Digital rehabilitation is a rapidly expanding discipline that involves using digital technology to provide treatments for the rehabilitation process that are affordable, accessible, and user-friendly (Mehl et al., 2018). In contrast to traditional therapy, it seeks to deliver new perspectives on person-centered training and a variety of novel experiences during rehabilitation (Robbins et al., 2018). The field of extended reality (XR) technology is one example of advancements; it creates an interactive experience between the digital and physical worlds (Wonggom et al., 2020).

XR is an umbrella term for immersive technologies, including virtual reality (VR), augmented reality (AR), and mixed reality (MR) (Tu et al., 2023). VR refers to a computer-generated, immersive environment that enables users to interact with digital content and mimics real-world experiences in a virtual environment (Henderson et al., 2007; Weiss et al., 2006; Schultheis and Rizzo, 2001; Wilson et al., 1997). It can be categorized into three levels of immersion: non-immersive experiences, which are delivered through desktop screens; semi-immersive experiences, which involve projection displays; and fully immersive experiences, which rely on head-mounted displays (HMD) to provide a comprehensive sensory environment (Mujber et al., 2004). AR adds digital content to the physical world, enriching real-life experiences by overlaying virtual information on physical objects in real space (Vinolo et al., 2021). In comparison, MR is represented as a more sophisticated iteration of VR and AR by adding interactive experiences with virtual objects in a real-world environment (Palumbo et al., 2022). MR users experience the physical and virtual content co-existing, and the virtual objects as actually present in their physical environment (Žilak et al., 2022). Consequently, XR technology enables the customization of all 3D components in the space, not only objects but also environments, therefore it improves the ability to observe and evaluate actions in real-world settings while in rehabilitation training (Goh et al., 2024). Moreover, XR has been widely proven to offer added value during rehabilitation treatment in patients’ clinical outcomes, such as enhanced motor functions (Laver et al., 2017; Mubin et al., 2019; Saposnik and Levin, 2011), increased community participation (Gorman and Gustafsson, 2022), and improved psychological and cognitive wellbeing (Massetti et al., 2018).

As one of the most used personalized elements in XR training programs, avatars are often added to provide various real-time virtual feedback (Dewez et al., 2021; Liu et al., 2022) and embody human presence in a virtual environment for interactive and effective rehabilitation during training (Wonggom et al., 2020). Numerous studies have highlighted the efficacy of avatars in promoting motor recovery (Liu et al., 2022; Liu et al., 2020) and supporting mental health (Burton et al., 2015). These avatars not only facilitate profound higher levels of user engagement with the virtual environment but also play a crucial role in motor skill reacquisition by enabling self-correction through avatar observation, thereby improving clinical rehabilitation outcomes (Liu et al., 2020; Numa et al., 2015).

Building upon the basic benefits of avatars in XR, this technology has gradually advanced towards 3D and 4D Digital Human Modeling (DHM), which represents human beings with physical appearance, movements, and behaviors in computerized, digital, and virtual visualization (Scataglini and Paul, 2019). While conventional avatars cannot offer the same high realism and lifelike interactions in virtual spaces as 3D and 4D DHM, 3D DHM can only represent a static visualization with fixed parameters, such as body shape (Scataglini and Paul, 2019). DHM in 4D, on the other hand, enables a dynamic capture of the full body shape with both 3D appearance and changes followed by time during a motor task. As a result, the benefits of DHM encourage and facilitate many research projects that aim to investigate this technology’s efficacy in enhancing clinical outcomes and the rehabilitation process.

While conventional avatars cannot offer the same high realism and lifelike interactions in virtual spaces as 3D and 4D DHM, 3D DHM can only represent a static visualization with fixed parameters, such as body shape (Scataglini and Paul, 2019). DHM in 4D, on the other hand, enables a dynamic capture of the full body shape with both 3D appearance and changes followed by time during a motor task. As a result, the benefits of DHM encourage and facilitate many research projects that aim to investigate this technology’s efficacy in enhancing clinical outcomes and the rehabilitation process.

