Emerging technologies in interventional pain management: a scoping review on current innovations and future directions

Abstract

Background

Emerging digital and bioelectronic technologies are rapidly transforming interventional pain management, but their clinical roles and evidence base remain unclear. A scoping review of these innovations is needed to map the current landscape and identify gaps for future research.

Objective

To conduct a scoping review of emerging technologies applied to interventional pain management, and to summarize their clinical applications, outcomes, and methodological limitations.

Methods

This scoping review followed the PRISMA-ScR guidelines and was structured according to the Population–Concept–Context (PCC) framework. A comprehensive search was performed in PubMed, Scopus, Web of Science, Embase, and CENTRAL from inception to January 1, 2026, using predefined search strategies targeting emerging technologies and interventional pain procedures. Titles, abstracts, and full texts were screened in a three-stage process, supplemented by backward citation searching. Data were charted on study and patient characteristics, technology type, interventional pain application, and reported outcomes. Methodological quality and risk of bias were appraised using Joanna Briggs Institute (JBI) tools appropriate to each study design.

Results

Four major themes emerged from the reviewed literature. (1) Precision patient selection and intelligent clinical decision support using multimodal artificial intelligence models, (2) physiologic closed‑loop neuromodulation and neural dosing guided by evoked compound action potentials, (3) AI‑assisted procedural guidance for ultrasound‑guided interventions, and (4) digital therapeutics, wearable technologies, and remote monitoring platforms supporting long‑term management. Across these domains, early evidence suggests that AI‑driven tools and data‑integrated neuromodulation systems may improve patient selection, procedural accuracy, therapy optimization, and longitudinal monitoring in interventional pain care. However, most studies remain limited by small sample sizes, heterogeneous outcome measures, and insufficient external validation.

Conclusions

Emerging technologies demonstrate potential for precision, personalization, and effectiveness in interventional pain management. Nevertheless, additional multicenter, longitudinal research with standardized outcomes is essential before extensive clinical implementation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12871-026-04017-1.

Introduction

Pain constitutes a complex sensory and emotional experience that significantly impacts quality of life (QoL) and functional capacity. The subjective nature of pain and the heterogeneity of patient responses constrain traditional management approaches [1]. In critically ill populations, such as ICU patients, uncontrolled pain induces sympathetic activation and a hypermetabolic stress response, which have downstream effects on glycaemia, immunity, wound healing, and inflammation. This underscores the importance of comprehensive assessment utilizing numeric and behavioral scales, in addition to nociception-monitoring devices, and advocates for multimodal pharmacologic and nonpharmacologic management. Within this context, real-time monitoring and Artificial Intelligence (AI)-driven tools may enhance detection capabilities and facilitate individualized care [2–4].

Using data from 146 countries and 510,247 respondents, a study examined whether global pain levels changed during the COVID-19 pandemic. Analyses showed that 33.3% of people reported pain in 2019, compared with 32.8% in 2020, 32.5% in 2021, and 34.1% in 2022. None of these year-to-year differences were statistically significant, indicating that the pandemic did not meaningfully alter overall pain prevalence worldwide. However, pain significantly declined from 2019 to 2020 among adults over age 55, individuals who were widowed, and those without children in the household. Overall, the pre-pandemic pattern of elevated global pain persisted throughout the pandemic period [5]. In Iran, recent cross-sectional studies similarly document a high prevalence, for instance, reporting that 23.7% of adolescents endure chronic pain [6].

Recent advancements in minimally invasive methods and digital therapies, including extended reality (XR), offer promising alternatives to traditional healthcare. The incorporation of AI shows significant potential for precision pain medicine through machine learning, natural language processing, computer vision, and expert systems to identify biomarkers, predict treatment outcomes, and enhance personalized treatment plans [1, 7].

AI applications in pain management mitigate the limitations of self-report by providing objective, continuous, and multimodal assessments through algorithms trained on clinical, imaging, physiological, and patient-reported data. These tools can categorize pain types, monitor progression, and help predict treatment outcomes, thereby facilitating more standardized and personalized therapeutic interventions [8]. Parallel advancements in AI-assisted medical imaging exemplify the automated classification of nodules and lesions across ultrasound, computed tomography (CT), and magnetic resonance imaging (MRI). Such developments diminish dependence on operator expertise and support real-time clinical decision-making. They are particularly relevant to interventional pain procedures, where imaging accuracy is of utmost importance [9].

The integration of AI with remote patient monitoring (RPM) enhances these functionalities: wearable and implantable sensors that feed data into AI algorithms enable continuous surveillance, facilitate early detection of therapy tolerance or failure, and support dynamic adjustments of neuromodulation or pharmacologic regimens. These advancements may improve adherence, increase accessibility, and lead to better long-term outcomes for patients with chronic or complex pain [4, 10, 11].

Interventional pain management has made rapid progress due to advances in minimally invasive procedures, neuromodulation, augmented reality, AI, and RPM. These pioneering technologies possess the potential to enable more precise, adaptable, and personalized interventions, thereby improving the efficacy and safety of treatments for both acute and chronic pain. Despite the growing interest, existing evidence remains fragmented and heterogeneous, frequently limited to small-scale, single-center studies with short follow-up periods. There is a significant lack of comprehensive data concerning the comparative effectiveness, long-term outcomes, and optimal patient selection criteria for these innovative interventions. Moreover, the integration of digital health tools, predictive analytics, and AI into interventional pain care has not been systematically evaluated, resulting in gaps in understanding their practical clinical utility. This scoping review seeks to synthesize current literature on emerging technologies in interventional pain management, identify gaps in the evidence and research methodologies, and propose a framework for future investigations and clinical application.

Method

A scoping review is appropriate when the existing body of evidence is extensive, heterogeneous, and conceptually complex. It is also applicable when the objective is to inform a subsequent systematic review or when methodological limitations preclude conducting one [12]. The primary purpose of a scoping review is to map the extent and detail of the existing literature systematically and to identify knowledge gaps, thereby guiding future research [13]. Furthermore, this study is documented in accordance with the PRISMA extension for scoping reviews (PRISMA-ScR). (Supplementary Material 1) [14]. Given the rapidly expanding and diverse body of literature on emerging technologies in interventional pain management and the study’s primary objective of examining current innovations and future developments, a scoping review was deemed the most appropriate methodological approach.

A review protocol was not prospectively registered in a public registry such as the Open Science Framework (OSF), as scoping reviews are not eligible for registration on PROSPERO. The lack of prospective protocol registration is acknowledged as a methodological limitation; however, the eligibility criteria, search strategy, and data charting framework were defined a priori by the review team and remained unchanged throughout the conduct of the review.

