Scoping review of AI-driven wearable technologies for rehabilitation and functional assessment in Parkinson’s disease: a protocol

Abstract

Background

Parkinson’s disease (PD) is a common neurodegenerative disorder characterised by high prevalence and disability rates, severely impairing patients’ quality of life and imposing a substantial societal burden. Rehabilitation interventions are an essential component of PD management; however, conventional face-to-face rehabilitation is constrained by limited resources and poor adherence. In recent years, the integration of artificial intelligence (AI) with wearable technologies has offered new avenues for rehabilitation, enabling continuous monitoring, objective assessment and personalised feedback. Although relevant studies have emerged, most are limited by small sample sizes, short-term designs and a lack of comprehensive synthesis. This study aims to conduct a scoping review to summarise the current applications of AI-driven wearable technologies in PD rehabilitation and functional assessment, and to identify existing research gaps.

Methods and analysis

This review will follow the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews guidelines and employ the Joanna Briggs Institute three-step search strategy. After an initial pilot search, a comprehensive search strategy will be developed to systematically search 10 databases (CNKI, Wanfang Data Knowledge Service Platform, SinoMed, the Cochrane Library, PubMed, Web of Science, CINAHL, Scopus, Embase and the IEEE Xplore Digital Library). One researcher will independently perform data extraction, and another will independently verify the extracted data. Eligible studies will include original research articles, dissertations and conference papers published after 2020, involving PD patients and using AI-driven wearable devices for rehabilitation or functional assessment. Data will be synthesised narratively and presented using tables and figures.

Ethics and dissemination

This study involves only publicly available published data and therefore does not require ethical approval. Findings will be disseminated through peer-reviewed journal publications and presentations at academic conferences.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This scoping review will follow the Joanna Briggs Institute methodology and Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews reporting guideline to ensure methodological transparency.

• A comprehensive search strategy will be implemented across English and Chinese databases.

• Only studies published from 2020 onward will be included to reflect recent developments in artificial intelligence (AI)-driven wearable technologies.

• Due to language and resource constraints, only studies published in English or Chinese will be included, which may result in the omission of some relevant evidence.

• Methodological heterogeneity across studies, including differences in study design, outcome measures and AI approaches, including variations in algorithm complexity, may limit comparability.

Introduction

Parkinson’s disease (PD) is a common neurodegenerative disorder characterised by tremor, bradykinesia, rigidity and postural and gait disturbances.^{1 2} Its high prevalence and disability rates not only severely compromise patients’ quality of life but also impose a substantial burden on families and society.^{3 4} Rehabilitation interventions are an integral part of PD management; however, conventional approaches typically rely on face-to-face guidance, which is resource-intensive and often associated with poor patient adherence. These limitations are particularly evident in resource-limited settings, where access to structured rehabilitation services is constrained by shortages of specialised professionals, geographical barriers and limited availability of technology-supported interventions, resulting in suboptimal rehabilitation coverage and effectiveness.⁵,⁷

In recent years, the rapid advancement of artificial intelligence (AI) and wearable technologies has created new opportunities for PD rehabilitation. Wearable devices can continuously and in real time collect multidimensional data during daily activities and rehabilitation training, including gait, tremor and postural control.⁸,¹⁰ In this review, the term ‘AI-driven wearable technologies’ refers to wearable or wearable-type mobile devices that collect patient-generated data and apply AI algorithms either embedded within the device or applied externally during data analysis. When integrated with AI algorithms, these data enable objective assessment and dynamic feedback, as well as support for remote monitoring and home-based rehabilitation.¹¹ This approach enhances the continuity and individualisation of rehabilitation while showing potential for optimising resource allocation and improving adherence.^{12 13}

Although several studies in recent years have begun to explore the application of AI-integrated wearable technologies in the rehabilitation of PD, the existing evidence remains largely concentrated on technical feasibility or laboratory validation. Most studies are characterised by small sample sizes, short follow-up periods and predominantly single-centre designs.¹⁴,¹⁷ Current applications mainly focus on two directions: (1) quantifying kinematic parameters during rehabilitation training using wearable sensors—such as tremor amplitude,⁸ gait speed,¹⁸ gait variability¹⁹ and postural stability²⁰ and (2) developing AI-based motion monitoring or disease assessment models to identify motor impairments or track symptom progression, rather than directly evaluating the therapeutic efficacy of rehabilitation interventions.²¹,²³ Commonly used wearable devices include inertial measurement units,²⁴ pressure-sensing insoles,²⁵ accelerometers and smartwatches.²⁶ AI algorithms frequently employ support vector machines,²⁷ random forests or deep learning networks,^{28 29} primarily for detecting gait events, recognising freezing of gait or assessing tremor severity. Collectively, these studies have demonstrated the feasibility of integrating AI and wearable technologies for motion data acquisition and pattern recognition, thereby laying a technical foundation for objective rehabilitation assessment.

