This article presents the proposed application of digital twins in rehabilitation. Such a system could lead to an enhancement in the quality of rehabilitation care. It highlights the potential benefits of incorporating such a system into therapy and discusses challenges connected to the issue. The main goal of the article is to discuss publicly available datasets that could be utilized to create a digital twin and to point out their advantages and disadvantages. The presented datasets encompass diverse aspects of rehabilitation: from patient characterization with specific conditions to sensors and wearables utilization, exercise performance monitoring, extended reality integration, and characterization of healthy individuals as the target patient state. As no perfect dataset for digital twins was found, further research is suggested as the direction for achieving a serviceable digital twin in rehabilitation.
The field of rehabilitation is undergoing a significant transformation, with information technology (IT) playing an increasingly crucial role in supporting patients on their road to recovery. This trend is driven by several factors, such as improved accessibility and reach, enhanced engagement and motivation, data-driven insights, and personalized care or improved monitoring and feedback.
According to [
b.jantos@student.po.edu.pl (B. Jantos); m.tomaszewski@po.edu.pl (M. Tomaszewski) 0009-0001-9557-6787 (B. Jantos); 0000-0001-6672-3971 (M. Tomaszewski)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
Additionally, individuals with chronic or long-term health conditions have been unable to maintain their regular follow-up appointments, and healthcare professionals cannot attend all consultations. Telerehabilitation effectively addresses this gap by facilitating and expanding access to rehabilitation services.
Rehabilitation exercises can be repetitive and sometimes tedious. Technologies like virtual
reality (VR), augmented reality (AR), and gamification [
Wearable sensors and motion capture technologies can collect real-time patient progress
data by monitoring various parameters such as movement patterns, joint angles, muscle activity,
and overall physical performance. These advanced technologies [
A digital twin (DT) is a virtual representation of physical objects, systems, or processes
created using real-time data and simulation models. Digital twins serve as dynamic, realtime
digital counterparts of their physical counterparts, enabling continuous monitoring, analysis,
and optimization. The concept of digital twins integrates several advanced technologies,
including the Internet of Things (IoT), artificial intelligence (AI), machine learning (ML), and big
data analytics. The idea of digital twins originated in the manufacturing sector, where it was first
introduced by Michael Grieves in 2002 [
In manufacturing, digital twins are used to model and simulate production processes, helping
to identify bottlenecks, optimize workflows, and reduce downtime [
Smart cities utilize digital twins for urban planning, simulating various scenarios to help
planners and policymakers make informed decisions about infrastructure development, traffic
management, and resource allocation [
In the automotive industry, digital twins are employed in vehicle design and testing,
simulating performance under different conditions to reduce the need for physical testing and
accelerate the design process [
According to [
In healthcare, digital twins facilitate personalized medicine by simulating and predicting
individual treatment responses, enabling precise medical interventions tailored to each patient
[
Digital twins in rehabilitation refers to creating highly detailed and dynamic digital models of patients undergoing rehabilitation. These digital replicas are built using realtime data gathered from various sources, such as wearable sensors, motion capture systems, medical imaging, and patient health records. By continuously updating with live data, digital twins provide a comprehensive and evolving representation of the patient’s physical and physiological state. The core concept of digital twins in rehabilitation involves integrating advanced IT technologies. These technologies allow the digital twin to process and analyze the collected data, facilitating a deeper understanding of the patient's progress and needs. This continuous, realtime feedback loop enables healthcare providers to tailor rehabilitation protocols precisely, enhancing the effectiveness and efficiency of treatment plans.
It can be distinguished by plenty of application areas of DT in rehabilitation. The main ones include:
• Personalized therapy planning - Digital twins customise rehabilitation programs to suit
each patient's unique needs [
• Real-time monitoring, long-term monitoring, and feedback (f.e. enhanced recovery
tracking) - According to [
• Remote rehabilitation and telehealth - The Covid-19 pandemic has introduced new challenges for healthcare systems, particularly in the realm of rehabilitation. Patients recovering from Covid-19 often experience prolonged symptoms and require extensive rehabilitation to regain their pre-illness health status. Digital twins facilitate remote rehabilitation by enabling healthcare providers to monitor and guide patients from a distance. Patients can perform their exercises at home while their digital twin provides continuous data to their therapist. This capability is particularly valuable for patients with difficulty accessing inperson therapy sessions due to geographical, mobility, or pandemicrelated constraints.