Moreover, Massetti et al. (2018) evaluated the training programs that integrated VR with 3D DHM in people with spinal cord injuries and the results highlighted the potential of using DHM to promote strength, balance, gait, and motor recovery after rehabilitation. A more recent systematic review and meta-analysis by Scataglini et al. (2024) evaluated the accuracy, validity, and reliability of Markerless Camera-Based 3D Motion Capture Systems (MCBS) versus Marker-Based 3D Motion Capture Systems in gait analysis. Spatiotemporal parameters demonstrated excellent accuracy, validity, and reliability, with moderate agreement in hip and knee kinematic variables. While 3D digital health models were produced in both cases, the MCBS, with its 3D and 4D scanning capabilities, proved to be a more comprehensive tool for creating a personalized digital human model that considers size, form, and aesthetics while integrating gait analysis and rehabilitation into inclusive precision medicine.

Another systematic review (Chattopadhyay et al., 2020) also found evidence supporting the efficacy of integrating computer-controlled 3D DHM in patient-facing systems, demonstrating their ability to significantly enhance health outcomes in various types of target populations compared to traditional interventions, such as quality of life (Andrade et al., 2015), physical activity (Bickmore et al., 2013; Ellis et al., 2013), mental health (Wieser et al., 2010), psychological condition (Cassidy et al., 2016), patient education (Gunn et al., 2020). Additionally, only two gait analysis studies were found related to 4D DHM. De Rosario et al. (2023) (De Rosario et al., 2023) presented the biomechanical applications of using 4D scanning technologies, especially in assessing volumetric asymmetries and rehabilitation. While in another study by Meletani et al. (2024) (Meletani et al., 2024) demonstrated the 4D scanning system has comparable reliability and accuracy to inertial measurement unit systems in gait analysis. Both studies confirmed the validity and reliability of 4D DHM and their potential use for measuring biomechanical characteristics in rehabilitation. However, as 4D DHM is a novel method of rehabilitation, no reviews have been found that have thoroughly covered therapy and assessments in XR-based applications.

Therefore, the purpose of this systematic review was to present the first thorough analysis of current training approaches that use 3D–4D DHM in XR–based applications for rehabilitation. Regarding 3D–4D DHM in XR-based rehabilitation, this study made a distinctive contribution to identifying the gaps in up-to-date utilization of the technologies in clinical settings and inspiring further investigations and education by researchers and professionals into 3D-4D DHM in XR-based rehabilitation.

To the best of our knowledge, this is the first systematic review that comprehensively synthesizes the existing research on the effectiveness of 3D-4D DHM within XR-based rehabilitation. The findings across all included studies suggested that 3D DHM holds the potential to impact rehabilitation outcomes across several domains positively. These include functional outcomes, such as improved upper limb function (Tsekleves et al., 2016), gait performance (Nuic et al., 2018), balance function (Nuic et al., 2018), quality of life (Alemanno et al., 2019), and physical activity (Mitchell et al., 2023); physical outcomes, including decreased pain condition (Nuic et al., 2018; Alemanno et al., 2019), spasticity (Tsekleves et al., 2016), increased joint range of motion (Alemanno et al., 2019); psychological measures, such as improved depression and emotional regulation (Perpiñá et al., 2009; Nuic et al., 2018; Nosek et al., 2016; Falconer et al., 2017), reduced body image and weight concerns (Nosek et al., 2016; Keizer et al., 2016), and psychological impact (Mitchell et al., 2023); and general health outcomes, including improved body composition, and metabolic health (Mitchell et al., 2023; Sansoni et al., 2024). Moreover, when compared with conventional therapy, training involving 3D DHM showed largely greater improvements in eye movement of neglect patients (Kim et al., 2007), body image measures of anorexia nervosa patients (Perpiñá et al., 2009), eating disorder measures of bulimia nervosa patients (Sansoni et al., 2024), and diabetes-related indicators of type 2 diabetes patients (Faber and Fonseca, 2014). Additionally, 3D DHM was also demonstrated to be effectively utilized for detecting gait abnormalities (Booth et al., 2019; Zhu et al., 2022) and enhancing ownership of the embodiment experience (Barhorst-Cates et al., 2022; Keizer et al., 2016) during training, which highlights the potential as an innovative tool in XR-based rehabilitation. training.