Search strategy

A comprehensive literature search was conducted across multiple electronic databases, including PubMed, Scopus, Web of Science, Embase, and the Cochrane Library (CENTRAL), to identify studies on emerging technologies in interventional pain management. The search strategy combined controlled vocabulary terms (MeSH in PubMed, Emtree in Embase), free-text keywords, and input from subject experts to ensure comprehensive coverage of relevant concepts. Keywords related to pain management interventions (pain management, interventional pain management, nerve block, neuromodulation, spinal cord stimulation) were combined with terms representing emerging and innovative technologies (emerging technologies, digital technologies, artificial intelligence, telemedicine, machine intelligence). Boolean operators (AND/OR) were used to combine search terms, and truncation (*) was applied to capture variations of key concepts. The search was conducted in the database from its inception through January 1, 2026, with no language restrictions. The search strategy was adapted as needed for each database and is available in full in Supplementary Material 2. The search strategy was designed to ensure an optimal balance between sensitivity and specificity. To improve the precision and manageability of the retrieved literature, search terms related to both interventional pain management and emerging technologies were primarily applied to article titles. This title-focused approach was intended to identify studies in which the technological aspect constituted a central focus rather than a peripheral mention within the abstract or keywords. This strategy also yielded an acceptable Number Needed to Read (NNR), thereby maintaining efficiency while minimizing the exclusion of relevant literature. To further enhance completeness, we manually screened the reference lists of all included articles and relevant reviews to identify additional eligible studies. To ensure comprehensive coverage, review articles retrieved during the initial search were used solely for backward citation searching. The reference lists of these reviews were manually screened by two independent reviewers to identify additional eligible primary studies that met the inclusion criteria. Disagreements were resolved through discussion. This process yielded 1 additional study, which was incorporated into the final selection.

Inclusion and exclusion criteria

The inclusion and exclusion criteria for this scoping review were delineated in accordance with the PCC framework.

Population (P)

Studies involving patients receiving interventional pain management, including nerve blocks, spinal cord stimulation, intrathecal drug delivery, or neuromodulation, were included, whereas studies focusing exclusively on animal models or non-human subjects were excluded.

Concept (C)

In principle, studies investigating emerging or innovative technologies utilized in interventional pain management, such as AI, telemedicine, machine learning, virtual medicine, and other digital or computational technologies, were incorporated. Conversely, studies that focused exclusively on traditional pain management methods without a technological component were excluded.

Context (C)

Eligible studies encompassed those conducted within any clinical, experimental, or observational environment where interventional pain management was provided. Conversely, studies lacking practical application or details of implementation, such as purely theoretical or opinion-based papers, were considered ineligible.

Additional criteria included the absence of language restrictions, the inclusion of studies published in the database from its inception up to January 1, 2026, and the consideration of peer-reviewed articles, conference proceedings, as well as relevant grey literature. Review articles were employed solely for reference mining and not as primary data sources.

Study selection

The selection process for the studies included in this scoping review was conducted over three successive rounds. During the initial round, all retrieved records were independently screened by two reviewers (R.N. and G.V.), who evaluated titles and abstracts in a blinded manner against the predefined eligibility criteria. Each reviewer independently classified records as “include,” “exclude,” or “uncertain.” Discrepancies between the two reviewers were first resolved through discussion, and any remaining disagreements were adjudicated by a third senior reviewer (M.R.F.). Inter-reviewer agreement was monitored throughout the screening process.

In the second phase, the same two reviewers (R.N. and G.V.) independently performed full-text screening of all potentially relevant studies. Decisions regarding inclusion or exclusion were made in accordance with the predefined eligibility criteria, with reasons for exclusion documented. Any disagreements were resolved through consensus, and unresolved cases were referred to the third reviewer (M.R.F.) for final adjudication.

During the third phase, backward citation searching was performed by manually reviewing the reference lists of all included primary studies as well as relevant review articles identified during database searching. Citations identified through this process were imported into the same screening workflow and assessed against the eligibility criteria by two reviewers (R.N. and G.V.). One additional report was identified through this process and underwent full-text assessment but was ultimately excluded due to inappropriate study design or outcome; therefore, no additional studies were added to the final sample through citation searching.

Risk of bias assessments

The methodological quality of the included studies was assessed utilizing the Joanna Briggs Institute (JBI) Critical Appraisal Tools, appropriate for each study design. The evaluation was performed using the relevant JBI checklists for cohort studies, cross-sectional studies, case series, case reports, and diagnostic test accuracy studies. Each item on the checklist was rated as “Yes,” “No,” “Unclear,” or “Not applicable” based on the information reported in each study [15, 16].

Although critical appraisal is not a mandatory component of scoping reviews according to JBI methodology and the PRISMA-ScR statement, we elected to perform an optional critical appraisal of included sources to provide readers with contextual information regarding the methodological quality of the underlying evidence base in this rapidly evolving field. Importantly, the appraisal results were used descriptively to characterize the available evidence and were not used as criteria for inclusion or exclusion, nor were they used to weight findings in the synthesis. All eleven studies meeting the eligibility criteria were retained in the review regardless of appraisal score, in line with the mapping function of a scoping review.

A total of eleven studies were evaluated. The appraisal outcomes demonstrated that the overall methodological quality varied from moderate to high. Based on the predefined scoring criteria, studies achieving a score of 80% or above were classified as high quality, those scoring between 50% and 79% were deemed of moderate quality, and scores below 50% were regarded as indicative of low quality.

Data abstraction and collation

Data from the included studies were extracted using an analytical-descriptive approach as outlined by Arksey and O’Malley (2005) [13]. This process entailed a systematic review of each study to compile essential information, including authorship, country, study sample size, intervention type, technology details, and measurement outcomes. The extracted data were compiled into a structured table to enable comparison and to evaluate how each study addressed the objectives of the scoping review.

Following data extraction, the studies were systematically organized thematically. An initial thematic analysis was conducted to identify recurring patterns and emerging trends, which were then reviewed and validated by the research team. The proposed themes were discussed collaboratively to ensure consensus and confirm their relevance to the review’s overall objectives.

