers has not yet taken place).

In terms of relevance to the SDGs, the literature was most strongly aligned with the themes of Good Health and Wellbeing (8), Industry, Innovation, and Infrastructure (7), and Responsible Consumption and Production (6), and to a lesser extent with the themes of Quality Education, Sustainable Cities and Communities, and Partnerships for the Goals (one publication each).

4. Thematic Synthesis of AI-Based Rehabilitation DTs

4.1. IoMT and Bio-Integrated Sensing

The globalization of AI-based DTs for intelligent rehabilitation and physiotherapy reflects the rapid convergence of healthcare, advanced engineering, and digital communication technologies. Initially, research was local and discipline-specific, focusing separately on rehabilitation sciences, wearable sensors, or computational modeling without tight integration [1]. The second phase involved the expansion of interdisciplinary collaborations, where biomedical engineering, AI, and wireless communications began to converge into a unified research framework. At this point, early IoMT implementations enabled remote data collection using wearable and implantable devices [2]. The third phase is characterized by the emergence of integrated systems, in which AI-based DTs support continuous patient monitoring and personalized rehabilitation strategies. In this phase, multimodal sensors and multiphysics modeling combine electromagnetic, biomechanical, and physiological data to improve system accuracy and clinical relevance [3]. The fourth stage encompasses global scalability and platformization, where cloud infrastructures and wireless, bio-integrated sensor networks enable real-time rehabilitation regardless of geographic boundaries. At this level, DL and hybrid predictive models optimize therapy adaptation, movement analysis, and digital biomarker extraction [4,5]. The fifth stage highlights governance and standardization challenges, including interoperability, cybersecurity, ethical implementation of AI, and equitable access to digital healthcare technologies [6,7,8]. This development path demonstrates how rehabilitation technologies are evolving toward globally connected, intelligent ecosystems that integrate computational intelligence, wireless healthcare, and personalized medicine.

The pace and extent of globalization of AI-based DTs in rehabilitation were primarily shaped by technological maturity, regulatory frameworks, and the availability of interoperable healthcare infrastructure such as the Internet of Things (IoMT) [9]. Progress is also strongly dependent on the pace of development of AI, wireless communication networks, and multiphysics modeling, which determine how closely digital systems can replicate human physiological processes [10]. Economic factors such as healthcare financing, market incentives, and differences between high- and low-resource regions further influence the speed of global adoption of these technologies. Ethical, legal, and cybersecurity considerations can accelerate adoption through trust-building standards or slow it down due to necessary safeguards for patient safety and data privacy. Cultural acceptance among medical professionals and patients, as well as training opportunities in digital health literacy, also play a key role in determining the speed and depth of adoption. The globalization of these technologies is beneficial when it increases accessibility, personalization, and clinical outcomes, but it is not 100% positive, as premature or uneven implementation can exacerbate inequalities, introduce security risks, or create dependence on poorly regulated digital infrastructures [11,12,13].

Demographic factors play a key role in shaping the demand, design, and implementation of AI-based DTs in rehabilitation, influencing the prevalence of age-related conditions, disability rates, and rehabilitation needs across different populations. In particular, aging populations increase the need for scalable, personalized solutions such as IoMT-enabled DTs, while younger populations may prioritize preventative and sports-related applications [14,15,16,17]. At the same time, demographic differences in income, education, and access to healthcare technologies can significantly impact the equitable implementation and effectiveness of digital rehabilitation systems [18,19,20].

The future of IoMT in AI-based DTs for smart rehabilitation and physiotherapy will depend on seamless interoperability between wearable sensors, implantable devices, smart rehabilitation equipment, and cloud-edge computing platforms, enabling continuous acquisition of high-resolution physiological, biomechanical, and environmental data. These combined data streams will support dynamic DTs capable of continuously updating patient-specific models using AI and multiphysics simulations, thereby increasing the precision of diagnosis, treatment planning, and predicting rehabilitation outcomes. Advances in wireless bio-integrated sensor technologies, including flexible, skin-compatible, and energy-efficient sensors, will facilitate long-term, unobtrusive monitoring of movement patterns, muscle activity, cardiovascular function, and other clinically relevant biomarkers, both in clinical and home settings. Integrating IoMT with FL, explainable AI, and secure edge computing is expected to improve real-time decision-making while preserving patient privacy, reducing latency, and ensuring compliance with evolving healthcare data management regulations. Furthermore, the convergence of IoMT with multiphysics modeling will enable DTs to simulate complex interactions between musculoskeletal, neurological, and cardiovascular systems, allowing clinicians to evaluate personalized treatment strategies before implementing them in real patients. From a global perspective, these technological advances have the potential to expand access to high-quality rehabilitation services through remote monitoring, adaptive telerehabilitation, and intelligent clinical decision support, although their widespread implementation will depend on a unified framework for interoperability, cybersecurity, regulatory harmonization, and equal access to digital healthcare infrastructure.

4.2. AI, XAI, and Agentic Intelligence

Technological maturity is a key factor in determining the development and application of AI-based DTs in rehabilitation, as it defines the level of integration between sensors, modeling, and clinical decision support systems. It influences the IoMT’s ability to reliably collect high-quality, real-time physiological and biomechanical data from wearable and implantable devices [22]. As maturity increases, multimodal data streams can be combined more effectively, enabling more accurate simulations of patient-specific conditions using AI-based DTs [23]. This maturity also determines the effectiveness of multiphysics modeling in capturing the complex interactions between biological, mechanical, and electromagnetic processes [24]. Its role also extends to clinical translation, where more mature technologies support predictive analytics, personalized therapy design, and remote rehabilitation management [25]. However, technological maturity is uneven across regions and healthcare systems, leading to significant variation in implementation capabilities and outcomes. This diversity is reflected in differences in infrastructure readiness, computing resources, and regulatory environments [26]. Achieving interoperability between heterogeneous devices, platforms, and data standards to ensure seamless system integration is a key challenge. Additional challenges include protecting patient data privacy, ensuring cybersecurity, and maintaining algorithm reliability in clinical settings [27]. Limited or uneven technological maturity can limit scalability and equity, while higher maturity enables more effective, precise, and globally deployable rehabilitation solutions.

4.2.1. Emerging Role of XAI

XAI is becoming a key enabler for building trustworthy AI-based DTs in smart rehabilitation and physiotherapy, providing transparent and interpretable decision-making processes. XAI helps clinicians understand how AI models analyze multimodal data collected from IoMT-enabled wireless wearable and bio-integrated sensors, increasing confidence in personalized treatment recommendations [28]. In DT ecosystems, explainable models support the interpretation of biomechanical, physiological, and electromagnetic interactions generated using multiphysics modeling. By revealing the rationale behind movement assessment and digital biomarker extraction, XAI facilitates clinically relevant validation of AI predictions. Integrating XAI improves patient safety by enabling healthcare professionals to identify potential errors, biases, or uncertainties in adaptive rehabilitation algorithms [29]. It also strengthens AI’s ethical principles, increasing transparency, accountability, and fairness in AI-assisted rehabilitation systems across diverse patient populations. In remote physiotherapy applications, XAI increases user confidence by providing understandable explanations for therapy modifications and predictive health assessments. Furthermore, explainability supports regulatory compliance and interoperability by documenting AI decision-making pathways within connected healthcare infrastructures (IoMT) [30]. As AI-based DTs become increasingly autonomous, XAI will play a key role in balancing high predictive performance with clinical interpretability and human oversight. Therefore, the convergence of XAI, IoMT, multiphysics modeling, and wireless bio-integrated sensors is expected to accelerate the development of reliable, scalable, and globally accessible precision rehabilitation systems.