However, unfortunately, we did not find any studies that related to the 4D DHM utilized directly by patients during treatment sessions, besides 3D DHM. Since 4D DHM advances beyond 3D DHM by enabling dynamic capture of the full body shape, incorporating both the 3D appearance (as in 3D DHM) and the temporal changes that occur during motor tasks (Meletani et al., 2024). All sixteen included studies showed positive results with the use of 3D DHM, therefore, we believe that 4D DHM could also offer added value by providing real-time, highly realistic, and dynamic visual feedback during training, which needs further investigation in future research. As the study by Meletani et al. (2024) proved the convincing reliability and accuracy in estimating gait spatiotemporal parameters and joint kinematics, it also highlighted the ability of the system to distinguish between healthy individuals and people with pathological conditions. Therefore, these huge advantages of 4D DHM should not be underestimated, particularly its accuracy and the informative data it provides, which open new frontiers for using DHM in improving clinical outcomes.

Although no 4D DHM was included in this review, it is important to highlight the findings of Kammler-Sücker et al. (2021), who demonstrated the greater similarity to the user’s realistic appearance in DHM, participants had better engagement and ownership of the embodiment in DHM while training, and thereby, highlighting the realism and personalization of the DHM could significantly affect ROM and movement performance (Kammler-Sücker et al., 2021).

DHM within the XR environment is considered a powerful tool for facilitating clinical outcomes and promoting self-correction during training (Liu et al., 2020; Barhorst-Cates et al., 2022). Nine studies focused on participants with neuromuscular diseases (Tsekleves et al., 2016; Kim et al., 2007; Booth et al., 2019; Barhorst-Cates et al., 2022; Zhu et al., 2022; Nuic et al., 2018; Kammler-Sücker et al., 2021; Alemanno et al., 2019), with four of these studies showing a preference for utilizing non–immersive VR combined with DHM in designed appearances (Tsekleves et al., 2016; Zhu et al., 2022; Nuic et al., 2018; Alemanno et al., 2019). This preference may arise from the increased risk of falls among participants, as fully immersive VR, although it enhances the level of illusion, can isolate users from their real environment, potentially compromising their ability to manage performance and maintain balance effectively. However, it is also notable that no incidences of falls were reported in either fully immersive or semi-immersive settings, as studies were predominantly conducted with participants in a seated position (Xu et al., 2024; Kim et al., 2007; Barhorst-Cates et al., 2022; Kammler-Sücker et al., 2021) or walking with dual belts (Booth et al., 2019) to ensure their safety. In contrast with neuromuscular patients, individuals with psychological, mental, or internal conditions do not often experience decreased balance function or impaired gait performance. However, five (Mitchell et al., 2023; Perpiñá et al., 2009; Nosek et al., 2016; Hall et al., 2011; Falconer et al., 2017) out of seven studies in these populations were still conducted within a non-immersive VR environment on a computer, which might be caused by the concern about the accessibility and ecology of the training itself. Furthermore, only one study from Sansoni et al. (2024) was performed with a DHM in high-realism appearances and well-designed animation in an AR setting. Nevertheless, no significant different effects were found between AR (Sansoni et al., 2024) and VR (Perpiñá et al., 2009) as shown in Supplementary Table S1, due to lack of comparable parameters. In conclusion, despite the moderate to high risk of bias presented in the included studies, the majority (Mitchell et al., 2023; Perpiñá et al., 2009; Tsekleves et al., 2016; Zhu et al., 2022; Nuic et al., 2018; Nosek et al., 2016; Alemanno et al., 2019; Hall et al., 2011; Falconer et al., 2017) demonstrated a preference for utilizing non-immersive VR settings in combination with DHM for rehabilitation training.

