This scoping review was conducted in accordance with the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines (Checklist 1) [14]. The methodological framework was guided by the five-stage process originally defined by Arksey and O’Malley [15] and further refined by Levac et al [16] to ensure a rigorous and transparent evidence synthesis. To enhance research integrity and prevent duplication, the study protocol was prospectively registered on the Open Science Framework [17].

Eligibility criteria were defined using the PCC framework. The Population encompassed learners and practitioners at various stages of professional acupuncture training, including undergraduate and graduate students, clinical interns, trainees, and practicing experts (eg, licensed acupuncturists or traditional medicine doctors). Studies were included if the target population involved individuals acquiring, developing, or evaluating acupuncture skills within institutional or experimental training settings. Regarding the concept, this review categorized XR systems into nonimmersive VR (NIVR), semi-immersive VR (SIVR), and fully immersive VR (IVR) levels using a study-specific operational classification based on display modality, spatial registration, and interaction characteristics, adapted from the classical taxonomy of Ware et al [18] and Lombard and Ditton [19], as summarized in a review of VR research by Cipresso et al [20]. We included IVR (eg, HMD-based systems) and SIVR (eg, 3D stereoscopic displays and UHD touch tables) that achieved 1:1 spatial mapping, defined as true-scale correspondence between the tracked position of the learner’s head, hand, or needle, or the registered position of a real body or physical model, and the position of the virtual content. IVR required an HMD with head tracking and 3D input through tracked controllers, data gloves, hand tracking, or a tracked needle device. SIVR required a display that did not meet the IVR criteria, such as a life-size, stereoscopic, or handheld screen or a monocular head-worn display, with virtual content presented at true scale and either registered to a real body or model or manipulated with tracked, sensor-based, or haptic input; registration and direct manipulation were treated as alternative, not jointly required, routes to 1:1 spatial mapping. Systems operated only by mouse, keyboard, or touch on a desktop monitor or reduced-scale screen without registration to a body, 360 degree video without manipulation, and reports that did not describe the display device were classified as NIVR and excluded, unless the authors stated that learners practiced in a VR system, in which case immersion was assigned from the system description. The context was acupuncture training, including acupoint localization, needling skills, and acupuncture-related clinical reasoning. Inclusion was limited to peer-reviewed original research and full conference papers; no language restriction was applied. Theses, reviews, editorials, study protocols, trial registry records, book chapters, preprints, and conference abstracts were excluded.

We searched eight electronic databases: PubMed (National Library of Medicine interface), Embase, Scopus, IEEE Xplore, CNKI (overseas platform), RISS, ScienceOn, and OASIS. All database searches were run on September 19, 2026. Backward and forward citation searching of the included studies supplemented the database search.

The search was developed and is reported in accordance with PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Literature Search Extension) guidelines (Checklist 2) [21]. Search strategies combined a context block (acupuncture, acupoints, meridians, and East Asian medicine terms) with a concept block (VR, AR, MR, XR, haptic, HMD, and related device terms) using the Boolean operator AND. The population element and outcome terms were applied at the eligibility stage rather than in the search, because restricting searches by additional concepts reduces sensitivity [22]. Controlled vocabulary (MeSH in PubMed and Emtree in Embase) was combined with free-text terms in title, abstract, and keyword fields. No published search filters were used or adapted, and no search strategy from a previous review was reused.

Syntax was adapted to each database. IEEE Xplore limits a query to 25 terms, so the concept block was reduced to 13 core terms. OASIS indexes only Korean medicine literature, so the concept block was run alone. In CNKI, the subject field was searched using Chinese acupuncture and XR terms. No language, date, or study design limits were applied. The search strategies were internally reviewed by the review team before the searches were run, and the searches were not subsequently rerun or updated. The search strategy for each database, copied exactly as run, with the number of records retrieved at each step, is provided in Multimedia Appendix 1.

Records were exported from each database and deduplicated in Microsoft Excel by matching digital object identifiers and normalized titles within a publication-year window of 2 years. Screening then proceeded in 2 stages.