Results

Study Selection

As illustrated in the study selection flow diagram (Fig. 1), a total of 740 records were identified through database searching, including MEDLINE (n = 58), Web of Science (n = 265), Scopus (n = 195), Embase (n = 99), and CENTRAL (n = 123). After removal of 254 duplicate records, 486 unique records remained and were screened based on titles and abstracts. Of these, 455 records were excluded, including 159 that were not relevant to the research question and 296 that were letters, reviews, or other ineligible publication types. Consequently, 31 reports were sought for full-text retrieval; however, one report could not be retrieved. Thirty full-text articles were assessed for eligibility. In addition, one additional report was identified through other sources and evaluated, but it was excluded due to an inappropriate study design or outcome. Of the 30 full-text articles evaluated, 19 were excluded for the following reasons: inappropriate population (n = 7), inappropriate outcomes (n = 7), or inappropriate study design (n = 5). Ultimately, 11 studies met the inclusion criteria and were included in the final qualitative synthesis [17–28].

Fig. 1.

PRISMA Flowchart

Study characteristics

Table 1 summarizes the characteristics of the 11 included studies. Regarding study design, three studies were prospective observational or clinical investigations; one was a retrospective real-world cohort; two were cross-sectional questionnaire-based evaluations; one was a case series; one was a validation study; one was a post hoc pooled analysis; one was a multicenter predictive cohort; and one was a development and performance evaluation study. The studies originated from the United States (n = 3), Italy (n = 2), Turkey (n = 2), and Spain, Australia, France, and China (n = 1 each). Six studies focused on patients with chronic pain conditions, including failed back surgery syndrome and persistent spinal pain syndrome; one involved epilepsy patients using an EEG database; one examined surgical patients undergoing regional anesthesia; one included healthy volunteers; and two assessed healthcare professionals or lay evaluators without patient enrollment. AI applications encompassed machine-learning–based prediction models for SCS outcomes (n = 3), closed-loop SCS systems using electrophysiological feedback (n = 2), AI-assisted ultrasound guidance for regional anesthesia (n = 2), LLMs for patient education and information delivery (n = 2), a wearable or app-based adaptive neuromodulation system (n = 1), and an experimental EEG-based closed-loop neuromodulation platform (n = 1). Reported outcomes spanned pain intensity (n = 6), functional disability (n = 4), health-related QoL (n = 4), predictive accuracy or classification performance (n = 4), device or system performance indicators (n = 4), and user-rated reliability or comprehensibility (n = 2), reflecting substantial clinical and methodological heterogeneity.

Table 1.

Basic characteristics of the included studies in this review

First Author/year	Study characteristics Design, n = [study size], mean age:, and country:	Participants	Measured Outcomes (measurement tools)	Intervention / Technology details	Comparison
Andres et al. 2021 [18]	Prospective, consecutive, observational, single-center pilot study, 24, 48 ± 8, Spain	Adults with Failed Back Surgery Syndrome (FBSS)	Pain intensity (NRS, PD-Q); psychological variables (HADS); resilience (CD-RISC10); sleep/social support (MOS-SS); patient satisfaction; adverse events; imaging biomarkers (rs-fMRI functional connectivity, brain volumetry); ML model performance (accuracy, sensitivity, specificity)	Spinal Cord Stimulation (tonic SCS using Avista™ 8-pole electrodes and Montage™ generator); AI-based Clinical Decision Support System (CDSS) integrating rs-fMRI imaging biomarkers and clinical data using machine learning algorithms (LDA, SVM, LR, KNN, DT, GNB) to predict SCS response	NA
Bianco et al. 2025 [27]	Cross-sectional, observational, questionnaire-based evaluation study, 41, NA, Italy	Evaluators assessing AI-generated responses: pain physicians, healthcare professionals, and lay individuals; no patient enrollment	Reliability (6-point Likert scale); Accuracy (3-point Likert scale); Comprehensibility (3-point Likert scale); predefined acceptability thresholds (Reliability ≥ 4, Accuracy ≥ 2, Comprehensibility ≥ 3)	Artificial intelligence–powered chatbot (ChatGPT-4.0) generating responses to 13 frequently asked patient questions on radiofrequency ablation (RFA) and cryoanalgesia; evaluation of AI-generated patient education content across pre-, intra-, and post-procedural domains.	NA
Bianco et al. 2025 [26]	Cross-sectional, observational, questionnaire-based evaluation study of AI-generated content, 12, NA, Italy	Evaluators (pain physicians and non-healthcare professionals); no patient involvement	Reliability (6-point Likert scale); Accuracy (3-point Likert scale); Comprehensibility (3-point Likert scale); word count analysis of AI-generated responses	AI-based chatbot (ChatGPT-4.0) generating responses to 13 common patient questions regarding spinal cord stimulation (pre-, intra-, and post-procedural domains)	NA
Brooker et al. 2021 [17]	Prospective, multicenter, single-arm, open-label clinical study, Implanted patients = 50; 24-month completers = 38, 56.7 ± 12.2, Australia	Adults with chronic back and/or leg pain	Pain intensity (VAS); pain impact (Brief Pain Inventory); functional disability (Oswestry Disability Index); quality of life (EQ-5D-5 L); sleep quality (Pittsburgh Sleep Quality Index); opioid consumption (MME/day); responder rates (≥ 50% and ≥ 80% pain reduction); ECAP amplitude data; reprogramming frequency; adverse events	ECAP-controlled closed-loop spinal cord stimulation using the Evoke® system (Saluda Medical); real-time feedback adjusting stimulation to maintain target ECAP amplitude.	NA
Erdem et al. 2023 [23]	Case report/case series (3 cases); descriptive, observational study, 3, 41.3, Turkey	Adult patients (ASA II) undergoing upper extremity or breast surgery	Block success (clinical adequacy); number of attempts; block performance time; complications (vascular puncture, pneumothorax, LAST); postoperative pain (VAS at 0, 2, 6, 12, 24 h)	AI-powered ultrasound–guided regional nerve block using Nerveblox software; automated identification and labeling of anatomical structures during infraclavicular and PECS I–II blocks	NA
Green et al. 2025 [21]	Retrospective, observational, real-world cohort study, Total users = 2135 adults; total sessions = 187,930, 58.7 ± 15.6 years, USA	Adults (≥ 18 years) with self-reported chronic pain of various etiologies and anatomical sites	Primary: pain intensity (0–10 numeric rating scale, weekly via mobile app); Secondary: mood, physical function, social engagement, work activity, overall well-being (all 0–10 numeric scales); device engagement metrics (session frequency, duration, waveform, intensity); responder rates; usage patterns.	AI-software–dependent neuromodulation (ASDN) using EcoAI™: wearable, FDA-cleared system delivering adaptive TENS and EMS with embedded AI engine dynamically adjusting stimulation parameters (intensity, waveform, duration) based on patient-reported outcomes, physiological markers, and prior session data; remote patient monitoring via cloud-based platform.	NA
Gungor et al. 2021 [22]	Prospective accuracy/validation study, 40, 32.2 ± 5.25, Turkey	Healthy adults (20 male, 20 female), BMI 22.2 ± 3; no underlying disease;	Accuracy of AI software (Nerveblox) for real-time identification of anatomical landmarks during ultrasound-guided peripheral nerve blocks (UGPNB); validators (anesthesiologist + radiologist) scored labeled images on a 5-point scale for multiple blocks (interscalene, supraclavicular, infraclavicular, TAP); kappa statistics for inter-rater reliability; correlation with age/BMI	AI-assisted real-time ultrasound anatomy identification software (Nerveblox) for UGPNB; color-labeled overlays and text tags highlight predefined anatomical landmarks; scan success is measured as % landmarks identified (images taken at 100% scan success).	NA
Hadanny et al. 2021 [23]	Prospective cohort, 151, 54.8 ± 12.0, USA	Patients with chronic pain	Pain outcomes: NRS, PCS, MPQ, ODI, BDI	Machine learning–assisted prediction models for long-term spinal cord stimulation (SCS) response; 31 preoperative features including demographics, pain scores (NRS, PCS, MPQ, ODI, BDI), and psychological assessments; unsupervised clustering (K-means) to identify patient phenotypes; supervised classification using logistic regression, Random Forest, and XGBoost; nested 10-fold cross-validation for model training and internal validation; top 10 most influential features used for final prediction; outcomes predicted as probability of responder or high responder at 1-year follow-up.	NA
Levy et al. 2024 [24]	Post-hoc analysis of 3 clinical studies (EVOKE RCT, ECAP, Durability), 180 (EVOKE n = 56; ECAP n = 103; Durability n = 21), 59 ± 12, USA	Patients with chronic, intractable back and leg pain refractory to conservative therapy;	Pain intensity (VAS), physical function (ODI), health-related quality of life (EQ-5D-5 L, PROMIS-10), sleep quality (PSQI), emotional function (POMS), cumulative responder score (MCIDs across five core domains)	Closed-loop spinal cord stimulation (Evoke® SCS System, Saluda Medical) with ECAP-controlled feedback: - Automatic adjustment of stimulation current to maintain target ECAP - Neurophysiologic metrics: system utilization, percent time above ECAP threshold, ECAP dose, dose ratio, dose accuracy (RMSE).	Traditional fixed-output SCS (trial-and-error programming) and temporary SCS trial phase.
Li et al. 2024 [25]	Development and evaluation of a brain-machine interactive neuromodulation research tool (BMINT), 10, NA, China	Epilepsy patients (CHB-MIT Scalp EEG Database)	Seizure detection performance: Accuracy, Sensitivity, Specificity, AUC; Real-time system performance: time delay (ms), gamma oscillation tracking correlation	BMINT research tool integrating: - 8-channel neural recording module (24-bit, 2000 Hz) - Edge AI computing module (Jetson Nano, CPU + GPU) - Stimulation module (2-channel, adjustable biphasic pulses, max ± 10 mA, ± 14 V) - Intelligent computing framework supporting SVM, CNN, RNN; real-time neural decoding and stimulation - Closed-loop adaptive stimulation based on single-channel EEG seizure detection.	NA
Ounajim et al. 2021 [28]	Multicenter predictive study using machine learning vs. lead screening trial, Training: 91 / Testing: 12, 47.7 (9.5), France	Patients with persistent spinal pain syndrome type 2 (PSPS-T2), post-spinal surgery, refractory to conservative therapy, are eligible for SCS.	Primary Outcome: SCS efficacy at 12 months measured using the composite Global Health Improvement Score (GHIS), derived from: - VAS (Visual Analogue Scale) - EQ-5D-3 L - Oswestry Disability Index (ODI) - Montgomery–Asberg Depression Rating Scale (MADRS)	Machine learning models for the prediction of SCS response: - Logistic Regression (LR) - Regularized Logistic Regression (RLR, ridge) - Support Vector Machine (SVM, RBF kernel) - Naive Bayes (NB) - Artificial Neural Networks (ANN, two hidden layers × eight nodes) - Classification and Regression Tree (CART) - Random Forest (RF) - Gradient-Boosted Trees (GBT) Training data: ESTIMET study (n = 91) External validation: AIVOC study (n = 12).	Traditional 5–10-day lead screening trial. Machine learning predictions compared against actual trial outcomes; ML models (RLR, SVM, GBT, RF) generally outperformed the screening trial in predicting 12-month response.