The most commonly used XAI techniques in patient digital twins include feature attribution methods such as SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations), as well as attention mechanisms, salience maps, and naturally interpretable ML models, which help clinicians understand how patient-specific data influence AI predictions [31,32]. These approaches are used to increase transparency, validate clinical decisions, identify relevant physiological and biomechanical factors, detect potential model biases, and support regulatory requirements for reliable AI in healthcare. In DT applications in rehabilitation, XAI enables clinicians to interpret predictions about movement quality, therapy progress, digital biomarkers, and personalized intervention strategies derived from multimodal IoMT sensor data. Future work is expected to focus on causal and counterfactual explainability, uncertainty-aware explanations, and interactive, human-centric XAI technology that provides clinically relevant feedback in real time within patients’ ever-evolving DTs. As basic AI models and multimodal digital twin platforms become more common, XAI will move beyond being a tool for post hoc explanations and become inherently explainable, adaptive, and continuously learning systems that seamlessly integrate transparency, security, fairness, and clinical decision support.

4.2.2. Rise of Agentic AI

AI agents are emerging in AI-based DTs in rehabilitation and physiotherapy as a shift from passive predictive models to autonomous, goal-oriented systems capable of planning and implementing therapeutic interventions. In the global perspective of intelligent rehabilitation, these agent-based systems operate in IoMT ecosystems, where multimodal, wearable, and bio-integrated sensors continuously transmit physiological and biomechanical data to digital replicas of patients in real time [33]. Architecturally, AI agents are embedded in layered frameworks combining edge computing for low-latency decision-making, cloud-based intelligence for large-scale learning, and DT simulation engines that maintain synchronized models of patient condition. Within this framework, agents can dynamically interpret movement patterns, detect anomalies, and autonomously adapt rehabilitation protocols based on feedback loops between the virtual and physical environments [34]. Integration with multiphysics modeling allows agents to reason about coupled electromagnetic, mechanical, and biological processes, improving both the accuracy of device interactions and the adaptation of therapy to individual patient needs [35]. In domain-specific applications, agent-based AI supports personalized physiotherapy by generating adaptive exercise programs, predicting recovery trajectories, and enabling remote supervision without constant physician intervention. These systems also enhance interoperability between heterogeneous medical devices by acting as coordination layers that harmonize data standards, communication protocols, and clinical decision logic. From a cybersecurity and ethics perspective, agent-based AI introduces new requirements for trustworthy autonomy, requiring explainability, secure data processing, and bias-aware learning in globally distributed healthcare infrastructures. Future directions point to fully autonomous rehabilitation ecosystems, where agent-based DTs collaborate with medical specialists as co-pilots, continuously optimizing therapy through reinforcement learning and federated intelligence. The convergence of agent-based AI, IoMT, multi-physics modeling, and wireless bio-integrated sensors is driving the transformation towards scalable, intelligent, and highly personalized rehabilitation systems (Figure 13).

The architecture presents a multi-layered AI-powered DT ecosystem in which multimodal data acquired from IoMT devices, wearable technologies, and wireless bio-integrated sensors is continuously transmitted to synchronized digital patient replicas via edge and cloud computing infrastructure, enabling real-time monitoring and low-latency communication. Autonomous AI agents constitute the intelligence layer of this system, continuously monitoring the patient’s condition, analyzing biomechanical and physiological information, predicting recovery paths, detecting anomalies, and coordinating adaptive rehabilitation strategies through closed-loop feedback between the physical patient and the virtual DT. DT integrates multiphysics modeling with data-driven AI to simulate coupled mechanical, biological, and electromagnetic processes, enabling AI agents to optimize therapeutic interventions based on the individual patient’s condition while maintaining high-fidelity virtual representations. The coordination layer enables interoperability between heterogeneous medical devices, communication protocols, clinical information systems, and distributed healthcare services, while leveraging explainable AI, secure data processing, and bias-aware learning to support reliable, autonomous decision-making. The information flow demonstrates how the convergence of autonomous AI agents, the Internet of Machines, multiphysics modeling, wireless, integrated biosensors, edge cloud intelligence, and continuous feedback enables the creation of scalable, personalized, and intelligent rehabilitation ecosystems that support healthcare professionals through adaptive therapy planning, remote monitoring, and continuous optimization of patient outcomes.

4.3. Multiphysics Modeling

Next-generation DTs in healthcare require multiphysics modelling, combining electromagnetic, biomechanical and physiological processes, to ensure accurate, patient-specific predictions. Such coupling is based on governing equations, including Maxwell’s equations for electromagnetic fields, continuum mechanics formulations for tissue deformation, and physiological transport and regulation equations describing biological functions. Solving these tightly coupled models in real time requires advanced numerical methods and high-performance computing due to their significant computational complexity. This highlights the role of specialised simulation environments and multiphysics software platforms, which enable co-simulation, data exchange and synchronisation across heterogeneous physical domains. This highlights the importance of cloud and edge computing as a means of distributing computational loads while meeting latency requirements in clinical decision support systems. This highlights the difficulty of the challenges associated with implementing real-time multi-physics coupling on digital healthcare platforms, which are only now being addressed.

Practical constraints associated with implementing federated learning directly on bio-integrated wearable sensor devices, used in AI-based digital DTs within the context of smart rehabilitation and physiotherapy, are of significant importance. Although FL enhances patient privacy by storing sensitive data on edge devices, the local training of deep neural networks is constrained by the limited battery capacity, memory, computational power and thermal budgets of wearable sensors. These resource constraints can increase training latency, energy consumption and communication overhead, making continuous model updates on devices impractical for many rehabilitation applications. To address this technical bottleneck, it is worth considering hybrid cloud-edge architectures, in which lightweight inference and preliminary data processing take place on wearable devices, while computationally intensive model optimisation is delegated to edge servers or cloud infrastructure. This provides a more balanced assessment of FL’s capabilities, highlighting both its privacy benefits and the engineering trade-offs that need to be taken into account.