At the title and abstract stage, 2 reviewers (HIJ and HSP) independently screened all records, without access to each other’s judgments, against the predefined eligibility criteria, and disagreements were resolved by discussion. Records without an abstract were judged on the title, and abstracts were retrieved from the source database when the title was insufficient.

At the full-text stage, the same 2 reviewers independently assessed each report against the eligibility criteria, and disagreements were resolved by discussion. Exclusions were coded as nonimmersive system (F1), no educational purpose (F2), no implemented system or outcome, or a conference abstract (F3), ineligible publication type (F4), or full text not retrievable (F5).

Data charting was performed independently and in duplicate by 2 reviewers (HIJ and HSP) using a standardized charting form; the variables charted are listed under the Data Items section. The charting form was pilot-tested on a subset of included sources before full charting to calibrate its application. The completed forms were compared item by item, and disagreements were resolved by a third reviewer (KHK).

For each included source, the following data items were charted: author, year of publication, country, XR modality and level of immersion, study design, participants and sample size, training content and task, reported outcomes and metrics, hardware and software, and limitations. Country was assigned according to the corresponding author’s affiliation, and Hong Kong was counted within China.

Consistent with established scoping review methodology, formal critical appraisal of the methodological quality or risk of bias of individual sources of evidence was not performed, as the objective of this review was to map the extent, range, and nature of the available evidence rather than to evaluate its quality; this scope decision is discussed further in the Limitations section.

Charted data were summarized descriptively and narratively. Study characteristics were tabulated and reported as frequencies and percentages by country, year of publication, and level of immersion. Included studies were then grouped by the core educational task they targeted, namely acupoint localization, needling skills, and clinical reasoning, and the reported technical and educational outcomes were summarized narratively within each group. Each study was assigned to one domain according to the principal training objective stated by the authors, and outcomes for other tasks were described narratively. To describe the study designs underlying the available evidence, each source was additionally assigned to 1 of 4 evidence tiers: tier 1 for comparative or experimental designs; tier 2 for technical validation; tier 3 for system or hardware development; and tier 4 for case, survey, or qualitative reports, and the distribution of these tiers across modalities and educational domains was displayed as an evidence gap map. These tiers reflect study design categories used for descriptive mapping and are not a validated hierarchy of methodological quality or risk of bias. Finally, identified limitations and future directions for formal curriculum integration were synthesized.

Figure 1 illustrates the study selection process. The database searches identified 6176 records (PubMed, n=548; Embase, n=1576; Scopus, n=1456; IEEE Xplore, n=798; CNKI, n=855; RISS, n=450; ScienceOn, n=476; OASIS, n=17). Before screening, 1171 records of ineligible document types were removed through database filters, including 106 Embase trial registry records, and 1627 duplicates were removed. The remaining 3378 records were screened by title and abstract, and 3291 were excluded. Of these, 260 addressed treatment in patients or self-care use and 2288 were unrelated to acupuncture education. A further 99 described no XR training system, 642 were ineligible publication types, and 2 were duplicates identified during screening.

Of the 87 reports sought for retrieval, 7 could not be obtained. Of the 80 reports assessed in full text, 55 were excluded. Reasons were a nonimmersive system (n=30), no educational purpose (n=9), no implemented system or outcome, or a conference abstract (n=15), and an ineligible publication type (n=1).

Citation searching identified 647 records. These records were screened by title and abstract against the database search results and the same eligibility criteria; 646 were excluded because they duplicated a database record or did not describe an XR training system for acupuncture, and the remaining record was assessed in full text and included [23].

In total, 26 studies were included in this scoping review, 25 from the database search and 1 from citation searching (Figure 1).

A geographical analysis of the 26 included studies showed that research conducted in the East Asian region accounted for 96.2% (n=25) of the total. By country, China produced the highest number of studies with 18 publications [4,23-39], followed by South Korea with 4 [10,40-42] and Taiwan with 2 [9,43]. A single study was reported from the United States [44], and one from Japan [45] (Figure 2). Six studies were published before 2020, in 2008 [45], 2013 [37], 2015 [44], 2017 [39], and 2019 [26,35], and the remaining 20 studies (76.9%) were published between 2020 and 2026 (Figure 3) [4,9,10,23-25,27-34,36,38,40-43].

