Global Insights a Bibliometric Analysis of Artificial Intelligence Applications in Rehabilitation Worldwide

Abstract

Rehabilitation plays a vital role in helping patients recover functionality after illness or injury. However, challenges remain in providing customized, accessible rehabilitation services. Artificial Intelligence (AI) techniques like machine learning are emerging as promising tools to enhance rehabilitation. This study aimed to conduct a bibliometric analysis to synthesize global growth trends, research foci, and collaborative patterns in AI rehabilitation research.

Keywords

Artificial intelligence; Bibliometric; Global; Rehabilitation; Machine learning.

Introduction

Rehabilitation plays an essential role in helping patients regain functionality and improve quality of life after injury, illness, or disability. However, providing comprehensive and individualized rehabilitation services remains challenging due to constraints such as limited accessibility, high costs, and lack of customization [1]. (AI) has emerged as a promising solution to address some of these challenges and enhance rehabilitation outcomes. (AI) has already been successfully incorporated into many healthcare domains, from diagnostic imaging to genomic testing [2]. This has spurred interest in leveraging AI's capabilities of automation, adaptation, and data analytics to make rehabilitation more effective and efficient. Although still an emerging application area, recent research indicates growing use of AI techniques in rehabilitation to develop smart assistive devices, optimize therapy plans, and improve assessments.

One active area of AI research aims to develop robotic solutions that can physically assist patients during therapy [3]. Robotic exoskeletons and prostheses use AI algorithms to intuit patient movements and provide customized support. Such intelligent devices increase the intensity and accessibility of physical rehabilitation. VR and AR are also being combined with AI to create more adaptable and gamified simulations for cognitive and motor rehabilitation [4].

Additionally, AI shows promise for more accurate quantification and analysis of patient condition and progress. Machine learning techniques can automate assessment of gait and movement disorders to support rehabilitation planning and evaluation [5]. AI may thus enhance both the quality and efficiency of rehabilitation services.

However, challenges remain regarding the safety explain ability, and ethics of applying AI algorithms in rehabilitation [6]. More research is needed to develop transparent and fair AI systems that clinicians can trust. This bibliometric study aims to understand the current global landscape of AI applications in rehabilitation through indepth analysis of research publications. Mapping out active topics, trends, and collaborative networks can provide insights to guide future AI development and implementation for rehabilitation.

Materials and Methods

This bibliometric analysis utilized the Scopus and WOS databases to comprehensively review literature on AI applications in rehabilitation [7]. Scopus and WOS were selected because they index millions of publications across healthcare, computer science, engineering, and other pertinent fields, allowing thorough capture of interdisciplinary rehabilitation AI research [8].

Scopus contains over 70 million records sourced from more than 5,000 international peer-reviewed journals, books, and conference proceedings [8]. Scopus offers sophisticated search and filtering options to pinpoint relevant documents through Boolean operators, wildcards, field tags, and more. Citation tracking and detailed export features enable in-depth bibliometric analysis.

Meanwhile, WOS provides access to over 18,000 journals spanning science, social science, arts and humanities via several indices, including the Science Citation Index (SCI) expanded and conference proceedings citation index [9]. WOS offers cited reference searching, analysis tools, and data exports valuable for bibliometric reviews.

Together, Scopus and WOS provided comprehensive coverage of AI and rehabilitation literature for identifying publication and citation trends, productive author’s institutions, influential publications, and international research networks essential to understand this emerging field. The interdisciplinary focus of these databases enabled gathering insights across computer science, health sciences, engineering, and related areas.

Research strategy

The research strategy was developed based on an extensive review of literature on AI applications in rehabilitation. Separate searches were conducted in WOS and Scopus using a combination of relevant title keywords [10]. For the WOS search, we used ("artificial intelligence" OR "machine learning" OR "deep learning") in the title field AND "rehabilitation*" in the title field. For Scopus, the search query was: (TITLE ("artificial intelligence" OR "machine learning" OR "deep learning") AND TITLE ("rehabilitation*"). These searches focused the results on documents with clear relevance to AI techniques applied specifically in rehabilitation contexts. The research strategy was limited by (1) the study period from 2000 to 2024 and (2) research or review articles as well as conference papers published in peer-reviewed journals. No language restriction was imposed.