Wireless, bio-integrated sensors are expected to play a key role in the evolution of AI-based DTs, enabling continuous, non-invasive, and precise monitoring of physiological, biomechanical, and biochemical parameters during rehabilitation and physiotherapy. New generations of flexible, stretchable, and skin-conforming sensors will provide real-time measurements of muscle activation, joint kinematics, tissue tension, cardiovascular activity, and metabolic biomarkers, creating comprehensive datasets for dynamic, patient-specific DT models. Integrating these sensor platforms with AI and multiphysics modeling will facilitate adaptive simulations that will continuously improve recovery predictions, optimize therapeutic interventions, and identify early indicators of complications or treatment failure. Future work is expected to focus on self-powered sensors, multimodal sensor architectures, and wireless edge communications, enabling continuous, long-term monitoring while minimizing patient burden and improving data reliability. From a global healthcare perspective, advances in wireless bio-integrated sensors will enable scalable, personalized, and remotely accessible rehabilitation ecosystems, but their successful implementation will require robust standards for interoperability, cybersecurity, clinical validation, and equitable access across healthcare settings.

4.4. Clinical and Home Rehabilitation

Clinically validated AI-based DT ecosystems should integrate multimodal patient data from wearable and bio-integrated sensors, medical imaging, electronic medical records, laboratory results, and biomechanical assessments into a continuously synchronized virtual patient model [36]. Such ecosystems require prospective, multi-center clinical validation using standard rehabilitation outcome measures, robust comparisons with conventional physiotherapy, and understandable AI algorithms to ensure transparency, reliability, and clinician trust [37]. Interoperability should be achieved through internationally recognized healthcare data standards and secure communication protocols, enabling seamless information exchange between hospitals, rehabilitation centers, home care systems, and telemedicine platforms [38]. The ecosystem should also include patient-specific multiphysics models that simulate musculoskeletal, neuromuscular, cardiovascular, and biomechanical responses to optimize rehabilitation strategies in real time [39].

AI-based DTs in smart rehabilitation and physiotherapy are emerging as advanced systems that combine real-time patient data, computational modeling, and predictive analytics to support personalized care [40,41]. In this context, multiphysics modeling refers to computational approaches that simulate coupled physical, biological, and electromagnetic processes in the human body and connected medical devices, enabling more realistic representations of tissue mechanics, neural activity, and device-tissue interactions [42]. Wireless bio-integrated sensors encompass implantable or wearable sensor systems that are embedded in or tightly coupled to biological tissues and capable of continuously transmitting physiological signals such as motion, pressure, temperature, or biochemical markers. Integrating these technologies with the IoMT enables continuous, distributed data acquisition from heterogeneous sources in clinical and home environments [43,44,45,46]. AI-based DTs can then assimilate these multimodal data streams to create personalized virtual models that reflect the patient’s physiological state in real time. This combination allows clinicians to simulate rehabilitation scenarios, predict treatment outcomes, and optimize therapy parameters before implementing them in practice [47,48,49,50,51]. Multiphysics modeling increases the fidelity of DTs by capturing complex interactions between biomechanics, electrophysiology, and external therapeutic devices such as stimulation systems. Wireless, bio-integrated sensors extend clinical utility by enabling minimally invasive, continuous monitoring that improves data accuracy and temporal resolution. Together, these technologies support early detection of functional decline, adaptive rehabilitation planning, and remote patient management with reduced clinical burden [52,53,54,55]. Their convergence is expected to enable the creation of fully autonomous, closed-loop rehabilitation systems, although challenges remain related to computational costs, data standardization, and clinical validation [56,57,58,59].

AI-based digital health technologies for smart rehabilitation can be effectively integrated into smart home and smart terrain infrastructures, leveraging continuous data streams from the IoMT, enabling real-time patient monitoring in everyday environments [60]. In smart homes, wireless, bio-integrated sensor systems can discreetly track physiological and biomechanical parameters, enabling automatic adjustments to personalized rehabilitation programs based on user behavior and health status [61,62,63]. AI-based digital health systems can process this data, simulating patient conditions and recommending adaptive interventions without the need for constant clinical supervision [64,65,66]. At the broader level of smart terrain, such systems can connect households, healthcare providers, and community services into a unified digital health ecosystem [67,68,69,70]. This enables coordinated care delivery, early detection of health deterioration trends across the population, and more efficient allocation of medical resources. Integrating these technologies transforms rehabilitation into a distributed, proactive, and context-aware service, embedded in everyday life [71,72].

4.5. Interoperability, Regulation, and Governance

Regulatory frameworks play a key role in shaping the development, implementation, and global uptake of AI-based DTs in rehabilitation by defining security, privacy, and accountability requirements. They define how IoMT data is collected, transmitted, and stored in compliance with patient protection regulations [73]. These frameworks influence the validation and certification processes for AI-based DTs, ensuring clinically safe and reliable predictive models used in rehabilitation. They also set standards for the approval of medical devices, particularly wearable and implantable technologies used for remote monitoring and therapy [74]. A key role of regulation is to balance innovation with patient safety, preventing the premature implementation of unverified technologies in clinical environments.

Regulatory variation across countries leads to varying approval timelines, legal requirements, and technology adoption rates. This variation can slow global scale but also allows for adaptation to local healthcare priorities and ethical norms. Among the key challenges is harmonizing international standards to support cross-border interoperability and data exchange [75]. Another challenge is regulating rapidly evolving AI systems, whose adaptive behavior may not fit traditional certification models [76]. Strong and well-coordinated regulatory frameworks can foster trust and security, but excessively fragmented or restrictive systems can stifle innovation and slow the global impact of digital rehabilitation technologies [77].

The availability of interoperable healthcare infrastructure, such as the IoMT, is a fundamental enabler for the implementation of AI-based DTs in rehabilitation and physiotherapy. It defines the ability of various medical devices, wearable sensors, and implantable systems to seamlessly communicate and share standardized data in real time [78]. A high level of interoperability enables continuous physiological and biomechanical monitoring, which is essential for accurate modeling and personalized therapy [79]. Such an infrastructure supports the integration of multimodal data streams, including motion tracking, biosignals, and environmental context [80]. It also increases scalability by enabling healthcare systems to connect across hospitals, home care facilities, and remote rehabilitation platforms [81]. However, interoperability varies significantly across regions due to differences in technological development, funding, and standards implementation [82]. This variability can lead to fragmented ecosystems where devices and platforms cannot fully communicate or exchange data [83]. Key challenges include a lack of universal data standards, incompatible communication protocols, and limited integration between legacy and modern systems [84]. Additional challenges include ensuring secure data exchange while maintaining patient privacy and regulatory compliance [85]. The availability of interoperable infrastructure directly determines the effectiveness, scalability, and global reach of digital rehabilitation technologies.