XR systems can be distinguished into 3 types based on the level of immersion experienced by the user: NIVR, SIVR, and IVR [18-20]. This review focused on immersive and semi-immersive technologies that meet the criteria of high-fidelity spatial interaction and 1:1 physical mapping. Fully immersive systems (IVR, n=19), which use HMDs and 3D input devices (eg, VR controllers or data gloves) or markerless bare-hand pinch interactions with hand-tracking algorithms (eg, HoloLens 2) to completely immerse the user in the virtual environment, represented the majority of the included studies [9,10,23-27,30-35,37-41,45]. Semi-immersive systems (SIVR, n=7) typically consist of large screens for projection and advanced interface devices (eg, haptic feedback or infrared cameras) that allow users to interact with the immersive environment while remaining aware of the physical world, providing partial immersion and high presence [18-20]. This category included a life-size interactive touch table for which an AR module was planned but was not yet implemented [42], large-screen projection combined with student practice in a VR system [4], a stereoscopic display with 3D glasses and an interactive stylus [28], and a VR system described with stereoscopic glasses or a headset [29]. Other semi-immersive systems used a tracked life-size physical model with a large display [36], a tablet interface registered 1:1 to the head [43], or a monocular head-worn display (Google Glass) registered to the face through its camera [44]. By modality, 15 studies used VR, 5 used AR, and 6 used MR.

Educational tasks were categorized into 3 domains: acupoint localization (n=13), needling skills (n=11), and clinical reasoning (n=2).

First, acupoint localization (n=13) involved learning acupoint locations on 3D virtual bodies or by overlaying virtual data onto a real body or physical model. Regarding technical precision, Chen et al [27] reported a mean positional error of 2.55 mm for 16 facial acupoints overlaid through HoloLens 2 on 8 volunteers. Chiou et al [9] reported a mean error of 5.01 (SD 2.64) mm using a “Six-Point Landmark-Based AR Registration” algorithm that maps acupoints directly onto the patient’s body without external markers, which the authors considered clinically acceptable. Several systems allowed learners to view the 3D relationship between acupoints and underlying structures such as muscles, bones, nerves, and vessels [4,25,28,42]. In the study by Sun et al [30], an MR group scored higher than a textbook group on point-location items of a knowledge test given twice, 1 week apart (P=.03). A tablet-based AR system with 6-point registration showed discrepancies of 0.6 to 3.9 mm between virtual and actual acupoints on a mannequin head, with larger depth errors along the axis perpendicular to the screen [43]. These values are not directly comparable because the studies measured error of facial acupoint estimation [27], registration error on the body [9], and virtual-to-actual discrepancy on a tablet screen [43]. One system identified acupoints with a tracked pointer on a life-size copper model [36], and another supported acupoint selection in a head-mounted VR environment [35]; neither reported learner outcomes. Two further head-mounted VR systems let learners explore meridians and acupoints on a 3D body with hand controllers [25,26]. One reported only informal feedback from teachers and students [25], and the other reported no user data [26]. Two comparative studies assessed localization skills. In a randomized comparison of 70 vocational students, a VR group scored higher than a conventional group on acupoint localization (90.26, SD 3.11 vs 84.12, SD 3.98 points; P<.001) [29]. Compared with earlier student cohorts (98 VR-group vs 122 earlier-cohort students), a stereoscopic VR system was associated with higher practical examination scores, of which acupoint localization made up 70% (88.66, SD 6.82 vs 86.59, SD 8.41 points; P<.05) [28]. A wearable AR system on Google Glass mapped personalized acupoints and recorded complete treatment sessions, with an accuracy bias of 7.7 to 8.6 pixels against an electronic acupoint detector in three volunteers [44]. In a survey of 180 students using a head-mounted VR acupuncture system, 86% reported liking the system and 89% reported greater interest [34].