Risk of bias assessments

The methodological quality of the included studies was assessed utilizing the Joanna Briggs Institute (JBI) Critical Appraisal Tools in accordance with their respective study designs. A total of 11 studies were evaluated, comprising cohort studies, cross-sectional studies, a case series, a case report, and a diagnostic study.

Overall, the quality of the studies ranged from moderate to high. Two studies were rated as high quality, while the remaining studies were classified as moderate quality. The diagnostic study by Gungor et al. (2021) achieved the highest score (9/10, 90%) [22], followed by the case report by Erdem et al. (2023) with a score of 7/8 (87.5%) [19].

Among the cohort studies, Andres et al. (2021), Brooker et al. (2021) [18], Green et al. (2025) [21], and Hadanny et al. (2021) [23] met 8 of 11 appraisal criteria (72.7%), indicating moderate methodological quality. Similarly, Levy et al. (2024) [24] and Ounajim et al. (2021) [28] satisfied 7 of 11 criteria (63.6%).

The case series by Li et al. (2024) [25] fulfilled 7 out of 10 criteria (70%), while the two cross-sectional studies conducted by Bianco et al. (2025) [26, 27] each met 5 of 8 criteria (62.5%), reflecting moderate quality. Importantly, none of the included studies were rated as low quality (< 50%) (Table 2).

Table 2.