Interoperability, regulation, and governance are fundamental pillars for the successful implementation of AI-based DT ecosystems in smart rehabilitation and physiotherapy. Interoperability enables seamless integration and secure data exchange between wearable sensors, IoMT devices, electronic health records, medical imaging systems, and clinical decision support platforms, in accordance with internationally recognized standards such as HL7, FHIR, DICOM, and IEEE 11073. Standardized data models and communication protocols ensure that heterogeneous healthcare systems can exchange consistent, semantically relevant information while maintaining interoperability between institutions and providers. Regulatory compliance is equally crucial, as AI-based DT platforms must comply with medical device regulations, software validation requirements, cybersecurity standards, and data protection laws before they can be safely implemented in clinical practice. Governance frameworks should clearly define responsibilities for data ownership, access control, accountability, algorithm validation, model maintenance, and clinical oversight throughout the digital twin lifecycle. Strong cybersecurity mechanisms, including encryption, authentication, secure communication, continuous vulnerability monitoring, and incident response strategies, are essential to protecting sensitive patient information and maintaining trust in distributed healthcare infrastructures. Ethical governance should ensure transparency, understandability, fairness, and the reduction of bias in AI models, while maintaining patient autonomy through informed consent and appropriate control over personal health data. FL and privacy-preserving AI techniques further strengthen governance by enabling collaborative model development across healthcare settings without the need to exchange raw patient data, improving both security and regulatory compliance. Continuous post-implementation monitoring, independent auditing, and periodic model recalibration are essential to identify model drift, maintain clinical accuracy, and demonstrate ongoing compliance with evolving regulatory requirements and international standards. Together, robust interoperability, a comprehensive regulatory framework, and transparent governance provide the foundation for trustworthy, scalable, and globally deployable ecosystems capable of supporting safe, personalized, and data-driven rehabilitation and physiotherapy.

5. Proposed GIR-DT Framework and Validation Roadmap

We propose a global intelligent rehabilitation digital twin (GIR-DT) platform—a hierarchical architecture that combines edge AI deployed on mobile devices with cloud-based digital twins synchronized via the IoT infrastructure, enabling continuous monitoring in both clinical and home environments. In the proposed platform, federated learning (FL) enables collaborative training of AI models across healthcare settings without transmitting confidential patient data. This improves privacy, cybersecurity, and international scalability, reduces energy consumption within the Green AI paradigm, and simultaneously supports compliance with regional data protection regulations. A digital rehabilitation coordination layer continuously integrates AI predictions, multiphysics simulations, clinicians’ clinical expertise, and patient-reported outcomes to dynamically personalize exercise intensity, therapy progression, and risk management throughout the rehabilitation process. This proposed ecosystem provides a scalable path to globally standardized, clinically validated, safe, and interoperable AI-based rehabilitation, supporting precision physiotherapy, continuity of care, improved functional recovery, and evidence-based decision-making across healthcare systems (Figure 14).

The novel aspects of the proposed GIR-DT framework arise from the fact that the seven-layer architecture extends existing DT models by introducing several new components:

• Hierarchical edge and cloud intelligence, enabling low-latency AI inference on wearable devices while maintaining computationally intensive DTs in the cloud.
• A dedicated data integration and standardization layer, enabling interoperability through standards such as HL7, Fast Healthcare Interoperability Resources (FHIR) and Digital Imaging and Communications in Medicine (DICOM), rather than assuming compatibility.
• FL in the AI layer, enabling institutions to collaboratively improve models without exchanging raw patient data.
• A rehabilitation coordination layer, which combines AI predictions with multiphysics simulations to dynamically adapt rehabilitation protocols instead of relying on rigid treatment plans.
• Support for hybrid clinical and home rehabilitation, enabling seamless transitions between inpatient care and remote physiotherapy.
• A management layer, emphasizing continuous clinical validation, cybersecurity, regulatory compliance, ethical AI, and international standardization, makes the platform suitable for implementation across diverse healthcare systems.

This framework presents a seven-layer architecture in which multimodal physiological and biomechanical data acquired from wearable, wireless, bio-integrated, and IoMT sensors are processed using hierarchical edge and cloud intelligence, enabling low-latency AI inference at the edge while simultaneously supporting computationally intensive DT simulations in the cloud. A dedicated data integration and standardization layer harmonizes heterogeneous information, leveraging healthcare interoperability standards such as HL7, FHIR, and DICOM, enabling the secure and seamless exchange of clinical, imaging, and sensor data across healthcare platforms. The AI and DT layers incorporate federated learning, multiphysics modeling, and predictive analytics to collaboratively improve machine learning models without sharing raw patient data, while dynamically adapting rehabilitation protocols based on continuously updated, patient-specific DT simulations rather than rigid treatment plans. The rehabilitation coordination layer integrates AI predictions with simulation results to support personalized treatment planning in both the clinical and home environments, ensuring continuity of care and real-time adaptation of treatment strategies. A senior management layer manages the entire ecosystem, ensuring continuous clinical validation, cybersecurity, ethical AI, regulatory compliance, and international standardization. This enables reliable clinical decision-making, scalable implementation, and interoperability across heterogeneous healthcare systems worldwide.

In the proposed GIR-DT platform, physiotherapists use a dashboard as an interface to support clinical decision-making, which visualizes synchronized patient DTs, real-time data from wearable sensors, AI-generated risk assessments, progress in rehabilitation, and personalized treatment recommendations. The dashboard does not replace clinical assessment, but integrates AI predictions, multi-physics simulation results and patient-reported outcomes to support evidence-based decisions regarding exercise progression, treatment modifications and risk management in both clinical and home settings. Effective use of the platform requires targeted training in navigating the dashboard, interpreting AI-generated findings, understanding confidence metrics, and appropriately implementing recommendations into routine clinical practice, without requiring specialist knowledge of AI modelling or computational methods. By automatically aggregating and analyzing multimodal data, the GIR-DT platform reduces the time spent on manual data verification and documentation, enabling clinicians to focus more on patient interaction and personalized care. Continuous remote monitoring also enables physiotherapists to identify clinically significant changes between scheduled appointments, facilitating timely interventions while reducing the need for unnecessary face-to-face follow-up appointments. The platform has been designed to streamline clinical processes, ensure consistency in decision-making across all healthcare facilities, and improve continuity of care without significantly increasing the workload of healthcare staff.

A scientifically and clinically rigorous validation protocol should evaluate the entire GIR-DT ecosystem, not just the AI model itself. The protocol should verify sensor reliability, data communication, DT fidelity, AI prediction accuracy, clinical effectiveness, cybersecurity, and interoperability in both clinical and home environments (Figure 15).

The proposed Global Intelligent Rehabilitation DT Validation Protocol (GIR-DT-VP) is a seven-step framework designed to systematically evaluate AI-based DT ecosystems across technical, computational, clinical, and regulatory components.