Second, needling skills (n=11) focused on training for precise control of needle depth, angle, and stimulation intensity. Recent studies have evolved beyond simple visual cues to enhance skills through “sensory feedback” and real-time collision detection. In the study by Guruge et al [10], vibrotactile feedback that intensified with insertion depth was active in both practice sessions, and mean depth error decreased from 6.41 mm in the first session to 3.58 mm in the second. Experts scored higher than novices (69.25, SD 21.64 vs 56.26, SD 23.37 points), which the authors interpreted as support for construct validity. Yan et al [24] integrated a haptic device that replicated tissue resistance while learners needled back acupoints selected from case sheets; no learner evaluation was reported. Sun et al [30] implemented a multilayer collider algorithm to gauge insertion depth and angle in real time; across 2 trials, task time and the number of mistakes decreased, whereas angle and depth accuracy did not change significantly (P=.41 and P=.06). Early research by Kanehira et al [45] combined a head-mounted display with a magnetically tracked training device to judge the position, depth, and angle of a virtual needle in real time, without participant testing. Xu and Li [4] described a teaching reform in which an instructor’s headset view of high-risk acupoints was projected to the class before students practiced needling in a VR system with automatic assessment; no comparison group was reported. Most recently, a randomized comparison by Wu et al [31] found that an IVR system (PainEaseVR) was associated with higher posttraining practical-skills scores than conventional instruction (mean 32.52, SD 2.28 vs mean 28.62, SD 2.77 on an 18-item checklist; t₅₈=5.947; P<.001). In a randomized comparison of 70 students, a group taught in a shared head-mounted VR scene outscored a conventional group [32]. Scores were 86.84, SD 5.04 vs 78.26, SD 7.11 points for theory and 91.37, SD 3.16 vs 85.23, SD 4.21 points for operational skills (both P<.001). An MR system that combined ultrasonic computed tomography with HoloLens 2 guidance was tested with 31 participants [38]. It received higher System Usability Scale scores and lower NASA Task Load Index mental demand than a two-ring interface. Hand input was also captured with a data glove and 3D tracker, tested informally with 12 students [37], or with machine learning gesture recognition that reached about 92.65% average accuracy in the VR application [39].

Third, clinical reasoning (n=2) evaluated integrated decision-making, such as syndrome differentiation and treatment selection, through case-based scenarios. According to Jiang et al [23], the group taught with VR achieved a mean case analysis score of 93.88 (SD 4.20) compared with 84.65 (SD 4.68) in the traditional group (P=.001). In a survey of 79 students after an HTC Vive Pro acupuncture anesthesia scenario, 99% reported that the experience helped their clinical thinking [33].

The reviewed studies utilized a wide range of hardware to construct educational systems, from high-performance HMDs (HTC Vive, HoloLens 2, and Magic Leap One) to tablets, stereoscopic monitors, and specialized UHD multitouch tables (Table 1). To enhance tactile realism or optimize rendering performance on standalone devices, one study integrated a custom ESP32-based wearable haptic unit [10], and another applied polygon retopology to complex 3D anatomical models [30]. Despite these advancements, several barriers to implementation in actual educational settings were identified.

First, regarding physical constraints, learners reported that headsets were heavy or uncomfortable during prolonged use [25,33], and the stereoscopic glasses in one system limited the field of view and allowed only 1 learner per workstation [28]. Furthermore, optical and tracking constraints in standalone headsets, including a narrow field of view, limited display space, and reduced tracking precision during low-range spatial interactions, were noted as potential obstacles to continuous needle manipulation and spatial orientation [30]. Some virtual models also lacked internal anatomical detail, such as muscles and bones [26]. Second, concerning technical limitations, most current haptic and sensory feedback mechanisms, which primarily rely on vibrotactile cues or simple collision algorithms, fall short of replicating the complex physical resistance experienced during actual tissue penetration, thereby limiting the realistic simulation of the deqi sensation [10,30]. Third, the high cost of acquiring 3D scanners and AR glasses was reported as a barrier to large-scale institutional adoption [9]. Other reported constraints were specific to individual systems. In a VR acupuncture anesthesia project, students reported weak realism of hand needling without force feedback, and only one clinical case was available [33]. Tablet-based AR required manual alignment and showed depth error along the screen axis [43], and the ultrasonic computed tomography system required prior scanning and optical marker tracking [38].