Summary of JBI Critical Appraisal Results

Study ID	JBI Checklist	Yes	Total	Score (%)	Overall Appraisal	Inclusion Status
Erdem et al. 2023 [23]	Case Reports	7	8	87.5%	High Quality	Included
Andres et al. 2021 [18]	Cohort Studies	8	11	72.7%	Moderate Quality	Included
Brooker et al. 2021 [17]	Cohort Studies	8	11	72.7%	Moderate Quality	Included
Green et al. 2025 [21]	Cohort Studies	8	11	72.7%	Moderate Quality	Included
Gungor et al. 2021 [22]	Diagnostic Studies	9	10	90%	High Quality	Included
Hadanny et al. 2021 [23]	Cohort Studies	8	11	72.7%	Moderate Quality	Included
Li et al. 2024 [25]	Case Series	7	10	70%	Moderate Quality	Included
Bianco et al. 2025 [27]	Cross-Sectional	5	8	62.5%	Moderate Quality	Included
Bianco et al. 2025 [26]	Cross-Sectional	5	8	62.5%	Moderate Quality	Included
Levy et al. 2024 [24]	Cohort Studies	7	11	63.6%	Moderate Quality	Included
Ounajim et al. 2021 [28]	Cohort Studies	7	11	63.6%	Moderate Quality	Included

Judgments are based solely on reported information. “Unclear” indicates insufficient information to make a definitive judgment

Overall appraisal: ≥80% = Included; 50–79% = Moderate quality; <50% = Low quality. All assessments follow the official JBI Critical Appraisal Tools guidance (JBI, 2020)

Note on heterogeneity and synthesis structure

The eleven included studies are highly heterogeneous in design, populations, interventions, and outcome measures. For this reason, results are not pooled and should not be interpreted as directly comparable across studies or themes. Within each theme below, the synthesis is structured to first describe what the included studies actually reported (Evidence from included studies), then summarize the methodological gaps the studies themselves expose (Identified gaps), and finally to clearly label any forward-looking discussion as authors’ inferences rather than as findings of the included studies.

Theme 1: precision patient selection and intelligent clinical support

Recent evidence suggests that predicting patient responses to spinal cord stimulation (SCS) using data-driven methods is feasible. The integration of multimodal data, including clinical assessments, psychometric evaluations, structural and functional neuroimaging, and behavioral and wearable data, substantially improves predictive accuracy. A small imaging study on 24 participants conducted at a single center demonstrated that combining imaging biomarkers, such as functional connectivity and regional brain volumes, with clinical variables increased predictive accuracy from approximately 71–75% to 95.8% using Linear Discriminant Analysis (LDA). However, the findings were constrained by the limited sample size and the application of leave-one-out cross-validation [18]. Larger cohorts employing patient clustering followed by classification models, including logistic regression (LR), regularized LR (RLR), support vector machines (SVM), random forest (RF), and gradient-boosted trees (GBT/XGBoost), attained area under the receiver operating characteristic curves (AUCs) ranging from approximately 0.65 to 0.83. The cluster-specific logistic regression achieved an AUC of approximately 0.757, whereas random forest and gradient-boosted trees achieved approximately 0.83 in external validation. Interpretable analyses identified psychometric assessments, physiological and clinical evaluations such as response to TENS, hypoesthesia, changes in positional pain as significant predictors, thereby supporting the practical utility of explainable Clinical Decision Support Systems (CDSS) in guiding patient selection and the planning of stimulation parameters [23, 28].

Concurrently, LLMs demonstrate high readability and coherence when addressing patient queries but only moderate technical accuracy. Studies evaluating GPT-4–style models on questions about radiofrequency ablation and SCS reported readability scores of 4.8–5 of 5, whereas accuracy scores were only 2.5–2.7 of 5. These findings indicate that LLMs are suitable for patient education and engagement but not appropriate for autonomous therapeutic decision-making without clinician oversight. Accordingly, the most effective deployment strategy integrates safety-constrained LLMs for patient-facing education with explainable, multimodal CDSS to support clinician decisions in a hybrid framework [26, 27].

Methodological challenges persist, including limited datasets from small or single centers, particularly in imaging research, restricted data diversity, heterogeneous definitions of outcomes such as PD-Q thresholds versus percentage NRS reduction, and variability in validation methodologies, including leave-one-out and nested cross-validation. The reporting of calibration, uncertainty, and feature-selection procedures is inconsistent, thereby impeding cross-study comparisons. Further risks include model overfitting and bias; additionally, LLMs pose a risk of factual hallucinations and require domain-specific updates.

Inferences and future directions

The implementation of a hybrid architecture comprises a clinician-facing CDSS layer that provides interpretable, multimodal recommendations, complemented by a LLM layer specifically constrained for patient education, frequently asked questions, and basic triage. Operational considerations include harmonizing multicenter datasets, standardizing clinical and imaging protocols, employing privacy-preserving training methods such as federated learning, and reporting predictive uncertainty and calibration metrics. External validation using independent datasets and long-term randomized trials is imperative to establish the system’s efficacy, safety, and generalizability.

Future directions include developing safety-hardened LLMs for patient education, expanding large, multicenter datasets for CDSS training, integrating explainable AI techniques, and translating these advances into clinical workflows with clearly defined regulatory pathways. This hybrid approach optimizes the immediate benefits of automated patient education and engagement whilst maintaining clinician oversight of therapeutic decisions, thereby facilitating safe and scalable implementation in clinical practice.

Theme 2: closed‑loop neuromodulation and neural dosing

The studies collectively demonstrate that physiologic closed-loop SCS systems, which use the measured neural response to stimulation, specifically, the evoked compound action potential (ECAP), can deliver a quantifiable and consistent “neural dose.” In the Avalon/Evoke prospective multicenter cohort [17], 50 implanted patients followed up for up to 24 months showed a mean reduction in back-pain Visual Analogue Scale (VAS) scores from 81.3 mm to 18.6 mm which represents approximately 77.3% reduction. Additionally, 85.3% of participants met the responder criterion of higher than 50% reduction, and 67.6% were classified as high responders with higher than 80% reduction. The duration within the therapeutic window, as determined by ECAP, was notably high, with median values of 96.7%, 84.9%, and 90.2% across various time points, indicating strong maintenance of targeted neural activation under closed-loop control [17]. The BMINT research platform delineated a compact hardware–software architecture comprising multi-channel recording, Jetson Nano edge AI computing, and a stimulation module. It exhibited system-level latency of approximately 2.83 milliseconds [25]. It demonstrated real-time deployment of machine-learning-based detection and closed-loop responses, including CNN seizure detection with an AUC of roughly 97.6%, in simulation settings [25]. Analyses aggregated from ECAP-controlled closed-loop trials demonstrated a pooled maximal analgesic effect (MAE) of approximately 79% reduction in VAS scores. The studies exhibited high system utilization and indicated that nearly 99% of the time was spent stimulating above the ECAP threshold. Furthermore, the analysis encompassed practical neurophysiological dose metrics that correlate with clinical outcomes [24].