• Simulation, predictive analytics, and outcome forecasting - Review [
• Enhanced patient engagement and motivation - Engagement and motivation are critical to successful rehabilitation. Digital twins can provide visualizations and simulations that help patients understand their progress and the impact of their efforts. By visualizing improvements and future potential outcomes, patients are more likely to stay motivated and committed to their rehabilitation programs.
• Economic importance and resource optimization - By utilizing digital twins, healthcare
providers can optimize the use of limited resources, focusing on patients needing the most
intensive support while enabling others to engage in self-guided rehabilitation under remote
supervision [
◦ Biomedical data: Integration of physiological data such as heart rate, respiratory rate, oxygen saturation, blood pressure, and sleep patterns is essential. This data can be collected using wearable sensors, home monitoring devices, and hospital telemetry systems.
◦ Functional data: The assessment of the patient's functional abilities, including mobility, balance, strength, and cognitive function, is crucial. This can be done through standardized assessments, physical therapy evaluations, and selfreported questionnaires.
• Patient-specific modeling: ◦ Physiological models: Computational models representing the patient's physiological systems, such as the cardiovascular, respiratory, and immune systems, should be developed. These models must incorporate the patient's specific characteristics and COVID-19 history.
◦ Functional models: Models capturing the patient's functional abilities, including musculoskeletal dynamics, neuromuscular control, and cognitive performance, need to be constructed. These models should be personalized based on the patient's assessment results.
◦ Integration of patient data: The collected clinical, biomedical, and functional data should be integrated into the patient-specific models. This integration will allow the digital twin to reflect the patient's current state and respond to changes in their condition.
• Real-time monitoring and simulation: ◦ Continuous data stream: A continuous data stream from wearable sensors, home monitoring devices, and other sources must be established to update the digital twin in realtime. This ensures that the twin accurately reflects the patient's current physiological and functional status.
◦ Simulation of interventions: The effects of potential rehabilitation interventions, such as exercise programs, medication adjustments, and alternative therapies, should be simulated on the digital twin. This allows clinicians to predict how patients respond to different treatment options.
◦ Predictive analytics: Machine learning algorithms should be utilized to analyze the data and predict potential complications, such as secondary infections, long-term sequelae, or relapses. This enables proactive interventions and personalized risk management.
• Visualization and communication: ◦ Interactive dashboards: An interactive dashboard that visualizes the patient's digital twin, including their physiological parameters, functional status, and predicted outcomes, should be developed. This dashboard must be accessible to clinicians, therapists, and patients themselves.
◦ Real-time alerts: Real-time alerts notifying clinicians and patients of any significant changes in the patient's digital twin, indicating potential concerns or the need for intervention, should be implemented.
◦ Personalized communication tools: Tools facilitating personalized communication between clinicians, therapists, and patients using the digital twin as a shared reference point should be developed. This promotes collaborative care and patient engagement.
The literature review highlights that the initial step in creating a digital twin for rehabilitation purposes is the collection of appropriate datasets. Comprehensive data collection is fundamental as it provides the essential input needed to build accurate and reliable digital models. This article, therefore, identifies and presents examples of publicly available datasets that are particularly relevant for this purpose. These datasets encompass a wide range of information, including physiological, clinical, and functional data, which are critical for constructing a detailed and dynamic digital twin. By utilizing these resources, researchers can lay a solid foundation for the development and testing of a digital twin prototype specifically designed for the rehabilitation of post-COVID patients.
Physiological datasets might include measurements of heart rate, respiratory rate, oxygen saturation, and other vital signs that can be continuously monitored using wearable sensors. Clinical datasets could provide insights into patient demographics, medical history, the severity of COVID-19 infection, and current symptoms, which are crucial for tailoring the digital twin to individual patient profiles. Functional datasets may offer information on mobility, balance, strength, and cognitive function, all of which are important for assessing and improving rehabilitation outcomes.
The rest of the article delves into these datasets in detail, discussing their origins, contents, and how they can be effectively used in the context of digital twin development. Additionally, it explores the potential of these datasets to support various stages of the digital twin lifecycle, from initial modeling to real-time updates and predictive analytics.