To overcome these barriers, future research directions emphasize improving portability by offloading computation to a paired smartphone [44], incorporating realistic tissue resistance gradients or pseudo haptic feedback beyond simple vibration [10,30], and developing data-driven algorithms that dynamically adjust for individual anatomical variations such as body mass and gender differences in acupoint mapping [30]. Furthermore, developing level-based customized programs [4] and professional validation by expert groups [42] were proposed as next steps for XR-based acupuncture education. The distribution of research and the XR technologies applied across each educational domain are visualized in the evidence gap map (Figure 4).

The distribution of XR modalities differed by domain. As shown in the evidence gap map (Figure 4), AR research was confined to acupoint localization (n=5), and none of the 5 studies used a comparative design. MR studies centered on needling (4 of 6), whereas VR studies addressed localization (n=6), needling (n=7), and clinical reasoning (n=2). Both technologies produced comparative (tier 1) evidence for needling skill training, in VR studies [31,32] and MR studies [30,38], whereas no AR study addressed needling. Among studies assigned to localization, comparative evidence came only from VR studies [28,29], although an MR needling study also compared point-location scores with textbook learning [30]. Clinical reasoning was examined only within VR [23,33]. Overall, 7 of the 26 included studies used a comparative or experimental design (tier 1). The remaining 19 studies comprised technical validation (tier 2, n=6), system or hardware development (tier 3, n=10), and case, survey, or qualitative reports (tier 4, n=3).

These results address the 3 research questions on the range of XR technologies, their educational objectives, and the evidence on technical accuracy, educational efficacy, and implementation barriers.

This scoping review mapped immersive and semi-immersive XR training systems for acupuncture education published between 2008 and 2026. AR systems were used only for acupoint localization, MR systems mainly for needling, and VR systems for localization, needling, and clinical reasoning. Most studies reported technical or usability outcomes, and only 7 of 26 studies used a comparative design, so the educational benefits of these systems remain preliminary.

This study specifically focuses on immersive XR technologies, excluding nonimmersive desktop-based simulations. Historically, numerous studies have explored computer-based virtual acupuncture systems using standard monitors with a mouse, a camera, or desktop haptic input [46-48]. While these desktop applications contributed to visualizing meridians and acupoints [49,50], they lack the spatial immersion required for a realistic “first-person” training experience. We prioritized immersive technologies because acupuncture training depends on an understanding of 3D spatial relationships between surface acupoints and internal anatomical structures, rather than 2D visual memorization [10]. A recent meta-analysis reported that VR-based education was associated with higher knowledge and skill scores than traditional methods [11].

The included literature showed different distributions of XR modalities across educational tasks; these patterns should not be interpreted as evidence of modality-specific superiority. AR studies were concentrated on surface localization, overlaying acupoint information onto the user’s real-world view without external markers [9,44], although this focus remained on surface-level navigation rather than complex procedural tasks (Figure 4), whereas the included VR and MR studies more often addressed needling depth, angle, or 3D anatomical relationships. By rendering skin semitransparent or displaying tissue layers during virtual needling [28,45], these systems allow students to intuitively grasp insertion depth through visual or haptic cues [10]. Across XR modalities, a further distinction exists between IVR and SIVR systems. While both meet this review’s criteria for 1:1 spatial mapping, full visual immersion applies primarily to IVR, whereas the SIVR systems in this review use large displays, stereoscopic monitors, touch tables, tablets, or a monocular head-worn display that keep the physical surroundings in view [4,28,36,42-44].