Given that visual feedback is a crucial element in influencing performance and maintaining balance in individuals with neurological conditions (Chen et al., 2021), thus, the patient’s performance could vary depending on the different visual content. Further investigations were conducted. Barhorst-Cates et al. (2022) found that using a first-person view of DHM yielded better outcomes and higher ownership of the DHM, although the differences between the first-person view, mirroring view, and anatomical view were not statistically significant. This lack of significant difference might be attributed to the relatively low difficulty of the tasks involved. Xu et al. (2024) explored the third-person view of DHM and found that it generated a strong sense of ownership among stroke patients, though this study lacked a comparison group with an alternative view. Considering the level of evidence, both two studies bring moderate risk of bias to this review, whereas Xu et al. (2024) obtained an EBRO level as an A2 and Barhorst-Cates et al. (2022) obtained a B. Consequently, we agree that a third-person view of DHM could elicit clinical results by inducing a stronger sense of ownership. However, non-immersive VR with a first-person view of DHM was also able to enhance better clinical outcomes in training with DHM (Barhorst-Cates et al., 2022). The results of the EQ in the study by Barhorst-Cates et al. (2022) indicated higher ownership scores with the first-person view, suggesting that this point of view of DHM may better align with participants’ perceptions of their own bodies. This finding is consistent with previous studies from Slater et al. (2009), Slater et al. (2008), Petkova et al. (2011). Therefore, to further clarify the outcomes of these two perspectives, fuure research should address the comparative effectiveness of first-person versus third-person views of DHM to determine which method offers the most significant clinical benefits.

Moreover, two studies (Xu et al., 2024; Keizer et al., 2016) further compared the synchronous and asynchronous visuo-tactile stimulation of the actual and virtual body (DHM). Xu et al. (2024) reported that both synchronous and asynchronous visuo-tactile enhanced brain activation, however, the synchronous visuo-tactile stimulation led to stronger brain activation compared to the asynchronous condition. This was accompanied by an increased event-related desynchronization amplitude in the α and β frequency bands (Xu et al., 2024), which enhanced motor neuron activity, potentially improving motor function restoration and aiding movement initiation ability in stroke patients during motor imagery enhancement (77, 78). While the findings of Keizer et al. (2016) align with those of Xu et al. (2024), demonstrating a positive effect of DHM in both synchronous and asynchronous visuotactile stimulations, with a significant reduction in the misestimation of all body circumferences except for height and abdomen width (Keizer et al., 2016). Participants reported more embodiment of DHM with synchronous stimulation compared to asynchronous stimulation, however, did not show significantly larger amounts of changes in body size estimation between two stimulations, which indicated that the synchronicity of the visuotactile stimulation is not important for changing the experience of body size (Keizer et al., 2016). In short, DHM itself could potentially enhance brain activation, boost motor restoration, and decrease the misestimation of body circumferences. The DHM with synchronous visuo-tactile stimulation could further improve better embodiment of the DHM, compared with asynchronous visuo-tactile stimulation.

Another interesting remark is that while most of the results indicated the positive effectiveness of DHM, its role in interventions for psychological conditions might differ from its role in neurorehabilitation. In psychological and mental health interventions, DHM primarily helps people with anorexia nervosa (Perpiñá et al., 2009; Keizer et al., 2016), bulimia nervosa (Perpiñá et al., 2009; Sansoni et al., 2024), borderline personality disorder (Falconer et al., 2017), and intellectual disabilities (Hall et al., 2011) by addressing the multisensory processing deficit, changing distorted body image perception, and using the perspective-taking function, encouraging participants to engage and reflect more deeply with their emotional feelings and cognitive experiences, thereby enhancing self-awareness to facilitate outcomes (Falconer et al., 2017; Sansoni et al., 2024; Brizzi et al., 2023). In contrast, neurorehabilitation and neuromuscular rehabilitation for people with neglect (Kim et al., 2007), hemispheric disease (Tsekleves et al., 2016; Barhorst-Cates et al., 2022; Zhu et al., 2022), LBP (Kammler-Sücker et al., 2021; Alemanno et al., 2019), PD (Nuic et al., 2018), CP (Booth et al., 2019), and dyskinesia (Xu et al., 2024), DHM primarily functions as a real-time imitation and simulation tool for movement and performance feedback, allowing users to enhance the quality of their performance through self-correction (Liu et al., 2020; Escarti and Guzman, 1999). In this way, DHM helps patients regain control over their physical movements by fostering a more active and engaged learning process within the brain, particularly in cases where motor functions are impaired (Xu et al., 2024). Overall, DHM can provide different effects to enhance self-awareness, self-recognition, and self-correction over various pathologies in the rehabilitation process.