Furthermore, a comprehensive real-world dataset obtained from an AI-guided Transcutaneous Electrical Nerve Stimulation (TENS) and Electrical Muscle Stimulation (EMS) platform (EcoAI) [21] demonstrated population-level, multidimensional improvements in symptoms in which median pain reduction from 4.1 to 1.9 over 24 months and responder rates per domain approximately of 66–71% was achieved. Additionally, it identified pragmatic dosing strategies, including an optimal session frequency of two to four times daily and an optimal session duration ranging from 20 to 59 min, which are pertinent for the scalable application of non-invasive neuromodulation [21, 24]. Despite promising findings, notable gaps remain: [1] the lack of standardized protocols for reporting and defining neural dose metrics, such as normalized ECAP amplitudes, dose ratio thresholds, and dose accuracy conventions, impeding cross-study comparisons and clinical implementation; [2] a relative scarcity of randomized, long-term comparative trials that directly evaluate closed-loop versus open-loop SCS in relation to sustained clinical outcomes and safety; and [3] the ongoing need for rigorous in-vivo validation of hardware and algorithmic latency, robustness, and fail-safe mechanisms beyond bench and simulation testing (a domain where BMINT’s latency data are encouraging but require clinical validation).

Inferences and future directions

Opportunities exist in: establishing consensus on neural-dosing terminology and target ranges based on aggregated device-log analytics; deploying hybrid research platforms, such as BMINT integrated with clinical devices and ECAP-logging SCS systems, to optimize adaptive control mechanisms; translating extensive real-world datasets into hypothesis-driven clinical trials to validate non-invasive dosing paradigms; implementing federated or privacy-preserving learning techniques to aggregate device logs across multiple centers; and undertaking studies on cost-effectiveness and regulatory pathways to facilitate the adoption of physiologic closed-loop SCS, as well as to develop evidence-based neural dosing guidelines. Collectively, these strategies offer concrete steps to transform ECAP-based closed-loop systems from promising innovations into standardized, evidence-based clinical practices.

Theme 3: AI-assisted procedural guidance

The studies collectively demonstrate that machine-learning–based assistance for ultrasound-guided procedures, notably ultrasound-guided peripheral nerve blocks (UGPNB), can reliably identify sonoanatomic landmarks, streamline procedural workflow, and be deployed in near-real time. In a clinical case series, the Nerveblox software generated color overlays of four to five predefined anatomic regions (pectoralis major/minor, subclavian artery/vein, pleura, first rib). It facilitated successful infraclavicular and PECS blocks in three patients without block-related complications and with a rapid onset of surgical anesthesia [19]. A formal validation study involving 40 healthy volunteers yielded 480 paired raw and labeled images. The investigation demonstrated a high degree of concordance among expert validators, with mean landmark accuracy ratings of approximately 4.9 out of 5, and inter-rater kappa coefficients ranging from 0.95 to 0.98 across various block types. These results endorse the reliability of the system’s labeling performance under optimal acquisition conditions [22]. Additional research on the integrated hardware–software platform (BMINT) demonstrates that edge-AI architectures can execute real-time signal processing and deploy deep learning models with system latency of approximately 2.83 milliseconds, while achieving exemplary classification performance in simulations (CNN AUC approximately 97.6%). This highlights the technical viability for applications such as real-time assistive guidance and synchronous closed-loop control [25]. Together, these datasets demonstrate both clinical use cases and the underlying computational capabilities required for AI-assisted procedural guidance.

Significant gaps persist including, involvement of healthy volunteers or small clinical series, which make their generalizability to patients with complex anatomies, such as pediatric, geriatric, obese, or post-surgical distortion, unproven. Additionally, the correlation between automated landmarking accuracy and patient-centered outcomes, such as reduced complications, fewer needle passes, and shorter procedure durations requires randomized and pragmatic effectiveness trials. Moreover, standardized reporting metrics are currently absent. From a technical standpoint, robustness must be demonstrated across varying image qualities, probe orientations, and operator techniques. Furthermore, human–AI interaction paradigms, specifically how AI suggestions are presented and overridden within the clinician-in-the-loop workflow, need optimization.

Inferences and future directions

The opportunities available are creating extensive, diverse, and annotated ultrasound repositories to improve model generalizability, conducting randomized controlled trials (RCTs) and implementation studies to substantiate clinical benefits and cost-effectiveness, integrating edge-AI devices to enhance data privacy while providing low-latency assistance, developing AI-augmented simulation and training curricula to promote rapid skill acquisition, and collaborating with regulators and device manufacturers to certify AI tools as medical devices, establishing concrete safety and performance standards. If these measures are diligently pursued, AI-assisted ultrasound guidance is well-positioned to become a validated adjunct, thereby enhancing safety, efficiency, and education in regional anesthesia and other ultrasound-guided interventions.

Theme 4: digital therapeutics, wearables, and remote monitoring

In contrast to Theme 2, which concentrates on device‑level physiological regulation and ECAP‑based neural dosing, this theme explores the broader digital health ecosystem associated with neuromodulation. This encompasses wearable devices, remote digital therapeutics, home‑based engagement tools, and real‑world monitoring systems that facilitate therapy extension beyond the implanted stimulator. The ECAP‑controlled SCS systems previously discussed produce neural‑dose metrics that can be integrated into these remote platforms; however, the focus here is on data integration, functional capacity in daily life, adherence, and long-term remote management, rather than real-time adaptive stimulation. Studies involving the Evoke closed-loop SCS system demonstrated significant long-term clinical enhancements. Additionally, opioid consumption was reduced substantially, from 62.9 to 29.1 MME/day [15]. Although these outcomes were obtained in the context of an implanted closed-loop stimulator, they underscore the additional significance of integrating recorded neural-dose data with remote monitoring instruments that monitor patient behavior, functionality, and medication consumption in everyday life [17].

Real-world data from the EcoAI platform involving 2,135 users across 187,930 treatment sessions demonstrated that self-reported pain and functional outcomes improved over time. The median pain score decreased. Furthermore, usage patterns, including sessions per day and session duration, were correlated with treatment efficacy [21]. Pooled analyses from three clinical SCS datasets with ECAP control indicated that the MAE in a combined population of 166 patients was approximately 79% pain reduction; implanted recording devices also provided targetable metrics such as “percent time above ECAP threshold” and “neural dose accuracy (RMSE in µV)” to assess therapy delivery (system active time approximately 91% and percent time above threshold approximately 99% at MAE). These findings demonstrate that integrating wearable and implantable devices with logging and AI algorithms enables quantification of neural dose and facilitates home and community monitoring, thereby linking treatment components to real-world patient functioning and medication use [24]. From a technological standpoint, research instruments in edge-AI, such as the BMINT platform, have demonstrated the capacity for real-time processing and deep learning algorithms to operate with minimal system latency (2.83 milliseconds) and substantial traditional accuracy (CNN Area Under Curve = 97.6%, with online simulations surpassing 99% accuracy), thereby facilitating adaptive, response-oriented remote therapy delivery [25].