• Step 1: Device and Sensor Validation. All wearable and bio-integrated sensors should be calibrated against certified medical reference devices. Signal quality, accuracy, repeatability, latency, power consumption, and robustness should be assessed in laboratory and field settings in rehabilitation settings.
• Step 2: Data Communication and Interoperability Validation. IoMT should be assessed for communication security, packet loss, timing accuracy, transmission latency, scalability, and compliance with international interoperability standards such as HL7 FHIR, DICOM, IEEE 11073, and ISO/IEEE communication protocols.
• Step 3: DT Fidelity Validation. The DT should be quantitatively compared to the patient’s actual physiological and biomechanical state using multimodal clinical data. Validation metrics should include, at a minimum, geometric similarity, biomechanical simulation error, temporal synchronization, physiological prediction error, and model update rate.
• Step 4: AI Validation. ML and DL models should be evaluated using external, multi-center datasets with prospective validation. Performance should be assessed in terms of accuracy, sensitivity, specificity, precision, F1 score, ROC-AUC, calibration, uncertainty estimation, reliability analysis, explainability, and robustness to noisy or incomplete data.
• Step 5: Clinical Rehabilitation/Physiotherapy Validation. The entire GIR-DT ecosystem should be validated in randomized, controlled clinical trials comparing AI-assisted rehabilitation with conventional physiotherapy. Primary outcomes should include functional recovery, pain reduction, quality of life, adherence, rehabilitation time, clinician workload, and patient satisfaction.
• Step 6: Home Rehabilitation Validation. The framework should be evaluated in a real-world home rehabilitation setting by assessing continuous monitoring performance, communication reliability, patient adherence, usability, remote intervention effectiveness, digital literacy requirements, and long-term system stability.
• Step 7: Global Governance and Continuous Learning Validation. Continuous post-implementation monitoring should assess cybersecurity, privacy protection, regulatory compliance, federated learning performance, model drift, international interoperability, and continuous improvement of clinical performance through periodic AI model updates and independent external auditing.

The validation framework goes beyond conventional ML metrics, explicitly taking into account the spatiotemporal characteristics of patients’ constantly evolving DTs. There is a need to assess temporal consistency, the accuracy of long-term predictions, adaptation to streaming multimodal data, and synchronization between the patient’s virtual and physical states over time. Validation should assess the robustness of dynamic model updates, real-time data assimilation, and the stability of personalized predictions under changing physiological conditions.

The validation process begins with the calibration and assessment of wearable and wireless bio-integrated sensors, followed by verification of secure IoMT communication, interoperability, and compliance with international healthcare standards, including HL7 FHIR, DICOM, IEEE 11073, and ISO/IEEE communication protocols. The framework subsequently validates DT fidelity through quantitative comparisons between virtual and real patient states, while AI models are rigorously assessed using multicenter external datasets based on predictive performance, explainability, robustness, uncertainty estimation, and resilience to noisy or incomplete clinical data. Clinical effectiveness is then evaluated through randomized controlled trials and real-world home rehabilitation studies, measuring functional recovery, pain reduction, quality of life, adherence, usability, remote monitoring reliability, clinician workload, and long-term operational stability. The final validation stage establishes a continuous governance framework that monitors cybersecurity, privacy protection, regulatory compliance, federated learning performance, model drift, international interoperability, and periodic AI model updates through independent auditing, ensuring trustworthy, scalable, and continuously improving DT ecosystems for global rehabilitation practice.

The proposed GIR-DT protocol introduces a hierarchical validation strategy in which each layer of the ecosystem is independently validated before conducting an integrated system validation. This modular methodology facilitates repeatability, regulatory approval, international standardization, and large-scale implementation, while ensuring that AI-based DTs remain clinically credible, technically robust, interoperable, safe, and adaptable in both hospital and home settings (

Table 2. Proposed validation matrix.

GIR-DT Layer	Validation Objective	Representative Metrics
Layer 1 Bio-integrated sensing	Sensor accuracy and reliability	Signal-to-noise ratio, RMSE, drift, battery life, sampling stability
Layer 2 IoMT communication	Reliable data transfer	Latency, jitter, packet loss, throughput, synchronization error
Layer 3 Data integration	Interoperability	FHIR conformance, semantic consistency, data completeness
Layer 4 AI and DT	Predictive performance	Accuracy, F1-score, ROC-AUC, calibration error, explainability, uncertainty
Layer 5 Multiphysics models	Model fidelity	Simulation error, biomechanical agreement, computational time
Layer 6 Clinical workflow	Clinical benefit(s)	Functional scales, recovery time, adherence, clinician acceptance
Layer 7 Governance	Safety and sustainability	Cybersecurity, privacy compliance, model drift, auditability, regulatory conformity

).

A significant contribution of this protocol is the layer-specific validation strategy. Instead of treating DT as a single black box, each layer of the architecture is independently verified and then validated as part of an integrated ecosystem. This approach promotes reproducibility, simplifies regulatory assessment, and enables progressive certification as individual components evolve. It also provides a framework that can be adapted to international standardization efforts, thus serving as the basis for future clinical guidelines and consensus recommendations for AI-based rehabilitation systems.

Performance thresholds have been introduced for each stage of validation. With regard to the performance of the artificial intelligence model, the following criteria are considered acceptable: accuracy ≥ 90%, F1 score ≥ 0.90, area under the ROC curve (AUC) ≥ 0.95, and sensitivity and specificity ≥ 90% for the classification of clinically relevant features. For continuous physiological and biomechanical predictions, the target error is defined as a root mean square error (RMSE) of less than 5% of the measurement range or a normalised root mean square error (NRMSE) of less than 10%, along with a mean absolute error (MAE) below clinically acceptable limits for the parameter in question. Validation of digital twin synchronization requires an end-to-end data latency of less than 100 ms for real-time rehabilitation applications, a synchronization accuracy of over 95 per cent, and data completeness exceeding 99 per cent. The multi-physics simulation component is validated by achieving agreement with experimental or clinical reference measurements within ±5%, while maintaining numerical stability and reproducibility in repeated simulations. Clinical validation is assessed on the basis of a statistically significant improvement in functional outcome measures, high utility scores assigned by both clinicians and patients (system utility score ≥ 80) and robust performance in external, multicentre datasets, with the aim of ensuring the generalisability of the results and reliable implementation in smart rehabilitation environments.

6. Challenges and Future Directions

Compared to conventional DT architectures in healthcare, which primarily integrate electronic medical records with static physiological models, the proposed AI-based DT for smart rehabilitation and physiotherapy introduces a continuously adaptive, patient-specific framework based on real-time, multimodal measurement and predictive intelligence. Existing DT frameworks based on IoMT connectivity primarily focus on remote patient monitoring and data aggregation, whereas the proposed architecture tightly couples IoMT streams with wireless, bio-integrated sensors to generate high-quality biomechanical and physiological representations during rehabilitation. Unlike biomechanical DT approaches, which primarily rely on musculoskeletal simulations, the proposed system utilizes multiphysics modeling to simultaneously capture biomechanical, electrophysiological, and tissue-level interactions, providing a more comprehensive picture of patient recovery. Unlike cloud-based DT architectures that implement centralized analytics with limited responsiveness, the proposed framework integrates edge AI with intelligent wireless sensors, enabling low-latency decision support, continuous model updates, and personalized therapeutic interventions. Existing AI-based DT rehabilitation systems often focus on isolated predictive tasks such as gait classification or outcome prediction, whereas the proposed architecture leverages explanatory and adaptive AI to continuously optimize rehabilitation protocols based on the patient’s dynamic responses. Furthermore, unlike conventional wearable-based rehabilitation systems that rely on rigid sensors with limited physiological range, the proposed DT utilizes wireless bio-integrated sensor technologies, enabling discrete, long-term acquisition of multimodal biomechanical and biosignal data. The convergence of IoMT, multiphysics modeling, a wireless bio-integrated sensor, and explainable AI creates a unified cyber–physical ecosystem that goes beyond the capabilities of existing DT architectures, enabling synchronized monitoring, simulation, prediction, and personalized intervention within a single intelligent framework. As a result, the proposed AI-based DT architecture represents a significant advancement over existing digital healthcare development models, offering higher levels of personalization, real-time adaptability, predictive accuracy, and intelligent closed-loop rehabilitation support for next-generation intelligent physiotherapy systems.