Validation and data export

The initial dataset retrieved from Scopus (CSV format) and WOS (text format) were exported and consolidated in Scientopy for validation and bibliometric analysis. After removing 48 non-relevant document types, the pooled corpus contained 497 publications. Of these, 215 (43.3%) were from WOS and 282 (56.7%) were from Scopus. Deduplication identified 182 duplicate entries, with 1 from WOS and 181 from Scopus removed. This resulted in a final dataset of 315 unique publications, with 214 (67.9%) from WOS and 101 (32.1%) from Scopus.

To validate the search strategy, the research productivity of five active authors in AI rehabilitation was checked against their publication counts in the filtered scientopy dataset. A Spearman correlation test found a significant strong agreement (p<0.001; r=0.96) between the two sets of numbers. This suggests high sensitivity and precision of the search queries in capturing relevant literature. The validated scientopy dataset was then used for all subsequent bibliometric analyses of publication trends, citations, authorship, and research networks.

The validated scientopy dataset was supplemented by exporting metadata to VOSviewer, a freely available computer program for bibliometric mapping [11].

Biblioshiny also used for identifying thematic mapping [12]. VOSviewer was utilized to construct visualization maps depicting collaborative networks and research hotspots [13]. In the resulting maps, node size indicates the frequency of occurrence of items like keywords, authors, or journals. Node color represents clustering of closely related items. The thickness of connecting lines signifies the strength of links between nodes, such as frequent co-authorship relationships.

Additional bibliometric indicators were calculated using the scientopy dataset, including total and mean number of citations received, and the h-index measuring scientific impact based on citation counts. VOSviewer complemented scientopy to provide both qualitative mapping and quantitative citation analyses for a comprehensive perspective on AI rehabilitation research [14].

Bibliometric indicators and data presentation

The scientopy and VOSviewer datasets were analyzed to extract relevant bibliometric indicators that characterize the growth and structure of AI rehabilitation research. Trends in annual publication output were visualized using line graphs generated in scientopy. Active countries, institutions, authors, and journals were ranked and presented in tabular format based on productivity metrics. Top-cited articles were compiled in a table (1) indicating total citations received.

Research themes and concepts were mapped through Biblioshiny to identify hot topics as well as declining or niche areas. Biblioshiny provides interactive visualizations of research topics and their connections over time based on text mining of titles, abstracts, and keywords. Co-occurrence analysis of author keywords in VOSviewer provided further insight into connections between research foci. Finally, collaborative networks were illustrated through VOSviewer visualizations of co-authorship and citation relationships. Together, these metrics, rankings, maps, and graphs provide a multidimensional perspective on the bibliometric landscape of (AI) applications in rehabilitation. Both quantitative indicators and qualitative visualizations were included for a comprehensive profile of this emerging interdisciplinary domain.

Results

General characteristics of the retrieved articles

The consolidated corpus for bibliometric analysis contained 315 publications on AI applications in rehabilitation research over the timespan from 1987 to 2024 April. These documents were sourced from 234 different outlets including journals, books, and conference proceedings. Analysis of annual research output showed gradual growth in this field with a 3.01% average annual increase in publications. The average age of documents was 3.34 years and the mean number of citations per paper was 6.952. In terms of document contents, the dataset contained 919 Author Keywords Plus and 274 Author Keywords reflecting the disciplinary breadth and research foci.

A total of 429 unique authors contributed to the literature, with only 4 single-authored documents. On average, each document had 1.45 co-authors, indicating a high degree of scholarly collaboration. The international collaboration rate was 3.492% based on cross-country co-authorships.