Finding publicly available data suitable for digital twin creation is a significant challenge due to the sensitive nature of such data. Perfect data for digital twin should be gathered per one patient as on observation for a longer time with carefully annotated timestamps. The data should be comprehensive and encompass various diseases, interventions, therapies, rehabilitation processes, medications, exercises, and dietary patterns. Additionally, it should capture the patient's lifestyle factors, which significantly impact their rehabilitation and recovery journey. Unfortunately, no suitable dataset has been identified that meets the specified criteria. Instead, we present samples of publicly available datasets worth considering in prototyping a digital twin. Datasets found during the literature review could be divided into the following categories, each one could be beneficial for constructing DT in different scopes: • Clinical data containing non-specific conditions, excluding mental health, children, and fetal data, in this category most of the datasets focused on ECG records and EEG signals.
• Clinical - general - category for datasets which includes a lot of general patient data like whole electronic health records data. Such a dataset could be crucial in terms of creating a digital twin.
• Clinical - imaging - datasets with images, mostly in DICOM format, or datasets with image annotations.
• Clinical - specific condition - data focused on subjects with specific conditions like stroke patients or diabetes patients.
• Gait, posture and movement - category for datasets that describe body position, movement specific, or pose 3D points. It could be beneficial when it comes to asses exercise quality or detecting the impact of the movement.
• VR and AR - category for datasets that concern the usage of extended reality methods. Only one dataset was found.
• Wearables + sensors - a category that gathers datasets focused especially on wearables and their ability to measure human health signals or exercise performance. Creating a digital twin requires a variety of datasets. In a perfect-world scenario, such datasets will mirror real patients and their reactions to therapy. However, it is challenging to prepare such a dataset, so it is an emerging task in such a domain. To highlight current possibilities, sample available datasets are presented that could be useful in the flow of digital twins in rehabilitation.
The presented diagram shows elements of digital twins in rehabilitation. Red squares were used for existing sample datasets that could be useful in creating a DT. Blue squares represent domains that lack such datasets. Even if some data exists, it is unsuitable for DT in rehabilitation or unavailable in digitalized versions in open access.
The first and fifth stages of the proposed system consist of datasets dedicated to the description of specific medical conditions per patient. Data should be especially focused on features that should be carefully monitored in particular situations to reflect those characteristics in digital twins.
The second element could focus on the element of therapy connected to monitoring exercises and body reactions. The third could harness possibilities of extended reality as a way to visualize exercises adjusted to the patient. The next part could be dedicated to exercising correctness so every subject will be allowed to check if their posture and movement are performed well. Those elements need clinical monitoring and medical evaluation, and they could be partially gathered by electronic systems but still, they have to be assessed by medical professionals.
All those parts need a reference point, which could be a digital twin of a healthy subject. To achieve that it is needed to gather a variety of data, presenting people in different stages of life and with diverse lifestyles.
As an entry point, the Patient-level dataset to study the effect of COVID-19 in people with
multiple sclerosis [
reported by medical professionals or 'patients' for data reported directly by the patients themselves.
Category of age: 0 for age (0,18); 1 for <18,50>; 2 for <51, 70>, 3 for > 70 “not_overweight” for patients with BMI <= 30; “overweight” for BMI > 30 “Yes” if patient was admissioned to the hospital due to No COVID-19 “Yes” for positive and confirmed COVID-19 diagnosis; “No” otherwise One of values “not_suspected”, “confirmed” - diagnosis of the subject “No” if no COVID-19 symptom was observed, “Yes” otherwise “suspected”, “Yes” if patient was in ICU as a result of COVID-19, Yes otherwise “No” “Yes” or “No” “Yes” if chills were present, “No” otherwise
Yes No No Yes Yes Yes Yes Yes
covid19_sympt_ pain Yes” if pain was present, “No” otherwise covid19_sympt_ fever covid19_sympt_loss _smell_taste covid19_sympt_ nasal_congestion covid19_sympt_ pneumonia covid19_sympt_ shortness_breath covid19_sympt_sore _throat covid19_ventilation current_dmt dmt_glucocorticoid edss_in_cat2
“No” otherwise Disease-Modifying Therapy (DMT) Status - this variable No captures a patient's current status regarding diseasemodifying therapy at the time of data entry. It has three categories: Yes: Patient is currently receiving DMT.
No: Patient is not currently receiving DMT.