In response to RQ1, AI-based digital twins integrate with IoMT by continuously collecting physiological, biomechanical, and environmental data from wearable and bio-integrated sensors. These real-time data streams are transmitted via wireless communication networks to virtual patient models that dynamically reflect the individual’s health status and rehabilitation progress [86]. AI analyzes data from multimodal sensors to detect movement patterns, assess functional performance, and predict recovery paths [87]. The digital twin continuously updates treatment recommendations, adapting rehabilitation exercises to individual patient responses and clinical goals [88,89]. This integration enables personalized, remote, and data-driven rehabilitation while improving clinical decision-making, patient engagement, and access to healthcare [89].

In answer to RQ2, multiphysics modeling enables DTs to simulate the interconnected biomechanical, physiological, and electromagnetic processes that influence the performance of rehabilitation devices and patient outcomes. ML algorithms extract clinically relevant digital biomarkers from complex multimodal datasets, facilitating objective assessment of patient health and recovery. DL techniques improve motion recognition, posture assessment, gait analysis, and anomaly detection by automatically learning hierarchical representations from sensor signals. Predictive models estimate rehabilitation progress, identify potential complications, and support early intervention through individualized risk assessment. Together, these computational approaches create adaptive rehabilitation systems capable of continuously optimizing treatment strategies based on real-time patient data [90,91,92].

To answer RQ3, the implementation of AI-based DTs is limited by the heterogeneity of medical devices, inconsistent communication protocols, and limited interoperability between health information systems. Protecting sensitive patient information requires robust cybersecurity mechanisms, privacy-preserving data processing, secure authentication, and compliance with healthcare regulations. Ethical challenges include ensuring algorithm transparency, minimizing bias, preserving patient autonomy, and establishing accountability for AI-supported clinical decisions. Managing high volumes of continuous, multimodal data demands, scalable storage, efficient processing, standardized data formats, and high-quality data management. Removing these technical, ethical, and organizational barriers is essential to creating reliable, trustworthy, and globally accessible digital rehabilitation ecosystems [93,94].

To answer question 4, new research focuses on XAI, FL, edge intelligence, and fundamental models, which aim to improve the transparency, privacy, and performance of DTs. Advances in flexible wearable electronics, implantable bio-integrated sensors, and next-generation wireless communication technologies will enable more accurate and continuous patient monitoring. Integration of multimodal digital biomarkers, cloud-based collaboration, and personalized predictive analytics are expected to improve the accuracy of rehabilitation and long-term patient management. Future digital twin platforms will increasingly support interoperable healthcare infrastructure through standardized architectures, semantic data models, and secure information exchange across clinical environments. Achieving equitable global adoption will require affordable technologies, inclusive AI models, international standards, multidisciplinary collaboration, and policies that promote accessibility, ethical governance, and sustainable innovation in healthcare [95,96].

The results support the working hypothesis that integrating AI, IoMT, multiphysics modeling, and wireless bio-integrated sensors into DTs can significantly improve the precision, adaptability, and scalability of rehabilitation services. These results are consistent with previous studies demonstrating that AI-based DTs enhance continuous patient monitoring, predictive analytics, and personalized therapeutic interventions, while also extending these concepts by incorporating FL, edge AI, and a coordinated DT architecture in the cloud. The proposed GIR-DT platform extends previous digital rehabilitation frameworks by introducing a hierarchical ecosystem that integrates clinician expertise, patient-reported outcomes, and physics-based computational models into a unified decision support environment. The results further support previous research highlighting the importance of interoperability, cybersecurity, and ethical governance of AI as essential prerequisites for the safe implementation of intelligent healthcare systems operating across institutional and national boundaries. The inclusion of FL supports the hypothesis that decentralized AI training can simultaneously improve privacy protection, regulatory compliance, international collaboration, and computational efficiency without compromising predictive performance. Furthermore, integrating Green AI principles and edge computing extends existing literature by demonstrating that sustainable AI architectures can reduce computational resource consumption while providing continuous, real-time rehabilitation support. From a healthcare systems perspective, the proposed framework reinforces previous evidence suggesting that digital rehabilitation ecosystems can improve continuity of care, optimize resource utilization, and facilitate equitable access to specialized physiotherapy services across healthcare settings. The presented results indicate that the convergence of AI, IoMT, multiphysics modeling, wireless bio-integrated sensors, and DT technologies represents a promising direction for future precision rehabilitation, while also emphasizing the need for further clinical validation, standardization, and interdisciplinary collaboration to translate these conceptual advances into routine clinical practice [97].

6.1. Technological Implications

The proposed GIR-DT platform creates a scalable technological framework that integrates edge AI, cloud-based DTs, and IoMT infrastructure, enabling continuous and personalized rehabilitation in clinical and home environments. FL significantly enhances the platform’s technological robustness by enabling collaborative training of AI models across geographically distributed healthcare facilities while maintaining patient privacy and compliance with regional data protection regulations. The hierarchical edge-cloud architecture reduces communication latency and computational burden, enabling real-time inference on mobile devices and synchronization of comprehensive DT models with cloud computing resources [98]. The convergence of wireless bio-integrated sensors, multimodal wearable devices, and implantable sensors facilitates the continuous collection of biomechanical and physiological data, enabling highly adaptive rehabilitation strategies based on the patient’s current condition [99]. Multiphysics modeling provides a comprehensive computational framework that simultaneously captures electromagnetic, biomechanical, and physiological interactions, thereby improving sensor performance, device safety, and overall system optimization. A digital rehabilitation coordination layer dynamically integrates AI predictions, multiphysics simulations, clinician insights, and patient-reported outcomes to personalize treatment intensity, progression, and clinical risk management throughout the rehabilitation process. Green AI principles incorporated into a federated architecture reduce computational energy consumption while maintaining high predictive accuracy, thus supporting the environmentally sustainable implementation of smart rehabilitation technologies on a global scale [100]. Standardized interoperability between AI algorithms, wireless communication protocols, electronic medical records, and DT platforms facilitates seamless data exchange and supports international coordination of rehabilitation services [101]. Advanced AI-based digital biomarkers extracted from continuously monitored patient data enhance predictive assessment, early detection of functional decline, and evidence-based clinical decision-making for precise physiotherapy [102,103,104]. The proposed GIR-DT ecosystem represents a technologically advanced and globally interoperable digital rehabilitation infrastructure that supports precision medicine, continuity of care, improved functional recovery, and the future convergence of AI, IoMT, wireless healthcare systems, and personalized physiotherapy.