Despite promising advances, significant gaps persist in translating these technologies into standard care. Firstly, data harmonization and generalizability, considering variations in hardware, programing protocols, self-reported measures, and diverse populations, remain inadequately established. Secondly, evidence demonstrating the correlation between device-derived metrics (such as “ECAP dose” and “time above threshold”) and long-term, patient-centered outcomes (including improvements in daily functioning and reductions in hospitalization and costs) necessitates randomized, longitudinal studies. Operational and regulatory considerations encompass privacy protection, secure data transmission, standardization of device logs, and mechanisms for reimbursement and insurance.

Inferences and future directions

Pursuable opportunities include: 1. the development of cross-disciplinary standards for reporting device metrics and “neural dose,” 2. the design of RCTs and pragmatic economic studies to establish superiority over current care approaches 3. the utilization of edge computing to reduce latency and enable on-device data retention 4. the application of AI models to predict dose adjustment requirements or provide engagement prompts, and 5. the integration of wearable and implantable data with electronic medical records and remote rehabilitation programs to create hybrid care pathways. Overall, the landscape of wearable and digital therapeutics offers a strategic avenue to enhance access, personalization, and quantification of therapy; however, establishing robust clinical evidence, data standardization, and operational and regulatory frameworks remains essential.

Critical analysis and summary of findings

Across the four themes, the eleven included studies describe heterogeneous applications of emerging technologies in interventional pain management. Within selected single-center cohorts, machine-learning–based models combining clinical and imaging variables reported improved predictive accuracy for SCS response. ECAP-controlled closed-loop SCS studies reported high responder rates and reduced opioid use over 24 months in implanted cohorts. AI-assisted ultrasound studies described high concordance for anatomic landmark identification, predominantly in healthy volunteers and small case series. AI-driven wearable TENS/EMS data and LLM-based patient education evaluations reported improvements in self-reported outcomes and acceptable reliability/comprehensibility, with lower scores on technical accuracy. Because of substantial differences in study design, populations, interventions, and outcome measures, these findings are not directly comparable across themes and should be interpreted as a descriptive map rather than as evidence of comparative effectiveness. Most studies were small or single-center, and AI-related models were rarely externally validated (see Fig. 2).

Fig. 2.

Critical Analysis and Summary of Findings

Nevertheless, significant limitations remain. Many studies carried out at individual centers involve relatively small cohorts, employ various outcome measures, and are prone to biases inherent in AI models and digital platforms. There is a need for larger multicenter longitudinal studies, supported by standardized reporting and regulatory frameworks, to establish clinical efficacy, safety, and generalizability more effectively. Although the findings are promising, they should be interpreted with prudence, and further evidence is required to endorse broad clinical adoption.

Discussion

The primary objective of this scoping review was to synthesize current evidence concerning emerging technologies in interventional pain management and to delineate prospective directions for future research. Our findings suggest that AI-based predictive models, closed-loop SCS, AI-assisted procedural guidance, and wearable digital therapeutics collectively enhance precision, personalization, and treatment efficacy. Compared to earlier single-modality or small-scale studies, our review underscores the additional value of multimodal data integration, longitudinal outcomes, and real-world datasets, thereby providing more robust evidence for patient selection, therapy optimization, and remote monitoring. While prior reports demonstrated feasibility or short-term benefits, our synthesis emphasizes long-term analgesic effects, responder stratification, neurophysiologic dosing, and practical deployment considerations. Notably, AI-assisted procedural guidance and wearable platforms exhibit promise in improving safety, adherence, and individualized therapy delivery, although gaps persist in generalizability, standardized outcome metrics, and regulatory frameworks. Overall, these findings corroborate prior literature by demonstrating the potential of emerging technologies and extend understanding through larger-scale, multimodal, and real-world evidence that support clinical translation and hybrid clinician–AI approaches.

Whereas Antel et al. (2025) conducted a systematic review that broadly mapped AI applications across acute and chronic pain management, our scoping review focuses specifically on interventional pain procedures, including neuromodulation, nerve blocks, and intrathecal delivery. It also encompasses non-AI technologies, such as closed-loop electrophysiological systems and LLM-based patient education tools, thereby offering a complementary perspective [1].

The findings of this scoping review indicate that emerging technologies in interventional pain management, including closed-loop SCS, AI-assisted procedural guidance, and wearable or remote digital therapeutics, possess considerable potential to improve precision, personalization, and treatment efficacy. These results are consistent with the Biopsychosocial Model of pain, as they address not only biological mechanisms (such as neural stimulation and quantifiable “neural dosing”) but also psychological and social dimensions through patient education facilitated by LLMs and remote engagement tools [29, 30].

Furthermore, the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT) frameworks help contextualize clinical and operational adoption processes. They emphasize that successful technology implementation depends on the alignment between technological capabilities and user acceptance within healthcare systems [31]. Notably, the integration of multimodal patient data, predictive algorithms for treatment response, and adaptive neuromodulation platforms exemplifies the movement toward precision pain management, in which therapy is tailored to individual patient characteristics and real-world response metrics [31, 32]. Collectively, these findings indicate that incorporating multidimensional interventions, user-focused design, and individualized treatment approaches can form a comprehensive conceptual framework to guide the future development and deployment of emerging technologies in interventional pain management [33].

Recent advances in the management of chronic pain increasingly incorporate virtual reality (VR), telehealth, robotics, AI, and interventional procedures. Cerda et al. (2024) presented a narrative review that highlights telehealth and emerging VR technologies, underscoring their potential to facilitate remote physical assessments, biofeedback interventions, and personalized asynchronous care. Nevertheless, the long-term efficacy and scalability of these technologies remain limited [34]. Supporting this, Goudman et al. (2022) conducted a systematic review and meta-analysis of 41 studies, demonstrating significant improvements in pain and functioning following VR interventions, thereby emphasizing VR as a promising adjunct for managing chronic pain [35]. Matthie et al. (2022) similarly reported that VR, augmented reality (AR), and mixed reality (MR) enhanced pain management outcomes in 78% of randomized controlled trials. However, they emphasized the importance of longer follow-up durations and the assessment of accessibility [36]. Viderman et al. (2023), in an umbrella review encompassing 21 systematic reviews involving over 17,000 patients, reaffirmed the efficacy of VR across various contexts, including chronic, perioperative, and periprocedural pain, thereby demonstrating its broad applicability [37].