However, from the perspectives of rehabilitation medicine and nursing, these studies still exhibit notable limitations. First, existing evidence is largely derived from laboratory-based or short-term studies that focus primarily on algorithmic performance and device accuracy, with limited evaluation of clinical utility and external validation.^{14 15} Moreover, many studies involve small sample sizes, short follow-up durations and predominantly single-centre designs.^{8 18 19 21 26 27} Second, most study populations consist of individuals with mild to moderate PD, with insufficient consideration of disease stage, comorbidities and individual variability, thereby limiting the generalisability of findings.^{14 15 18 19} Third, intervention protocols vary widely in type, frequency and duration, and outcome assessment tools lack standardisation, making it difficult to compare results across studies.^{16 17} Aside from a small number of investigations into virtual reality training, cueing strategies and biofeedback systems, there remains a shortage of high-quality, multicentre trials with long-term follow-up to rigorously evaluate the clinical effectiveness of AI-driven wearable rehabilitation programmes.^{14 16 17} In addition, multidimensional outcomes—such as patient adherence, usability, quality of life, clinician workload and resource utilisation—are often under-represented in current research frameworks, and external validation of algorithms as well as real-world evidence remains limited.^{8 14 15 26} At present, there are virtually no large-scale studies assessing the sustainability and clinical benefits of AI-enabled closed-loop feedback systems in home-based rehabilitation environments, nor is there systematic evidence on cost-effectiveness or implementation strategies.

Taken together, the existing literature shows fragmented research objectives, high methodological heterogeneity and a lack of unified evaluation frameworks, as well as insufficient interdisciplinary integration across engineering, rehabilitation and nursing domains. Most prior studies have primarily approached this field from an engineering or informatics perspective, focusing on algorithm accuracy and sensor performance, while the clinical applicability, patient experience and service model innovations from rehabilitation and nursing perspectives have not been systematically synthesised. To address this gap, this scoping review will systematically summarise current applications of AI-driven wearable technologies in PD rehabilitation interventions and functional assessment, synthesise available evidence and identify research gaps. By establishing an integrated framework that bridges technological, clinical and nursing dimensions, this review further aims to clarify what has been achieved and what remains underdeveloped, and to highlight priorities for future study design, outcome standardisation and reporting. The findings aim to inform nursing practice and guide future research.

Methods

Overview

To achieve the study objectives, we will conduct a scoping review in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA).³⁰ The methodological steps adopted are detailed in the following subsections.

Step 1: formulation of research questions

Following an initial search of the literature on PD rehabilitation and an exploration of the applications of AI-integrated wearable technologies, the following research questions were formulated:

1. What types of AI-driven wearable technologies have been applied in PD rehabilitation and functional assessment?

2. For what primary purposes are AI-enabled wearable devices used in PD rehabilitation?

3. What physiological and kinematic signals are typically collected by these devices?

4. What roles, potential value and reported performance do AI-driven wearable technologies have in optimising rehabilitation models, assessing patients’ functional status and promoting individualised interventions?

Step 2: identification of eligible studies

Search strategy

The literature search will follow the Joanna Briggs Institute (JBI)-recommended three-step search strategy for scoping reviews.³¹ First, a preliminary search will be conducted in representative databases using initial keywords to identify relevant synonyms and subject terms, supplemented and expanded with reference to the MeSH thesaurus. Second, a comprehensive search strategy will be developed based on the identified keywords and index terms and systematically applied across all included databases, including CNKI, Wanfang Data Knowledge Service Platform, SinoMed, the Cochrane Library, PubMed, Web of Science, CINAHL, Scopus, Embase and the IEEE Xplore Digital Library. In addition, grey literature sources will be searched, including conference proceedings and dissertations, to capture relevant unpublished or non-peer-reviewed evidence. Third, the reference lists of all included studies will be manually screened to identify additional relevant literature. Duplicate records will be removed prior to the study selection process.