With respect to document types, most publications were journal articles (175), followed by conference papers (67), proceedings papers (44), and reviews (26). Only 4 documents were single-authored works. The collaborative nature and cross-disciplinary appeal of AI and rehabilitation research is evidenced by the diversity of contributing authors and dispersed publication outlets spanned by this analysis.

Evolution and growth of publications

Analysis of publication outputs over time shows that AI rehabilitation research has gone through four growth phases (Figure 2). Only sporadic publications (3 documents) appeared in 2007-2008. This represents the initiating period. Research activity slowly began accelerating between 2013-2015, marked by 6 publications. The field then entered a period of rapid expansion from 2016-2019, with publication counts rising from 13 to 37. The most recent and ongoing phase from 2020-present exhibits exponential growth, with yearly output jumping to 50+ papers.

Overall, WOS contained the majority of publications (212) compared to scopus (99). The chronological publication trends highlight that AI-rehabilitation is still an emerging domain, with 89% of papers published from 2016 onwards. The sharp upward trajectory in the last few years reflects increasing attention and productivity in applying AI to enhance rehabilitation assessment, devices, therapies, and access. Further growth is anticipated as more researchers enter this interdisciplinary area.

Table 1. The general characteristics of the retrieved articles.

Top 10 active countries

Presents the top 10 countries with the most publications on AI rehabilitation. China leads with 31 papers, followed distantly by the United States (15), Taiwan (8), Italy (7), India (6), Malaysia (5), Canada (3), Germany (3), Japan (3), and the United Kingdom (3). Notably, China did not have any publications until 2009 but has since rapidly emerged as the dominant country, contributing close to 10% of total output. In contrast, early pioneering work came from Japan and Germany in 2007-2008. The US maintained steady low productivity until recently showing an uptick from 2018-onwards. In general, Asian countries including China, Taiwan, India and Malaysia demonstrate growing research activity in this domain. The geographic distribution indicates AI rehabilitation attracts global interest, albeit with a focus in China reflecting wider national prioritization of AI research and development (Figure 2).

Top 10 most active journals

Presents the top 10 journals with the most publications on AI rehabilitation. IEEE Access published the highest number of papers (11), followed by sensors (8), IEEE transactions on neural systems and rehabilitation Engineering (5), computational and mathematical methods in medicine (4), and frontiers in neuroscience (4). Additional journals with 3 papers each included computers in biology and. Medicine, BMC medical informatics and decision Making, PLOS ONE, computational intelligence and neuroscience, and diagnostics.

The list contains prestigious journals covering medical informatics, biomedical engineering, neuroscience, sensors, and computational methods. Most are published by leading publishers like IEEE, Hindawi, Elsevier, Springer Nature, Frontiers, and MDPI. Impact factors range widely from 2.3 to 9.2. Overall, AI rehabilitation research appears across diverse journals, indicating its broad relevance. However, IEEE Access and IEEE Transactions emerge as preferred outlets likely due to the engineering and computing aspects of this field.

Table 2. The top 10 most active journals.

Citation analysis

The 315 articles in the corpus received a total of 2189 citations, with a mean of 6.95 citations per document and an h-index of 32. Table 3 lists the top 10 most cited articles, published in prestigious journals like IEEE journal of translational engineering in health and medicine, computers in biology and medicine, IEEE transactions on neural systems and rehabilitation engineering, and scientific reports.

Table 3. The top 10 most cited articles.

The most cited paper (113 citations) presented an IOT-enabled stroke rehabilitation system using smart armbands and machine learning. Other highly influential works focused on deep learning frameworks for assessing exercises, arm movement classification, home-based rehabilitation systems, and mortality prediction after stroke rehabilitation. Overall, pioneering studies on AI and machine learning rehabilitation applications received the most citations, highlighting their significant research impact. The high citation counts indicate the importance ascribed to these studies advancing and demonstrating AI techniques for improved rehabilitation outcomes (Table 2).