The gap map (Figure 4) highlights that while VR is a versatile platform, no MR study addressed clinical reasoning, representing an important direction for future research. At the same time, the apparent pedagogical advantages of specific XR modalities should be interpreted in light of the uneven methodological rigor across studies. Only 7 of the 26 included studies used a comparative or experimental design. Most instead reported technical outcomes, such as positional accuracy or usability, rather than direct measures of learning. This imbalance was most evident for AR, where none of the 5 studies used a comparative design. Given this concentration of noncomparative designs, claims about the relative pedagogical advantage of any single XR modality should be regarded as preliminary until confirmed by larger, adequately powered comparative trials. Nonetheless, the observed division of labor across modalities may reflect a practical distinction in task demands: AR’s concentration on acupoint localization addresses a primarily visual-spatial task, for which overlay-based guidance may be particularly useful, whereas VR and MR’s focus on needling depth and angle addresses a psychomotor task that typically requires repeated, feedback-driven practice to refine. This distribution also resembles reviews of head-mounted XR in surgical [12] and medical [13] education, which have mainly described procedural and anatomical training rather than higher-order clinical reasoning. Clinical reasoning instead requires integrative, case-based judgment that has traditionally been supported by nonimmersive virtual patient formats; the absence of MR studies in this domain may, therefore, reflect a weaker match between immersive spatial interfaces and the integrative, case-based nature of diagnostic reasoning, rather than an inherent technical shortfall of MR itself.

Immersive XR may also address two limitations of traditional training, safety and standardization. VR and MR may allow rehearsal of high-risk needling without needle insertion in human participants, potentially reducing exposure to iatrogenic injuries such as pneumothorax [5]. Unlike an earlier monitor-based system that overlaid acupoints on a camera image of the learner’s hand [47], immersive XR can capture quantitative data, including positional accuracy, insertion depth and angle, and procedure time, in 3D environments [10,30], which may support standardized evaluations such as the Objective Structured Clinical Examination, although safety and assessment outcomes were not evaluated in the included studies. In one comparative study, learners rated the VR system highly for interest and interactivity, although satisfaction was surveyed only in the VR group [28]. For educators, XR systems may offer a platform that helps bridge theoretical knowledge and clinical practice. These technologies may facilitate active engagement through 3D anatomical interaction and immediate feedback. Although these features are promising, implementation barriers remain, including high equipment costs [9] and the need to match XR content to learners’ prior anatomical knowledge [4]. Therefore, future efforts should focus on embedding XR within competency-based education to support standardized training and objective assessment.

Despite these advancements, several challenges remain. The most notable is the lack of realistic haptic feedback. Although several desktop-based systems have implemented force-feedback using mechanical haptic arms [51,52], integrating such sophisticated haptics into wireless, immersive HMD environments remains technically difficult. Most included systems relied on visual cues or basic vibrotactile responses [10,30], and only 2 used force-feedback devices [24,28]; these approaches cannot fully replicate the deqi sensation or tissue resistance found in physical simulators. Haptic interfaces compatible with head-mounted XR are needed to close this gap. Accessibility is another concern, as advanced markerless registration often requires expensive hardware like 3D scanners [9]. Moreover, most models are currently restricted to specific body parts; whole-body models would be needed to cover the full meridian system. Including conference proceedings captured recent XR developments, but their variable peer review may have admitted less rigorously validated systems, which would tend to overstate technical maturity. The 26 included studies were concentrated in East Asia (25/26), so the map may not cover XR training developed elsewhere or indexed in databases that were not searched, including Japanese databases such as J-STAGE. Finally, formal risk-of-bias appraisal was not performed, consistent with the mapping objective of a scoping review; the evidence tiers therefore describe study design and should not be interpreted as ratings of methodological quality. Objective technical performance metrics, such as localization and depth error margins, were synthesized to characterize technical maturity, but these metrics do not substitute for formal risk-of-bias appraisal. Classification of immersion depended on how authors described their hardware.

This scoping review identified 26 studies that applied IVR, SIVR, AR, and MR training systems to acupuncture education, addressing acupoint localization (n=13), needling skills (n=11), and clinical reasoning (n=2). AR systems were used only for localization, MR systems mainly for needling, and VR systems across all three domains; no AR or MR study addressed clinical reasoning. Unlike earlier reviews of physical phantom models or of XR in general health professions education, this review maps immersive systems by modality, level of immersion, and educational task. Only 7 studies used a comparative design, and most reported technical or usability outcomes, so the educational benefits of these systems remain preliminary. Randomized trials comparing XR-based and conventional training on objective localization and needling outcomes, with retention follow-up, are needed before these systems are adopted for competency assessment. Trials comparing XR modalities on the same task would then show which modality suits which training objective.