The search strategy was developed by a researcher trained in systematic literature searching (XY) and encompasses four themes: wearable or wearable-enabled mobile devices, rehabilitation training and functional assessment, Parkinson’s disease, and artificial intelligence/machine learning. The Parkinson’s disease-related search terms also include key motor and non-motor functional manifestations, such as gait freezing and sleep-related functions. The search terms related to AI will cover both algorithms embedded within wearable devices and external AI models applied to analyse wearable data. Within each theme, terms will be combined with “OR,” and across themes they will be linked with “AND,” with necessary adaptations made according to database characteristics (shown in online supplemental appendix 1).

Eligibility criteria

The inclusion criteria for this scoping review are based on the Population–Concept–Context framework (see table 1). Studies involving patients with a confirmed diagnosis of PD will be considered eligible. The review will focus on rehabilitation medicine and clinical nursing practice, with particular emphasis on rehabilitation interventions, functional assessment and rehabilitation-related long-term follow-up, including home and remotely delivered applications. Conversely, studies unrelated to PD rehabilitation or functional assessment will be excluded, as will those primarily focusing on disease screening, diagnostic tool development or epidemiological investigations. This scoping review will include quantitative and mixed-methods studies published on or after 1 January 2020, encompassing experimental and observational study designs (such as cohort, case–control and cross-sectional studies). In addition, grey literature, including full-length conference papers and theses/dissertations, will also be considered to reflect recent developments in this field. This review focuses on AI-driven wearable technologies applied to motor and functional rehabilitation and assessment. Studies addressing behavioural or psychological outcomes without motor or functional relevance will not be included. The methodological quality of included studies will be appraised using the Mixed Methods Appraisal Tool.³²

Table 1. Eligibility criteria mapped to the PCC.

	Inclusion	Exclusion
Population	Patients diagnosed with Parkinson’s disease (PD), with no restriction on age, gender or disease stage. Studies including mixed populations will be eligible only if PD data can be extracted separately.	Studies without PD patients or where PD-specific data cannot be separated.
Concept	Studies using non-invasive, body-worn wearable or wearable-type mobile devices that either integrate AI algorithms or generate data analysed using AI techniques (eg, machine learning, deep learning). Devices include smartwatches, wristbands, smart insoles, smart clothing, exoskeletons, wearable sensor systems, electronic textiles and wearable-type mobile terminals. Applications must be related to rehabilitation or functional assessment, such as: Motor function assessment (gait, posture, tremor, joint mobility, etc) Rehabilitation training feedback and efficacy evaluation Monitoring disease progression or functional changes Optimisation of personalised rehabilitation programmes or predictive modelling.	Studies not applying AI methods (only traditional statistics/thresholds). Studies using non-wearable devices, handheld-only devices (eg, smartphones), near-body but non-wearable devices, implanted devices, wired devices requiring connection to external systems, or complex multielectrode systems that require professional operation. Studies focused solely on screening, diagnosis tool development or epidemiology.
Context	Rehabilitation medicine and clinical nursing practice, including long-term follow-up and home/remote rehabilitation scenarios.	Studies limited to diagnostic or risk prediction settings without rehabilitation relevance.
Type of source	Quantitative and mixed-methods original research articles, including experimental and observational studies (cohort, case–control and cross-sectional), as well as theses/dissertations and conference papers for which the full text can be accessed, published on or after 1 January 2020, will be included. Both Chinese and English language studies will be considered.	Reviews, case reports, preprints, study protocols, conference abstracts, posters, editorials and commentaries.

AI, artificial intelligence.

Step 3: study selection

The study selection process will be conducted in two stages: (1) title and abstract screening and (2) full-text screening.

In the first stage, one reviewer (SL) will import the retrieved records into EndNote V.X9 and remove duplicates. Subsequently, two reviewers (SL and SC) will screen the titles and abstracts to determine whether the studies should be excluded or included in the preliminary list. In the second stage, the same two reviewers (SL and SC) will perform full-text screening to establish the final list of included studies. To ensure consistency and rigour, any disagreements regarding eligibility will be resolved through discussion between the two reviewers. If consensus cannot be reached, a third reviewer (XY) will make the final decision. The search results will be fully reported in the final scoping review and presented using a PRISMA flow diagram (see figure 1).

Figure 1. PRISMA flow diagram of study selection. *Consider, if feasible, reporting the number of records identified from each database or register searched rather than the total across all sources. **If automation tools were used, indicate how many records were excluded by a human and how many by automation tools.