As the final consideration, only one study by Mitchell et al. (2023), utilized 3D DHM in XR environment on a large scale, involving three hundred type 2 diabetes patients. The findings demonstrated that DHM home telerehabilitation achieved clinical outcomes comparable to in-person care in controlling hemoglobin A1c levels, increasing physical activity, and enhancing patient engagement. Despite this RCT study having a moderate risk of bias, the study results highlighted the potential benefits of DHM, including reduced therapist workload and increased treatment portability for patients. In contrast, the remaining studies included in this review were conducted on no more than 30 patients (Xu et al., 2024; Perpiñá et al., 2009; Tsekleves et al., 2016; Kim et al., 2007; Booth et al., 2019; Barhorst-Cates et al., 2022; Zhu et al., 2022; Nuic et al., 2018; Kammler-Sücker et al., 2021; Nosek et al., 2016; Alemanno et al., 2019; Hall et al., 2011; Falconer et al., 2017; Sansoni et al., 2024; Keizer et al., 2016). Four studies (Xu et al., 2024; Barhorst-Cates et al., 2022; Kammler-Sücker et al., 2021; Keizer et al., 2016) explored the mechanisms underlying DHM rehabilitation benefits. For instance, stroke patients showed improved voluntary imitation (Kammler-Sücker et al., 2021) and accuracy (Barhorst-Cates et al., 2022), as well as enhanced motor imagery (Xu et al., 2024), while AN patients experienced improvements in body image disturbance (Keizer et al., 2016). The other eleven studies (Perpiñá et al., 2009; Tsekleves et al., 2016; Kim et al., 2007; Booth et al., 2019; Zhu et al., 2022; Nuic et al., 2018; Nosek et al., 2016; Alemanno et al., 2019; Hall et al., 2011; Falconer et al., 2017; Sansoni et al., 2024) focused on the validation and feasibility of using DHM in rehabilitation, yielding promising outcomes but highlighting significant limitations, such as DHM has not yet been widely adopted in patients’ daily routine training programs. Most applications remain experimental and are not accessible to the general public, likely due to technical complexity, high costs, and the absence of standardized training protocols. Addressing these barriers is critical for the successful clinical translation of DHM technology.

Regarding the limitations of this systematic review, the level of included studies was relatively low with only one RCT and one cohort study. The remaining consisted of five non-randomized case-control studies, six pre-post-experimental studies, and two case studies without comparison groups. Therefore, the lack of robust control groups or randomized designs might introduce bias and confounders into the results. Secondly, it should be noted that the sample sizes were also significantly insufficient in some studies, potentially limiting the generalizability of the findings and validity of the studies (Faber and Fonseca, 2014). While four studies had a large sample size population of more than fifty participants (Mitchell et al., 2023; Kim et al., 2007; Booth et al., 2019; Keizer et al., 2016), only Keizer et al. (2016) achieved a low-risk bias, with the others offering a moderate risk of bias for this review. In addition, besides Keizer et al. (2016), the overall risk of bias in all included studies ranged from moderate to high. Besides Xu et al. (2024) obtained a level of A2, the rest of the EBRO levels ranged from level B-level C. Apart from Xu et al. (2024), which was rated at level A2, the rest of the EBRO levels ranged from level B to level C. Consequently, these fair to low-quality studies may weaken the strength and reliability of the conclusions and diminish the statistical power of the review. Thirdly, the search strategy was restricted to English-language studies, and therefore, would result in publication bias and reduce the statistical power of the overall estimations.