Visualization of thematic trends

Thematic analysis using Biblioshiny uncovered five clusters of research concepts in AI rehabilitation literature. The largest cluster represents machine learning approaches, while a closely related cluster covers deep learning methods. This reflects the prominence of machine learning and deep learning as active topics, aligning with their uptick in publications after 2016. Two smaller clusters contain works on convolutional neural networks and support vector machines - more specific AI techniques.

The second largest cluster encompasses rehabilitation applications, with virtual reality and stroke rehabilitation as key themes linked to this group. Additionally, a small cluster connects rehabilitation training to machine learning methods, indicating their combined application. The prevalence of machine learning reflects its versatility across assessments, devices, therapies, and other areas within rehabilitation.

Biblioshiny temporal mapping provided further insights into the evolution of topics over time. It reveals that machine learning rapidly emerged as the dominant theme after 2016, overtaking earlier foci like robotics and neural networks.

Deep learning and newer AI applications in computer vision, natural language processing have also gained traction in recent years. However, machine learning has consolidated its position as the most widely used approach. The dynamic thematic visualizations highlight the shifting AI technology landscape as the field continues maturing. They provide greater nuance into trends and trajectories shaping rehabilitation applications (Figure 3).

Patterns in author keywords co occurrences

Presents the clustering analysis of author keywords. The co-occurrence analysis of author keywords revealed 7 clusters reflecting the thematic structure of AI rehabilitation research (Figure 4). The largest cluster (Cluster 1) encompassed deep learning, convolutional neural networks, CNN, and related methods. Cluster 2 connected braincomputer interfaces and rehabilitation robots. Cluster 3 grouped artificial intelligence, rehabilitation, and key applications like stroke. Machine learning formed its own cluster (Cluster 6), while Cluster 4 contained big data and rehabilitation research. Cluster 5 covered activity recognition and feature extraction, and Cluster 7 had classification terms.

The mapping shows deep learning as the predominant technique, while machine learning also occupies a distinct niche. Rehabilitation contexts like stroke and Parkinson's disease are linked to AI methods, with stroke rehabilitation being closely associated to the core rehabilitation cluster. The emerging clusters demonstrate the key AI technologies, health conditions addressed, and analytics tasks within this domain. The keyword grouping patterns provide additional granularity into the topics and relationships driving AI rehabilitation research (Figure 4).

Discussions

This bibliometric analysis revealed key insights into the rapid growth and evolving landscape of AI rehabilitation research over the past two decades. The publication trends uncovered in our study highlight the rising prominence of AI as a tool to transform rehabilitation practices.

The exponential increase in yearly outputs, especially after 2016, can be attributed to greater affordability and accessibility of high-performance computing resources necessary for deep learning algorithms [25]. Advances in sensor technologies, robotics, and VR/AR have also enabled more applications of AI in smart devices and immersive platforms for rehabilitation [26].

Additionally, wider digitalization across healthcare is driving adoption of big data analytics like AI to enhance decision-making and personalization [27]. Our findings reflect how these Technical capabilities and paradigm shifts have spurred AI innovation in rehabilitation.

However, most studies focus on developing novel algorithms rather than implementing AI in clinical settings [28]. More research on integrating AI seamlessly into rehabilitation workflows is essential for real-world adoption. Hybrid systems combining AI with clinician oversight may address barriers around trust and transparency [29]. Studies must also expand beyond technical validation to evaluate AI impact on patient outcomes through longitudinal randomized controlled trials. Our analysis found stroke rehabilitation to be a major focus for AI applications, which can be attributed to stroke's high incidence rate and significant disability burden globally [30]. But there are opportunities to direct AI towards underserved neurological conditions like traumatic brain injury and spinal cord injury. paediatric and geriatric populations also warrant more tailored AI solutions for rehabilitation.

While China leads publications in this field, international collaborations remain low at just 3.5%. Establishing networks like joint labs and data consortiums can enable cross-country studies on larger diverse datasets [31].