Step 4: data extraction and analysis

The included studies will be independently extracted by one reviewer (SL) using a standardised data extraction tool (shown in online supplemental appendix 2). Extracted items will include basic study information, characteristics of wearable devices and details of AI technologies. Prior to formal extraction, the tool will be piloted, and any necessary modifications will be made based on feedback from the research team. All modifications will be incorporated into the final version and applied consistently throughout the review process.

To ensure accuracy, all extracted data will be verified by a second reviewer (SC). If critical data are missing or supplementary information is required, the original study authors will be contacted. Reviewers responsible for data extraction are expected to have basic knowledge of the relevant field and an understanding of study design and statistical methods. It is recommended that data extraction be conducted independently by more than one reviewer to minimise errors and potential bias. If disagreement persists after discussion between the two reviewers, a third independent reviewer will make the final decision. The final data will be organised and analysed narratively in accordance with the predefined research questions, with results presented in textual, tabular and graphical formats.

Step 5: Data synthesis

The extracted data will be synthesised using a descriptive and narrative approach. Studies will be summarised according to general characteristics, wearable device features and AI applications. Wearable devices will be categorised by device type, sensor configuration, wearing location, application setting and monitoring or intervention duration, while AI-related data will be grouped by application purpose, algorithm category, validation strategy and performance metrics. Results will be presented using descriptive tables and figures to illustrate research patterns, methodological characteristics and translational gaps across studies.

Discussion

This scoping review aims to systematically map and synthesise the existing evidence on AI-driven wearable technologies in PD from the perspective of rehabilitation and functional assessment. It is expected to make the following concrete contributions: (1) Construction of an evidence landscape: studies will be structurally categorised by wearable device type, wearing location and sensor signals, application setting and intervention or monitoring duration, enabling future researchers to rapidly locate the rehabilitation tasks and assessment targets addressed by different technological approaches; (2) Development of a ‘task–signal–algorithm’ mapping framework: The review will summarise which signal features are primarily applied to specific rehabilitation-related outcomes and which categories of AI methods are employed to form key analytical workflows, providing a transferable reference for subsequent study design and outcome selection; (3) Distillation of methodological elements for clinical translation: by focusing on the reporting and comparison of validation strategies, performance metrics and interpretability or clinical usability information (such as threshold setting and reporting of decision-relevant indicators), the review will offer more concrete guidance for clinical teams and nursing professionals in selecting, evaluating and implementing these technologies in practice and (4) Identification and visualisation of evidence gaps from proof of concept to real-world closed-loop rehabilitation: the review will synthesise the current state of clinical translation of AI-enabled wearables, highlight areas of research concentration and scarcity, and propose priority directions for future studies to facilitate the transition of AI wearables from ‘monitoring/assessment tools’ to ‘closed-loop rehabilitation management solutions embedded within clinical workflows’.

Based on the above evidence synthesis, this scoping review is expected to enhance overall understanding of the applicability and advantages of AI-enabled wearable technologies in PD rehabilitation and functional assessment, thereby reducing uncertainty in technology selection and application. At the same time, the proposed evidence framework and methodological insights may facilitate greater standardisation and comparability in the design and reporting of future studies. Overall, these outputs are anticipated to support the more effective integration of AI-enabled wearables into clinical and nursing practice, promoting the transition of PD rehabilitation towards digitalised and intelligent care models.

Limitation

This review may also be subject to certain limitations. The included studies may exhibit substantial heterogeneity in terms of sample size, follow-up duration, study design, task paradigms and outcome measures, and current standards for rehabilitation assessment as well as for the application and reporting of AI methods have yet to be unified, which may to some extent limit direct cross-study comparisons and the generalisability of the conclusions. Nevertheless, precisely because of the fragmented nature of the evidence and methodological variability, a systematic synthesis and structured presentation of the existing literature is warranted, providing a foundational reference for future studies with greater methodological consistency and comparability.

Patient and public involvement

This study is a scoping review, and its objectives did not require the involvement of patients or the public.

Ethics and dissemination

As this study is based solely on publicly available published data and does not involve the collection of new human participant data, ethical approval is not required. On completion of the review, all data extraction forms, technical appendices and search strategies will be made publicly accessible. The findings will be disseminated through publication in peer-reviewed journals and presentations at academic conferences.