6.2. Economic and Organizational Implications

The proposed GIR-DT platform has significant economic implications, enabling more cost-effective rehabilitation through continuous remote monitoring, thereby reducing hospitalizations, unnecessary outpatient visits, and long-term healthcare expenses. Integrating edge AI with cloud-based DTs optimizes the allocation of computing resources, lowering infrastructure costs while maintaining high-performance clinical decision support across healthcare settings [105]. FL facilitates collaborative AI model development without centralized data storage, reducing expenses related to data transfer, cybersecurity management, and regulatory compliance, while also supporting international research collaboration. Implementing Green AI principles further reduces operational costs by minimizing energy consumption during model training and deployment, contributing to the emergence of financially sustainable digital healthcare ecosystems. From an organizational perspective, the digital rehabilitation coordination layer facilitates multidisciplinary collaboration by integrating clinicians, physical therapists, biomedical engineers, and AI specialists into a unified decision support environment that improves care coordination and workflow efficiency [106]. Standardized interoperability between IoMT devices, electronic health records, and DT platforms enables healthcare organizations to implement scalable rehabilitation services while reducing system fragmentation and administrative complexity [107]. The proposed architecture supports value-based healthcare by continuously monitoring patient outcomes and providing objective performance metrics that facilitate evidence-based reimbursement models and optimize healthcare resource allocation. The scalability of the GIR-DT ecosystem enables healthcare providers in both developed and resource-constrained regions to expand rehabilitation services without commensurate (often abrupt) increases in staffing or physical infrastructure, thereby improving organizational resilience and accessibility. From a financial perspective, predictive analytics and personalized rehabilitation programs contribute to faster recovery, shorter rehabilitation times, and fewer complications and hospital readmissions, generating long-term savings for healthcare systems and insurers [108]. The proposed globally interoperable rehabilitation ecosystem creates a sustainable economic and organizational framework that supports precision physiotherapy, efficient use of resources, equitable access to advanced rehabilitation technologies, and cost-effective digital transformation in international healthcare systems.

6.3. Social Implications

GIR-DT has the potential to significantly improve societal well-being by increasing equitable access to high-quality rehabilitation services, regardless of geographic location or healthcare infrastructure availability. Continuous remote monitoring supported by IoMT technologies enables patients to benefit from personalized rehabilitation in their homes, strengthening community-based healthcare and promoting independent living for older adults and people with chronic disabilities [109,110]. By integrating rehabilitation services with smart city and smart territory infrastructure, the GIR-DT ecosystem supports intelligent public health management through connected healthcare networks capable of delivering timely and coordinated interventions. The platform contributes to reducing regional healthcare disparities by providing standardized, interoperable, and clinically validated rehabilitation services that can be implemented in urban, rural, and underserved communities [111]. FL facilitates international collaboration between healthcare providers while protecting patient privacy, thereby promoting global knowledge sharing and accelerating the development of inclusive AI-based rehabilitation solutions. Implementing ethical AI principles, cybersecurity mechanisms, and compliance with regional data protection regulations strengthens public trust and social acceptance of intelligent digital healthcare technologies [112]. Personalized rehabilitation supported by AI-based DTs improves recovery, social participation, and quality of life, enabling patients to more quickly return to family, educational, professional, and social activities. At the regional and national levels, implementing interoperable digital rehabilitation ecosystems strengthens healthcare resilience by improving continuity of care during public health crises, population aging, and workforce shortages [113]. The proposed platform also supports digital inclusion by encouraging the widespread use of connected healthcare technologies and fostering digital health literacy among patients, caregivers, and healthcare professionals. Ecosystems such as GIR-DT contribute to the development of healthier, more inclusive, and technologically resilient communities, smart cities, regions, and countries by enabling accessible, personalized, and evidence-based rehabilitation within globally connected healthcare systems.

6.4. Ethical and Legal Implications

The proposed GIR-DT platform and similar solutions may raise significant ethical and legal concerns, requiring the integration of privacy by design, security by design, and trustworthy AI principles throughout the rehabilitation ecosystem [114]. The use of FL minimizes the transfer of sensitive medical information, enabling decentralized training of AI models, thus supporting compliance with the EU General Data Protection Regulation (GDPR) through enhanced data minimization, confidentiality, and patient control over personal data. Implementation of AI-based decision support systems in rehabilitation should comply, for example, with the requirements of the European Union Act on AI, particularly with regard to high-risk [115] AI systems used in healthcare, including risk management, transparency, human oversight, technical robustness, and post-market monitoring. The coordination layer for digital rehabilitation preserves meaningful human participation, ensuring that clinicians retain responsibility for treatment decisions, while AI provides understandable recommendations that support, rather than replace, professional judgment [116]. Compliance with the EU Directive on Network and Information Security (NIS2) requires healthcare organizations using the GIR-DT platform to implement comprehensive cybersecurity management, incident response procedures, supply chain security measures, and continuous risk assessments for connected medical devices and IoMT infrastructure. Ethical implementation of DTs also requires algorithm transparency, fairness, and continuous monitoring to minimize bias that could disproportionately affect vulnerable groups, including older adults, people with disabilities, and underserved communities [117]. Interoperability standards and internationally harmonized governance frameworks facilitate legitimate cross-border collaboration while ensuring data exchange complies with regional legal requirements and ethical principles [118,119]. Robust informed consent mechanisms should clearly explain how patient data is collected, processed, analyzed, and used to improve AI models, thereby strengthening patient autonomy and public trust in intelligent rehabilitation technologies. Integrating multiphysics simulations, wearable sensors, and AI-generated digital biomarkers requires rigorous clinical validation, accountability mechanisms, and a clear division of legal responsibilities between technology developers, healthcare providers, and regulators [120]. The proposed GIR-DT ecosystem demonstrates that the successful global implementation of AI-based rehabilitation technologies depends on balancing technological innovation with ethical governance, regulatory compliance, cybersecurity resilience, patient rights, and transparent, human-centered clinical decision-making.