Such global partnerships can spur innovation and expand AI access. Furthermore, greater engagement with industry players and entrepreneurs is imperative to drive commercialization.

Our research maps the AI rehabilitation landscape to guide future growth. Key priorities include moving from algorithm-centered studies to patient-centered design, addressing real-world implementation barriers, directing AI to high-need populations, fostering global collaborations, and accelerating translation and commercialization. Fulfilling this potential will require interdisciplinary, cross-sector partnerships. Our findings provide a roadmap for AI's approach in improving rehabilitation quality, outcomes and access.

Acknowledgments

The author would like to thank Faisal Hassan for helping in the validation step of the research strategy.

Code availability

Not applicable.

Author's Contributions

Azliyana Azizan initiated the concept, formulated the methodology, conducted data analysis, created graphics, and interpreted the data. Additionally, Azliyana Azizan authored and submitted the manuscript, with the final version being reviewed and approved by the author.

Funding

None.

Availability of data and materials

All data presented in this manuscript are available on Scopus database using the search query listed in the methodology section.

Declarations

Ethics approval and consent to participate, no approval is needed for this study.

Consent for publication

Not applicable.

Competing Interests

The authors declare that they have no competing interests.

References

1. Mitchell J, et al. Factors that influence the adoption of rehabilitation technologies: A multi-disciplinary qualitative exploration. J NeuroEngineering Rehabil. 2023;20:1-17.

   [Crossref] [Google Scholar] [PubMed]

2. Davenport T, et al. The potential For artificial intelligence In healthcare. Future Healthc J. 2019;6:94–98.

   [Crossref] [Google Scholar] [PubMed]

3. Ali O, et al. A systematic literature review of artificial intelligence in the healthcare sector: Benefits, challenges,    methodologies, and functionalities. JIK. 2023;81:1-19.

   [Crossref] [Google Scholar]

4. Lewis KO, et al. From static web to metaverse: Reinventing medical education in the post-pandemic era. Ann Med. 2024;561:20-23.

   [Crossref] [Google Scholar] [PubMed]

5. Fricke C, et al. Evaluation of three machine learning algorithms for the automatic classification of EMG patterns in gait disorders. Front Neurol. 2021;12:1-11.

   [Crossref] [Google Scholar] [PubMed]

6. Petersson L, et al. Challenges to implementing artificial intelligence in healthcare: A qualitative interview study with healthcare leaders in Sweden. BMC Health Serv Res. 2022;22;1-16.

   [Crossref] [Google Scholar]

7. Azizan A, et al. Bibliometric analysis on geriatric rehabilitation in scopus database (1948-2022). Topics in Geriatric Rehabilitation. 2024;40:60–68.

   [Crossref] [Google Scholar]

8. Falagas ME, et al. Comparison of pubmed, scopus, web of science, and google scholar: Strengths and weaknesses. Faseb J. 2008;22:338–342.

   [Crossref] [Google Scholar] [PubMed]

9. Kokol P. Discrepancies among scopus and web of science, coverage of funding information in medical journal articles: A follow-up study. J Med Libr Assoc. 2023;111:703–9.

   [Crossref] [Google Scholar] [PubMed]

10. Azizan A, et al. Reshaping healthcare: A bibliometric analysis of lessons learned in post-covid-19 health policy. National Public Health Journal. 2023;18:18–24.

    [Crossref] [Google Scholar]

11. Azizan A. Exploring the role of social media in mental health research: A bibliometric and content analysis. Journal of Scientometric Research 2024;131:1–8.

    [Crossref] [Google Scholar]

12. Aria M, et al. Bibliometrix : A R-tool for comprehensive science mapping analysis. JOI. 2017;11:959–975.

    [Crossref] [Google Scholar]

13. van Eck NJ, et al. Software survey: VOSviewer, a computer program for bibliometric mapping. Scientometrics. 2009;84:523–538.