Never Treated: Patient has never received DMT. “Yes” if patient is
“No” otherwise
0: EDSS score between 0 and 6 (no to moderate disability). 1: EDSS score above 6 (severe disability). taking glucocorticoid, “Yes” if subject is pregnant, “No” otherwise The biological sex of the patient. “Male” for males and No “Female” for females. Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes current_or_former_ smoker dmt_type_overall duration_treatment _cat stop_or_end_date_ combined covid19_outcome_ levels_2 has_comorbidities com_cardiovascular _disease com_chronic_kidney _disease disease com_diabetes com_hypertension com_immunodeficie ncy com_lung_disease com_malignancy com_neurological_ neuromuscular
Multiple Sclerosis (MS) Phenotype: “relapsing_remitting”: Relapsing-Remitting: “progressive_MS”: characterized by steady worsening of symptoms (includes secondary progressive and primary progressive). “other”: Includes clinically isolated syndrome (CIS), missing data, or patient/clinician uncertainty. “Yes” if subject is or was a smoker, “No” otherwise
0 - for subjects whose treatment is less than 11 years, 1 otherwise Date of stopping DMT 0 - for subjects with COVID-19 but not hospitalized No 1 - for subjects with COVID-19 and hospitalized 2 - for subjects hospitalized, in ICU and/or in a ventilation facility “Yes” if any comorbidities are present, “No” otherwise “Yes” if any cardiovascular comorbidities are present, “No” otherwise “Yes” if any chronic kidney diseases are present, “No” otherwise No Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes com_chronic_liver_ “Yes” if any chronic liver diseases are present, “No” otherwise “Yes” if diabetes is present, “No” otherwise “Yes” if hypertension is present, “No” otherwise “Yes” if immunodeficiency is present, “No” otherwise “Yes” if any lung disease is present, “No” otherwise “Yes” if any malignancy is present, “No” otherwise “Yes” if any neurological and/or neuromuscular comorbidities are present, “No” otherwise comorbidities_other
The different dataset that could be used to prepare a digital twin for subjects with medical
conditions is eICU Collaborative Research Database Demo [
3.2.
The second part of the proposed environment - wearables and sensors could be crucial to
creating digital twins dedicated to rehabilitation and physiotherapy, especially in remote
contexts. One of the datasets that could power the digital twin model is ScientISST MOVE:
Annotated Wearable Multimodal Biosignals recorded during Everyday Life Activities in
Naturalistic Environments [
The dataset would be more valuable if the experiments included more exercises performed
by a larger number of volunteers. Additionally, not every routine was performed by every
participant and some of the data were excluded from the dataset because of poor data quality.
Despite its imperfections, it remains a relevant resource for modeling the body's exercise
response with sensors. Such sensors could be useful in monitoring muscle activity during
rehabilitation or in gathering athlete’s fitness data for improving their performance [
The next part of the suggested schema is incorporating VR and AR, these technologies are
increasingly being incorporated into the healthcare processes [
A digital twin could further enhance patient support by incorporating a physiotherapy exercise dataset within its structure - corresponding to the fourth element of the diagram. An example of such a dataset is the Classification of Physiotherapy Exercises Dataset [28]. In this dataset, the data of 7 exercises (knee-rolling, bridging, pelvic tilt, “The Clam”, repeated extension in lying, prone punches, superman) performed by 30 subjects is presented. It contains data from three sensors: • • •
Obbrec Astra Depth Camera, Sensing Tex Pressure Mat,
Axivity AX3 3-Axis Logging Accelerometer (placed on the wrist and the thigh).
Although this dataset provides details about performing exercises, it does not describe the subjects and the exact conditions of the test. Those could be found in not linked article [29]. As the authors state – such data could show the differences between proper execution and patient execution. In terms of digital twins, it could be extended to model the outcomes of every exercise with its risks and opportunities.
The mentioned dataset is relatively small, although some ML algorithms achieve promising results even with a small number of cases [30]. It could be useful in terms of personal trainer assistance, however, it is not enough to build a DT alone. To be more useful it should cover also the impact of exercises on subjects over a longer period, about specific conditions. Another challenge is to create a vast dataset with a variety of performed exercises with multiple sensors executed by subjects with different possibilities – all of them should be supervised by a specialist to ensure high-quality data. 3.5.
The proposed system's fifth component lies at the intersection of medical expertise and electronic medical data collection. With the development of electronic health records, more and more decision processes could be digitized, however, in such sensitive areas as rehabilitation, it is necessary to include human doctors who will be responsible for making decisions, evaluating results, and designing further therapy. A whole digital twin cannot exist without specialized, human assessment – it could help in the process, and provide better communication and monitoring but it is not a replacement for professional care. 3.6.