6.5. Implications for Sustainability

The proposed GIR-DT platform and similar solutions contribute to sustainable healthcare by enabling remote, AI-assisted rehabilitation that reduces patient travel, lowers greenhouse gas emissions, and mitigates the environmental impact associated with conventional in-person care [121]. The hierarchical edge-cloud architecture supports Green AI principles by efficiently performing compute-intensive tasks, reducing data transmission, minimizing cloud computing requirements, and lowering overall energy consumption across distributed healthcare infrastructures. FL further enhances sustainability by reducing large-scale, centralized data transfers, thereby reducing network traffic, computing resource utilization, and energy requirements in AI model development. The convergence of IoT, wireless bio-integrated sensors, and future 6G communication technologies enables the delivery of highly efficient, ultra-low-latency healthcare services while supporting intelligent resource allocation and energy-aware network management [122,123,124,125,126,127]. Multiphysics modeling contributes to device sustainability by optimizing the electromagnetic, biomechanical, and physiological performance of wearable and implantable sensors, reducing material waste, extending device lifespan, and minimizing maintenance requirements. The proposed ecosystem aligns with Industry 5.0 principles by integrating human-centric AI, resilient digital infrastructure, and sustainable manufacturing in the healthcare sector, while laying the technological foundation for future Industry 6.0 ecosystems characterized by autonomous, adaptive, and environmentally optimized cyber–physical systems. Standardized interoperability between DT platforms and connected medical devices strengthens global healthcare supply chains by facilitating component compatibility, predictive maintenance, lifecycle management, and more efficient distribution of medical technologies [128]. Reducing reliance on large, centralized data centers and optimized AI computing can also indirectly reduce water consumption associated with cooling high-performance computing facilities, supporting a more sustainable digital infrastructure [129]. Continuous digital monitoring and predictive rehabilitation reduce unnecessary clinical interventions, optimize healthcare resource utilization, and reduce the consumption of medical supplies and associated logistics throughout the rehabilitation process [130]. The GIR-DT ecosystem demonstrates that the convergence of AI-based DTs, IoMT, multiphysics modeling, wireless bio-integrated sensors, Green AI, and emerging 6G technologies can support environmentally sustainable, resource-efficient, and resilient rehabilitation systems while addressing broader sustainability goals in next-generation healthcare and Industry 5.0/6.0.

6.6. Limitation of the Proposed Approach

Although this study presents a comprehensive, interdisciplinary perspective on AI-based digital twins in rehabilitation, the proposed GIR-DT platform remains primarily a conceptual architecture that requires extensive clinical validation in diverse patient populations and healthcare settings. Practical implementation of the platform may be limited by variability in IoMT infrastructure, wireless communication quality, computational resources, and digital health maturity across countries and healthcare systems [131,132,133]. The effectiveness of AI-based personalization depends on the availability of large, high-quality, multimodal datasets, and differences in data standards, sensor accuracy, and patient adherence can impact model performance and generalizability [134,135,136]. Although FL enhances privacy and regulatory compliance, challenges related to heterogeneous local datasets, communication overhead, model convergence, and interinstitutional coordination remain important research topics [81,137,138]. Integrating multiphysics modeling with real-time digital twin synchronization introduces significant computational complexity, which may limit implementation on resource-constrained edge devices and require further optimization [139,140,141,142]. Furthermore, the interoperability of wearable sensors, implantable devices, electronic health records, and cloud-based digital twin platforms has not yet been fully standardized, potentially limiting seamless integration across healthcare facilities and international environments [142,143,144,145]. The rapidly evolving regulatory environment, including new requirements for AI governance, cybersecurity, and medical device certification, may require ongoing adaptation of the proposed framework to ensure long-term regulatory and ethical compliance [146,147,148].

Although the current state of the literature is relatively limited, it is necessary to adopt a broader global perspective, rather than focusing exclusively on publications from a small group of countries. Research into DTs based on AI, IoMT, multiphysics modelling and wireless sensors integrated into the body is still in its infancy worldwide, and the available studies do not yet provide a comprehensive picture of the geographical situation. The potential for implementing these technologies across diverse healthcare systems, including in resource-constrained and developing regions, varies and depends not only on technical factors but also on organisational and economic factors, the preparedness of medical professionals, and the legal framework. Scalable cloud-edge architectures, interoperable IoMT platforms and energy-efficient wearable sensors can reduce barriers to the implementation of these solutions in settings with limited healthcare infrastructure. At the same time, we recognize that challenges such as unreliable network connectivity, limited computing resources, high implementation costs and a shortage of qualified staff may hinder large-scale deployment. It is also worth emphasizing the importance of cost-effective system design, open standards and flexible AI models that can operate under resource-constrained conditions. International cooperation, technology transfer and capacity-building initiatives.

6.7. Key Directions for Further Studies

Future research should include prospective multi-center clinical trials, cost-effectiveness analyses, user acceptance studies, and long-term, real-world evaluations to validate the clinical, technical, organizational, and societal benefits of the proposed GIR-DT ecosystem. Research is needed to develop understandable, trustworthy, and privacy-preserving AI methods, including FL and secure edge intelligence, to improve transparency, cybersecurity, and regulatory compliance in digital rehabilitation [149]. Further research should explore high-quality multiphysics digital twin models that accurately represent the complex interactions between biomechanical, physiological, and electromagnetic processes occurring during rehabilitation [150]. Long-term, multi-center validation studies are essential to assess the effectiveness, reliability, cost-effectiveness, and clinical impact of AI-based digital twin systems in diverse patient populations and rehabilitation settings [151]. Future work should also explore the integration of emerging wireless technologies, flexible bio-integrated electronics, digital biomarkers, and multimodal sensors to enable continuous, adaptive, and personalized rehabilitation with improved patient outcomes [152]. Interdisciplinary research should establish global standards, ethical governance frameworks, and equitable implementation strategies that ensure scalable, accessible, and sustainable AI-based digital twin ecosystems for precision rehabilitation and next-generation wireless healthcare [153,154].

8. Conclusions

This review demonstrates that the convergence of AI, DTs, IoMT, wireless bio-integrated sensors, and multiphysics modeling is transforming rehabilitation and physiotherapy from reactive clinical practice to intelligent, predictive, and personalized healthcare, thereby addressing the technological aspects of RQ1 and RQ2. The analysis also demonstrates that realizing this vision requires overcoming significant challenges related to interoperability, cybersecurity, ethical AI, data management, scalability, and equitable access, as highlighted in RQ3. In response to these challenges, this paper proposes the GIR-DT ecosystem as a holistic conceptual framework that integrates patients, clinicians, AI-based DTs, wearable and implantable sensor technologies, multiphysics simulation, cloud computing, and secure wireless communication within the ever-evolving rehabilitation infrastructure. The GIR-DT ecosystem expands existing DT paradigms by emphasizing global interoperability, explainable and trustworthy AI, standardized data exchange, adaptive therapeutic intelligence, and human-centered rehabilitation across diverse healthcare environments. Furthermore, the ecosystem provides a strategic foundation for future research by supporting multimodal digital biomarkers, real-time predictive analytics, personalized therapy optimization, and seamless integration with next-generation wireless healthcare networks, thus addressing the opportunities identified in RQ4. The proposed GIR-DT ecosystem offers a scalable vision for precision rehabilitation that can accelerate the development of a smart, safe, and sustainable digital healthcare infrastructure while improving clinical outcomes, patient engagement, and equitable access to rehabilitation services globally.