Human–AI collaboration for dysphagia rehabilitation from effectiveness to implementation complexity: a systematic review Authors: Wenwen Yang, Sufang Li, Yifei Du, Mengran Chen, Funa Yang, Fan Zhang, Ji Zhao, Yanqing Li, Xiaoxia Xu Published: 2026 ABSTRACT AbstractOropharyngeal dysphagia affects over half of neurological and oncological populations, yet rehabilitation is constrained by a global therapist shortage that human–AI collaboration has not demonstrably addressed. Here we report a systematic review of 31 studies (1012 participants; PROSPERO: CRD420251115997) evaluating AI-augmented swallowing rehabilitation in adults with oropharyngeal dysphagia, or in healthy volunteers testing systems designed for clinical application. We synthesised findings by aetiology and collaboration mode, assessing risk of bias and certainty of evidence (Grading of Recommendations, Assessment, Development and Evaluation, GRADE). AI-augmented interventions produce short-term gains in functional oral intake and physiological measures (GRADE moderate/low certainty), but these effects attenuate within weeks of cessation, and adherence declines sharply once clinician supervision is withdrawn. NASSS framework analysis reveals a central paradox: the adopter domain—digital literacy, cognitive impairment, interface usability—is the dominant implementation barrier (61.3% rated high), meaning the populations with the greatest need face the steepest barriers to adoption. AI algorithm performance is rated at very low certainty, with validation largely confined to healthy volunteers. These findings support advancement to pragmatic trials for supervised post-stroke rehabilitation but underscore that evidence for other aetiologies, unsupervised settings, and sustained outcomes remains insufficient.Subject terms: Computational biology and bioinformatics, Health care, Mathematics and computing, Medical research Introduction Safe swallowing requires sub-second coordination of over thirty craniocervical muscle pairs1,2—yet its rehabilitation remains critically under-resourced worldwide. Oropharyngeal dysphagia affects up to 30% of community-dwelling older adults and exceeds 50% in most neurological and oncological populations studied, including stroke, Parkinson’s disease, age-related frailty and head and neck cancer (HNC)3–8. Regardless of aetiology, dysphagia independently raises the risk of aspiration pneumonia, malnutrition, and death—post-stroke dysphagia alone confers a more than fourfold increase in pneumonia risk9,10, and aspiration pneumonia remains the leading cause of death in advanced Parkinson’s disease11,12. Because swallowing dysfunction frequently persists or progresses beyond the acute phase13,14, rehabilitation needs are chronic and escalating—accounting for an estimated US$4.3–7.1 billion in excess dysphagia-related inpatient costs per year in the United States15,16, a burden set to intensify as the global population aged 60 and older approaches 2.1 billion by 205017. Intensive swallowing rehabilitation promotes neuroplastic recovery and functional improvement18,19, but delivering it at adequate intensity hinges on a specialist workforce whose numbers fall far short of global need—and the deficit is widening. In the United States, speech-language pathologist employment is projected to grow by 15% between 2024 and 2034—well above the occupational average—yet still fall short of demand20. In low- and middle-income countries the deficit is orders of magnitude larger: the WHO estimates fewer than ten skilled rehabilitation practitioners per million population, and only 17% of low-income countries have even one speech–language therapist per million21,22. The consequence is that a large proportion of patients worldwide cannot access swallowing rehabilitation at sufficient intensity. Technology may help close this gap—but only if it augments, rather than supplants, the clinical expertise on which safe practice depends. Human–AI collaboration in rehabilitation embodies this principle, positioning AI not as an autonomous decision-maker but as computational support—real-time physiological monitoring, pattern recognition across