The final sixth element of the proposed schema includes a dataset encompassing healthy subject data. A well-suited example is 'Autonomic Aging: A dataset to quantify changes of cardiovascular autonomic function during healthy aging [31]. This comprehensive dataset features resting ECG recordings and continuous non-invasive blood pressure measurements from 1,121 subjects across 15 age groups (18-92 years). Additionally, it provides sex and BMI information. While offering a broad age range, the dataset could be further enriched with detailed data on mild activity parameters and subjects' medical history, including past illnesses and therapies.
The goal healthy digital twin also should be equipped with a measure that will quantify how far from an ideal healthy person is a physical twin – the patient. It could be useful to predict at the very beginning of the therapy and during the process estimated level of returning to health for the individual. Such a possibility would have a direct impact on health expenses, therapy planning, and chances of getting back to work for a person. This measure could be remodeled with digital twins depending on the version of therapy without the necessity of applying each one to the patient. In a perfect world scenario, this will save time for medical staff and patients and it will significantly improve life quality without the need for exposure to ineffective therapies.
The adoption of digital twin technology is transforming industries by enabling more informed decision-making, enhancing operational efficiency, and driving innovation. As technologies like AI, IoT, and big data continue to evolve, the capabilities of digital twins are expected to expand further. Future advancements may include more sophisticated simulations, greater integration with augmented and virtual reality, and broader application across emerging fields such as quantum computing and space exploration. Digital twins represent a pivotal shift towards a more connected and data-driven world, where physical and digital realms converge to unlock new possibilities and drive progress across diverse domains.
Creating a digital twin for the rehabilitation of post-COVID patients therefore requires the synergy of various technologies and approaches that together create a comprehensive and dynamic patient model that allows for precise monitoring and optimization of rehabilitation processes. A key element is the availability of specific public datasets that will allow the conducting of appropriate research experiments. For many reasons mentioned in the article, such collections are still missing. While illustrative datasets have been demonstrated, none are sufficient on their own. To prepare a better dataset for digital twins it is necessary to gather and process a significant amount of health records that describe a variety of patients during a long period, each described by specialists. This poses a challenge due to the ongoing slow digitization of medical records, the need for data anonymization, and the adherence to the highest data protection standards.
Another obstacle to creating a useful digital twin is designing a connection between the physical patient and the digital representation. It should be resistant to possible noise from gathered data but sensitive enough to alarm medical staff in case of emergency. his has the potential to significantly enhance the quality of telemedicine and expand patient access to professional rehabilitation services. Furthermore, such a system could potentially reduce therapy costs, shorten therapy duration, and assist specialists in therapy planning. To get closer to achieving a digital twin, additional studies are required in collaboration with medical professionals. Current Solutions using Virtual-Reality-Based Methods in Cardiac Surgery - A Survey, Computer Science, 25, 1, 2024, doi: https://doi.org/10.7494/csci.2024.25.1.5633 [27] J. Streepey and S. Dent, Body Sway When Standing and Listening to Music Modified to Reinforce Virtual Reality Environment Motion (version 1.0.0), PhysioNet, 2021. https://doi.org/10.13026/x32c-cz47 [28] Classificationof Physiotherapy Exercises Dataset,
www.kaggle.com. https://www.kaggle.com/datasets/rabieelkharoua/classification-ofphysiotherapyexercises-dataset [29] A. Wijekoon, N. Wiratunga, and K. Cooper, MEx: Multi-modal Exercises Dataset for Human Activity Recognition, arXiv (Cornell University), Jan. 2019, doi: https://doi.org/10.48550/arxiv.1908.08992. [30] M. Tomaszewski, P. Michalski, J. Osuchowski, Evaluation of Power Insulator
Detection Efficiency with the Use of Limited Training Dataset. Appl. Sci. 2020, 10, 2104, https://doi.org/10.3390/app10062104 [31] A. Schumann and K. Bär, Autonomic Aging: A dataset to quantify changes of cardiovascular autonomic function during healthy aging (version 1.0.0), PhysioNet, 2021. https://doi.org/10.130 Patient-level dataset to study the effect of COVID-19 in people with Multiple Sclerosis 26/2hsy-t491