    [Crossref] [Google Scholar] [PubMed]

14. Azizan A. Mapping the muscle mass: A birds-eye view of sarcopenia research through bibliometric network analysis. IJDSHS. 2023;7:134–143.

    [Crossref] [Google Scholar]

15. Yang G, et al. An iot-enabled stroke rehabilitation system based on smart wearable armband and machine learning. JTEHM. 2018;6:1–10.

    [Crossref] [Google Scholar]

16. Khodatars M, et al. Deep learning for neuroimaging-based diagnosis and rehabilitation of Autism Spectrum Disorder: A review. Computers in Biology and Medicine. 2021;139:104-949.

    [Crossref] [Google Scholar]

17. Liao Y, et al.  A deep learning framework for assessing physical rehabilitation exercises.  IEEE Trans Neural Syst Rehabil Eng.  2020;28:468–477.  

    [Crossref] [Google Scholar] [PubMed]

18. Panwar M, et al. Rehab-net: Deep learning framework for arm movement classification using wearable szensors for stroke rehabilitation. IEEE Trans Biomed Eng. 2019;66:3026–3037.

    [Crossref] [Google Scholar] [PubMed]

19. Chae SH, et al.  Development and clinical evaluation of a web-based upper limb home rehabilitation system using a smartwatch and machine learning model for chronic stroke survivors: Prospective comparative study. JMIR mHealth and uHealth.  2020;8:172-176.  

    [Crossref] [Google Scholar] [PubMed]

20. Lin WY, et al. Predicting post-stroke activities of daily living through a machine learning-based approach on initiating rehabilitation.  Int J Med Inform.  2018;111:159–164.  

    [Crossref] [GoogleScholar] [PubMed]

21. Scrutinio D, et al.  Machine learning to predict mortality after rehabilitation among patients with severe stroke. scientific reports.  Sci Rep.  2020;10:201-207.  

    [Crossref] [Google Scholar] [PubMed]

22. Vélez Guerrero MA, et al. Artificial intelligence-based wearable robotic exoskeletons for upper limb rehabilitation: A review.  Sensors.  2021;21:21-46.  

    [Crossref] [Google Scholar] [PubMed]

23. Zhu M, et al. Using machine learning algorithms to guide rehabilitation planning for home care clients. BMC Medical Informatics and Decision Making. 2007;7:20-30.

    [Crossref] [GoogleScholar] [PubMed]

24. Boukhennoufa I et al. Wearable sensors and machine learning in post-stroke rehabilitation assessment: A systematic review. Biomed Signal Process Control.  2022;71:103-197.  

    [Crossref] [Google Scholar]

25. Jiang F, et al. Artificial intelligence in healthcare: Past, present and future. SVN. 2017;2:230–243.

    [Crossref] [Google Scholar] [PubMed]

26. Wang C, et al. Artificial intelligence enhanced sensors enabling technologies to next generation healthcare and biomedical platform. Bioelectron Med. 2023;9:17-20.

    [Crossref] [Google Scholar] [PubMed]

27. Razzak MI, et al. Big data analytics for preventive medicine. Neural Comput Appl. 2019;32:4417–4451.

    [Crossref] [Google Scholar] [PubMed]

28. Lazarus MD, et al. Artificial intelligence and clinical anatomical education: Promises and perils. Anat Sci Educ. 2022;17:249-262

    [Crossref] [Google Scholar] [PubMed]

29. Zhang L, et al. The application of artificial intelligence in glaucoma diagnosis and prediction. Front. Cell Dev. Biol. 2023;11;23-25.

    [Crossref] [Google Scholar] [PubMed]

30. Dresser LP, et al. Artificial intelligence and the evaluation and treatment of stroke. Dela J Public Health. 2023;9:82–84.

    [Crossref] [Google Scholar] [PubMed]

31. Meng G, et al. Determination of driving forces for China’s energy consumption and regional disparities using a hybrid structural decomposition analysis. Energy. 2022;239:122-191.

    [Crossref] [Google Scholar]