Finally, the interventions and outcome measures used in the included studies showed considerable variations and combinations, combined with the limited number of comparable eligible studies, which made it difficult to conduct a meta-analysis for this systematic review. These variations of the outcome measures might be due to a lack of standardized designs for rehabilitation training and, therefore, complicate the comparison and interpretation of the results from the included studies. Despite these challenges, most findings of the studies were aligned with each other, highlighting the usability and positive effectiveness of DHM in assessment and training. This systematic review pointed out the urgent need for standardized designs or protocols in XR-based DHM rehabilitation settings to improve future research quality.

To address the limitations identified in this review, future research should focus on several points. Firstly, high-quality RCTs with a larger scale of participants are necessary, not only to increase the strength of the study but also to establish reliable results regarding the efficacy of 3D-4D DHM in XR-based rehabilitation. Secondly, future studies should add comparative groups with conventional treatment to reveal a clear understanding of the true added value of the interventions. Thirdly, future studies must develop standardized protocols to address the variability in study parameters within the same domain. This lack of standardization also explains why no studies utilizing 4D DHM could be included in this review because of the variety and non-comparable parameters.

In addition to the focus as described above, from a technological perspective, it is also essential to address the complexity of developing 4D DHM rehabilitation training systems to facilitate practical transition in clinical settings. To overcome this barrier, we suppose that close collaboration between rehabilitation therapists and computer science engineers is very critical for advancing application development. Such an interdisciplinary approach is important for filling the gap between technology superiority and practical transition to develop user-friendly and clinically applicable applications.

Finite element (FE) joint modeling could also serve as a complementary solution to address this issue. A FE model of the entire knee joint was developed and extensively investigated by Adouni et al. (2019), Adouni et al. (2020), their findings demonstrated the model’s capability to simultaneously predict macro-continuum joint mechanics, including compressive forces and stress distribution. Further validation by Faisal et al. (2019) confirmed the model’s ability to detect and record the mechanical response of cartilage in the knee, presenting its potential for understanding complex biomechanical behaviors in joint structures while movement.

Building upon these capabilities of the FE joint modeling, integrating 4D DHM with this technology could provide critical insights into dynamic movements and biomechanical analysis, serving as a complementary solution to address the current challenges in this new field. This combined approach has the potential to significantly enhance the precision, personalization, and overall effectiveness of rehabilitation interventions. Furthermore, it could strengthen the clinical relevance and applicability of 4D DHM, paving the way for its broader integration into rehabilitation training programs.

The implications of this systematic review are significant for the future of XR-based rehabilitation, particularly in integrating 3D and 4D DHM. The consistently positive results from 3D DHM across various patient groups indicate that this technology can enhance rehabilitation outcomes, especially when compared to conventional therapies. The potential of 4D DHM, which captures dynamic body movements, suggests that it could offer even greater benefits, though this requires further investigation. Despite the current lack of direct application studies for 4D DHM and some technological limitations, its accuracy and detailed feedback capabilities could open new avenues in clinical practice. However, the limitations of the included studies, such as small sample sizes and potential biases, highlight the need for more rigorous research to strengthen these findings and support the broader adoption of DHM in rehabilitation.

In conclusion, this review is the first to systematically analyze the effectiveness of 3D-4D DHM in XR-based rehabilitation, revealing consistently positive outcomes and sensitive assessments with 3D DHM across various patient groups. Although 4D DHM was still undiscovered, its potential for enhancing real-time feedback and improving clinical results is promising and warrants further research. Despite limitations, such as small sample sizes and moderate to high risks of bias, the findings suggest that DHM, particularly in non-immersive VR settings, could be a valuable tool in rehabilitation. Future studies should focus on optimizing DHM’s application, particularly comparing first-person and third-person views, and exploring 4D DHM’s full potential.