multiple signal streams, and individualised dosage adaptation—while clinicians retain contextual judgement, oversight of aspiration risk, and the therapeutic relationship23–25. We define a human–AI collaborative rehabilitation system as one integrating: at least one AI-enhanced component, such as adaptive parameter adjustment, multi-parameter pattern recognition, personalised protocol generation, or algorithm-driven real-time feedback and risk alerting; and at least one form of human clinical involvement, such as treatment plan formulation or approval, intervention supervision, parameter adjustment, or exception management. This definition spans a spectrum of collaboration intensity, from continuous clinician oversight with AI-augmented assessment to semi-autonomous AI operation under periodic clinical review. Swallowing lends itself to such collaboration: it generates multimodal physiological signals—electromyographic activity, lingual pressure, laryngeal excursion, deglutition acoustics26–28—that encode the rapid biomechanical sequences governing airway protection and bolus transit. These sequences unfold on millisecond timescales, beyond the reach of unaided clinical observation2 but amenable to computational analysis. Preliminary validation indicates that systems built on these signals achieve acceptable detection accuracy in specific populations and improve training precision in controlled settings29–31. Whether these early capabilities translate into real-world clinical benefit remains unclear. Systematic reviews show that specific swallowing interventions yield favourable group-level effects on impairment4,32–34, yet a Cochrane review—assessing functional endpoints—found no demonstrable reduction in mortality or long-term disability, with substantial inter-individual response heterogeneity that conventional clinical variables do not adequately explain35. This divergence reflects a persistent challenge in dysphagia research: physiological surrogates and functional outcomes do not reliably co-vary, so that gains in parameters such as muscle activation amplitude or lingual pressure do not consistently predict improvement in swallowing safety or oral intake. For AI-augmented systems, which must select optimisation targets from accessible physiological signals, this dissociation carries direct design consequences. Marked variation in baseline physiology across aetiologies compounds this problem: algorithms validated in post-stroke cohorts—from which most current evidence derives—cannot be assumed to transfer to neurodegenerative or oncological populations, which remain substantially under-represented. Demonstrating effectiveness is only the first translational hurdle. Implementation science shows that most AI systems performing well under controlled conditions fail to achieve routine adoption, for reasons rooted in the fit between technology design, user capacity, and care environment2,36. In swallowing rehabilitation, this challenge arises under conditions that distinguish it from other digitally augmented therapies. The defining impairments of the target population—neurological, cognitive, sensory—are precisely those that compromise technology interaction37, creating a fundamental tension between who most needs the system and who can most readily operate it. The stakes compound this tension: inadequate system performance during swallowing training can precipitate aspiration—a potentially fatal event—imposing fault-tolerance requirements that exceed those of most rehabilitation scenarios. The populations with the greatest need for these technologies—those in the most resource-constrained settings—face the steepest structural barriers to accessing them. Without deliberate countermeasures, these technologies risk entrenching the very inequities they were designed to alleviate38. These interdependent challenges remain unaddressed within a unified analytical framework. Previous systematic reviews examined individual technology types in isolation—surface electromyography (sEMG) biofeedback, gamified interfaces, mobile health platforms—yielding technology-specific conclusions that cannot account for why