Robotic technologies have also demonstrated potential, particularly in socially assistive robots and devices for managing procedural pain, as emphasized by Higgins et al. (2023). Of the 17 studies examined, 12 exhibited statistically significant reductions in pain measures; however, variability in populations, devices, and trial designs constrains the broader applicability of these findings [38]. In conjunction with digital and robotic interventions, interventional pain procedures offer durable analgesic effects. Eshraghi et al. (2022) [20] compiled evidence regarding percutaneous interspinous spacers and minimally invasive lumbar decompression, indicating substantial functional enhancements in patients with lumbar spinal stenosis. Meanwhile, Habib et al. (2023) documented that intrathecal pumps, neuromodulation, kyphoplasty, and targeted ablative procedures provide adequate and safe pain relief in oncology populations, thereby reducing reliance on opioids [39].

AI is being increasingly employed to optimize both assessment and intervention. Zhang et al. (2023) demonstrated AI’s capacity to enhance pain recognition, predict outcomes tailored to individual patients, and facilitate self-management [40]. Conversely, Leoni et al. (2024) emphasized AI-assisted image interpretation, procedural guidance, and augmented reality integration in spinal interventions [41]. Kamal et al. (2023) further elucidated the role of multi-agent systems within telehealth frameworks, emphasizing their capacity to facilitate coordinated, patient-centered decision-making for the management of chronic pain [42]. Strand et al. (2022) highlighted that the incorporation of innovative technologies within interventional, robotic, and AI-assisted fields enables the development of tailored, physiologically informed therapeutic approaches [43].

Beyond pain-specific applications, recent broader reviews on AI in healthcare emphasize the importance of hybrid and interpretable systems that combine data-driven models with expert knowledge and clinician oversight, particularly for high-stakes decisions where black-box predictions are insufficient [44]. Similarly, recent systematic evidence on AI-driven approaches in perioperative pain and psychological outcomes confirms the trends observed in our review: predictive analytics, real-time monitoring, and personalized interventions are emerging across surgical and chronic pain populations, but evidence remains limited by small samples and heterogeneous outcome measures [45]. Updated methodological guidance for scoping reviews further supports our approach: critical appraisal is optional, and the primary purpose of a scoping review is to map evidence and identify gaps rather than to draw effectiveness conclusions [46]. Together, these works reinforce the framing of our findings as a descriptive map of an emerging field that requires hybrid clinician–AI architectures, standardized outcome definitions, and prospective multicenter validation before broader clinical integration.

By integrating objective neurophysiological metrics with self-reported pain and functional assessments, our study maps gaps in the earlier literature regarding long-term efficacy, generalizability, and scalability. These observations should be interpreted as descriptive mapping rather than evidence of comparative effectiveness, and they highlight the need for further multicenter, prospectively registered, and standardized studies before integrated, technology-enhanced strategies can be recommended for routine chronic pain management.

Strengths and limitations

This study employed a scoping review to investigate technological innovations in interventional pain management comprehensively and to develop a framework for future research. Its strengths include an extensive search across multiple databases, transparent inclusion and exclusion criteria, and thematic analysis of studies involving various technologies, such as artificial intelligence (AI), neuromodulation, virtual and augmented reality (VR/AR), RPM, and wearable devices. Additionally, the included studies integrated clinical, imaging, and virtual-reality data to address both short- and long-term therapeutic outcomes. The limitations identified include a limited number of clinical studies, single-center or small-sample designs, heterogeneity in outcome definitions, and potential AI-related constraints that may affect accuracy and generalizability. Moreover, the lack of standardized reporting practices, the paucity of randomized long-term trials, and the variability in technological interventions present challenges for cross-study comparisons and clinical implementation. A further fundamental limitation of this review is the substantial heterogeneity of the included technologies, which span artificial intelligence-based clinical decision support, closed-loop neuromodulation guided by evoked compound action potentials, AI-assisted procedural ultrasound guidance, large language model-based patient education tools, and wearable digital therapeutics. The included studies also differ widely in populations (chronic pain patients, healthy volunteers, epilepsy datasets, and non-patient evaluators), study designs, intervention intensity, and outcome measures. While this breadth is consistent with the mapping function of a scoping review, it precludes meaningful quantitative pooling of effects and limits the generalizability of any cross-domain conclusions. Readers should therefore interpret the mapped evidence as descriptive rather than directly comparable across themes. Additionally, the review protocol was not prospectively registered (scoping reviews are not eligible for PROSPERO), which represents a methodological limitation. The optional critical appraisal performed using JBI tools should be interpreted as descriptive context rather than as a quantitative quality threshold for inclusion. Finally, the predominance of small, single-center studies and the absence of head-to-head randomized comparisons mean that conclusions regarding effectiveness or readiness for clinical implementation remain preliminary.

Conclusion

Emerging technologies in interventional pain management, including closed-loop SCS, AI-assisted imaging guidance, and wearable or digital therapeutics, represent an emerging and heterogeneous body of literature. This scoping review mapped the existing evidence and identified four broad thematic areas: precision patient selection, closed-loop neural dosing, AI-assisted procedural guidance, and digital therapeutics with remote monitoring. Across these themes, the available studies are predominantly small, single-center, and methodologically diverse, with limited long-term follow-up and few head-to-head comparisons. Consequently, this review describes the scope and characteristics of the current evidence rather than establishing comparative effectiveness or clinical readiness. Several research gaps were consistently identified, including the absence of standardized outcome metrics, the paucity of multicenter and longitudinal trials, and the limited evaluation of generalizability across populations and care settings. Future research should prioritize prospectively registered protocols, harmonized outcome definitions, multicenter prospective designs, and explicit reporting of model calibration, uncertainty, and external validation, before broader claims about clinical implementation can be supported.

Implications for clinical practice

As a scoping review, this work does not establish clinical effectiveness, and the implications described here should be interpreted as preliminary signals from a heterogeneous evidence base rather than recommendations for routine practice. The mapped literature suggests that, in selected settings and with appropriate clinician oversight, closed-loop SCS systems and wearable devices may support neural- or behavior-based monitoring and parameter adjustment, and AI-assisted imaging may support procedural training and decision-making during nerve blocks. Any future implementation of these technologies would require prospective comparative trials, harmonized outcome reporting, structured human–AI workflows, and adherence to regulatory and data privacy frameworks before integration into routine interventional pain practice can be supported.

Author contributions

[MRF]: Conceptualization, Methodology, Data Curation, Writing - Original Draft Visualization; [GV]: Conceptualization, Methodology, Resources, Data Curation, Writing - Original Draft; [RN]: Conceptualization, Methodology, Data Curation, Visualization, Review & Editing.

Funding

No external funding was received.

Data availability

The datasets generated or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable. This scoping review did not involve any individual person’s data, identifying images, or other personal or clinical details requiring consent for publication.

Competing interests

The authors declare no competing interests.