interventions efficacious in controlled trials have consistently failed to enter routine practice. This limitation is not merely additive; it is structural. Effectiveness and implementation complexity are coupled: features that enhance efficacy under supervision—high-fidelity physiological sensing, real-time adaptive algorithms—simultaneously raise barriers to deployment through calibration burden, digital-literacy requirements, and hardware costs. Conversely, design choices that simplify adoption—smartphone-only delivery, reduced clinician involvement—risk attenuating efficacy. Disentangling these trade-offs demands a framework that evaluates both dimensions jointly. The Non-adoption, Abandonment, Scale-up, Spread, and Sustainability (NASSS) framework39, which maps implementation complexity across seven interacting domains from condition characteristics to long-term sustainability (Fig. 1), offers a mature lens for this analysis40; what has been absent is its integration with effectiveness synthesis in a single review. NASSS framework-based complexity assessment of implementation barriers to human–AI collaborative dysphagia intervention. Barriers are mapped onto seven NASSS domains: D1 (Condition), D2 (Technology), D3 (Value Proposition), D4a (Adopters: Patients), D4b (Adopters: Therapists), and D5–7 (Organisation, Wider System, and Embedding). Surrounding panels detail domain-specific barriers synthesised from the included studies. Directional annotations between domains indicate cross-domain cascading interactions. Solid arrows denote primary influence pathways; dashed curved arrows denote cross-domain interactions; the dashed boundary at the bottom indicates additional complexity amplification in low- and middle-income country contexts. Here we report a systematic review with two converging objectives: to synthesise effectiveness evidence for human–AI collaborative swallowing rehabilitation across aetiologies, technology types, and collaboration modes; and to apply the NASSS framework to identify the systemic barriers separating controlled efficacy from routine adoption. We developed a taxonomy of human–AI collaboration modes to characterise how clinical and computational tasks are allocated across interventions, and examined whether mode of collaboration is associated with differences in outcomes and implementation complexity. In doing so, we sought to advance the field beyond the question of whether these technologies work, towards an account of how, for whom, and under what conditions they can be embedded in the rehabilitation of one of the most physiologically demanding sensorimotor functions in human medicine. Results Study selection The systematic search across PubMed/MEDLINE, Embase, the Cochrane Library, and Web of Science yielded 5236 records. Following removal of 3914 duplicates, 1322 unique records underwent title and abstract screening. Of these, 1112 were excluded as clearly not meeting the inclusion criteria, leaving 210 records for full-text assessment. Supplementary search strategies—including reference list searching, forward citation tracking, trial registry searching, and expert contact—identified 28 potentially relevant records, of which 12 proceeded to full-text assessment after screening. In total, 222 full texts were assessed (210 from database searching and 12 from supplementary searches), of which 191 were excluded for the following reasons: non-relevant interventional studies (n = 66), diagnostic AI systems without therapeutic integration (n = 33), conference abstracts with incomplete data (n = 30), non-relevant populations (n = 15), non-collaborative study designs lacking human–AI interaction (n = 42), and non-relevant outcomes (n = 5). Ultimately, 31 studies met all inclusion criteria and were included in the review. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram illustrates the complete study selection process (Fig. 2). PRISMA flow diagram of the search and study selection strategy. Study characteristics The included studies were published between 2016 and 2026: five (16.1%) during 2016–2019 and 26 (83.9%) during 2020–2026. Studies originated from diverse geographical regions, with 15 from the Asia-Pacific (48.4%; mainland China, n = 6; South Korea, n = 4; Taiwan, n = 3; Hong Kong and Japan, n = 1 each), six from North America (all from the United States), seven from Europe (Italy, n = 3; Turkey, n = 2; the United Kingdom and the Netherlands, n = 1 each), and three (9.7%) from multinational collaborations. Descriptively, facial recognition and gamified systems were more prevalent among Asia-Pacific studies, whereas sEMG biofeedback predominated among North American and European studies. The principal study designs comprised randomised controlled trials (Randomised Controlled Trial, RCTs; n = 13, 41.9%; including nine full-scale RCTs, two pilot RCTs, and two feasibility RCTs), non-randomised interventional studies (n = 8, 25.8%; including six quasi-experimental designs, one case-control study, and one longitudinal study), technology development and validation studies (n = 8, 25.8%), and qualitative studies (n = 2, 6.5%). Sample sizes ranged from 2 to 113 participants (median = 23.0; interquartile range [IQR]: 12.25–58.75). Study settings spanned the full care continuum: acute hospitals and inpatient rehabilitation facilities (n = 15, 48.4%), research laboratories (n = 6, 19.4%), home-based or home-extension settings (n = 4, 12.9%), community centres (n = 4, 12.9%), and hybrid hospital-to-home transitional models (n = 2, 6.5%). Detailed study characteristics, intervention technologies, and outcomes are summarised in Table 1. Table 1Study characteristics, intervention technology, and outcomesStudyDesign/SettingSample (n)PopulationTechnology SystemAI Algorithm and HardwareFeedback and InterventionDosing ProtocolComparatorFollow-upOutcomesZhang (2025) ChinaRCT (single-blind) Hospital → homeN = 84 IG = 42, CG = 42 Male: IG 57.1%, CG 69.0%Post-stroke dysphagia Moderate severity (GUSS 6.5–7.0) Age: 65.0 ± 9.7 vs 66.4 ± 13.1 y Onset: 1.72 ± 1.50 monthsAI-VG system Facial recognition + game biofeedback 3 games: Carrot Collection (lip), Maze Escape (tongue), Leaping Barriers (CTAR)MediaPipe 468-point 3D facial landmark detector Computer/tablet + camera Real-time trackingVisual biofeedback via game control Lip exercises (15 reps/side) Tongue exercises (15 reps/direction) CTAR exercises (15 reps)30 min × 5/wk × 4 wk Total: 20 sessions Home-based after initial hospital trainingConventional exercises (lip, tongue, CTAR) without AI-VG8 wk (4 wk Tx + 4 wk F/U)Primary: GUSS β = −4.02 (95% CI: −6.16 to −1.89), P < 0.001 ↑ Maintained at 8 wk (P < 0.001) Secondary: SSA: P = 0.006 ↑ FOIS: P < 0.001 ↑ MNA-SF: P = 0.034 ↑ SWAL-QOL: P < 0.001 ↑ Adherence: 90% (18/20 sessions) Satisfaction: 73/100Spoladore (2024) ItalyFeasibility study (system development)Home-basedN = 9 clinicians (validation) + 13 patients (ontology testing)Neurological dysphagiaVarying severity (DOSS/PAS) Age: NR Patients requiring dietary modificationKnowledge-based DSS Ontology-based semantic reasoningTwo applications: Clinician App + Patient App for safe recipe recommendationsOWL 2 DL ontology + 14 SWRL rules DOSS severity and PAS consistency mapping Computer/mobile + web browserVisual recipe display interface Personalised recipe filtering by: C1: DOSS severity level C2: PAS consistency Step-by-step cooking instructionsOn-demand use No fixed protocol System validation phase onlyNA (technical validation)NA (validation study)Primary: Recipe–clinician agreement >66.7% approval rate High agreement for most recipes Secondary: Clinician App: easy to use, useful Patient App: user-friendlyZhang (2025) ChinaPilot RCT (single-blind) Hospital inpatientN = 26 IG = 14, CG = 12 Male: IG 57.1%, CG 83.3%Post-stroke dysphagia GUSS-confirmed Age: 66.5 ± 8.3 vs 66.2 ± 15.0 y Ischaemic stroke: 78.6–83.3%FR-VG system Facial recognition video game Same 3 games as AI-VG (lip, tongue, CTAR exercises)MediaPipe 468-point 3D facial landmark detector Computer + camera + displayReal-time visual game biofeedback Game responds to facial movement Supervised hospital setting5×/wk × 4 wk Total: 20 sessions Direct therapist supervisionConventional swallowing rehabilitation without video game4 wk (Tx only, no F/U)Primary: GUSS P = 0.015 ↑ Secondary: FOIS: P = 0.004 ↑ Adherence: P = 0.032 ↑ FR-VG superior to conventional High acceptanceChan (2024) Hong KongMulti-centre quasi-experimental (self-controlled) Community centresEnroled: N = 113 Completed: N = 70 Male: 25.7%Healthy/pre-clinical elderly EAT-10: 4.67 ± 5.91 Age: 67.7 ± 4.9 y Community-dwelling ≥50 ye-Oral APP AI smartphone application 12 exercise types for tongue, lip and chin trainingERT 68-point facial landmark detector Real-time movement tracking Smartphone (iOS/Android) + front cameraVisual + audio feedback Video demonstrations Real-time landmark display Immediate accuracy feedback Fortnightly SLP consultations12 reps/set × 1 set/day 7×/wk × 8 wk Total: 56 sessions Home-based + periodic consultationNA (self-controlled)12 wk (8 wk Tx + 4 wk F/U)Primary: Oral function Bite force: P < 0.001 ↑ Chewing efficiency: P = 0.002 ↑ Tongue pressure: P < 0.001 ↑ Tongue endurance: P = 0.004 ↑ Secondary: DDK rates: P < 0.05 ↑ EAT-10/DRACE: NS Adherence: 97.87%Satisfaction: 80.3% ≥3/5Su (2025) TaiwanSingle-arm randomised pilot (feasibility)Home-basedN = 32Male: 21.9%Healthy middle-aged/elderlyPre-clinical (prevention)EAT-10 screenedNo swallowing disordersTongue Training AppAI mobile exergameFlow-based gameplay3 speed levels (slow/medium/fast)TeachableMachine (Google)1,200+ training imagesKeras/TensorFlow Lite + UnityAndroid smartphone + Garmin HRMVisual + audio feedbackGame prompts (‘great’, ‘nice’)Music-synchronised rhythmAdaptive speed progression30 min single session(feasibility testing)Within-subject speed comparisonNA (within-subject comparison)NA (single session)Primary: EngagementFSS2: P = 0.001 ↑NASA-TLX: P < 0.001Flow Distance: P < 0.001Cronbach α = 0.802Secondary:HRV measures: 25–58% significantK-means: 3 groups identifiedAdherence: 100%Medium speed optimalJung (2024)South KoreaRCT (3-arm)Home-basedEnroled: N = 90Completed: N = 76App: 25, Non-app: 25, Control: 26Male: 11.5–16%Community-dwelling elderlyAge: 65–85 yHealthy populationSmartphone exercise appCombined oral + whole-body exercise4 categories: oral health, expert hygiene, oral exercises, systemic exercisesStructured exercise guidance applicationSMS reminder systemSmartphone with appVisual exercise demonstrationsSMS reminders for adherenceOral + systemic exercise guidanceSelf-directed daily practiceApp: 60 min × 7/wk × 5 wkNon-app: 60 min × 2/wk × 5 wkNon-app: same exercises with supervised training 2×/wkControl: no exercise5 wk (Tx only, no F/U)Primary: Tongue pressure (IOPI)Anterior: +4.48 kPa, P < 0.05 ↑Posterior: +3.88 kPa, NSCheek: +2.76 kPa, NSSecondary:Saliva flow: +0.71 g/min, P < 0.05 ↑OHIP-14, GOHAI: NSDropout: 8.9% (8/90)Benfield (2023)UKRCT (feasibility)Hospital(acute stroke unit)N = 27IG = 12, CG = 15Male: IG 50%, CG 13%Acute stroke with dysphagiaWithin 4 wk of onsetAge: 73.3 ± 11.0 yMedically stable patientsBiSSkiT systemsEMG biofeedback softwareStrength training: 70% max amplitudeSkill training: amplitude + timing targets within 30 s windowNeuroTrac Simplex sEMG deviceTriode electrodes (submental)BiSSkiT softwareComputer + displayReal-time visual sEMG waveform feedbackTarget amplitude thresholdsTiming windows for skill trainingSwallowing exercises with biofeedback45 min × 5/wk × 2 wkTotal: 10 sessionsSupervised hospital settingConventional care (standard dysphagia management without biofeedback)12 wk (2 wk Tx + 10 wk F/U)Feasibility:84.6% completed >80% sessions92% comfortable/very comfortable75% appropriate frequencyClinical (within 2 wk):FOIS: +2 points ↑DSRS: +3 points ↑PAS 5 mL: +1 point ↑At 90 days: no group differencesPark (2019)South KoreaRCT (assessor-blind)HospitalEnroled: N = 46Completed: N = 37IG = 20, CG = 17Male: IG 65%, CG 50%Post-stroke dysphagiaVFSS-confirmedAge: IG 60.95 ± 11.19 y,CG 59.45 ± 9.34 ygbCTAR system‘Eating Little Fish’ gameGame-based CTAR with elastic resistance barLES 100 device with elastic resistance barMotion sensing game softwareDisplay monitor for gameReal-time visual game feedback‘Eating Little Fish’ metaphorAdjustable resistance levelsCTAR exercises with gamification30 min × 5/wk × 4 wkTotal: 20 sessionsSupervised trainingHead-lift exercise (HLE)Conventional Shaker exercise (isometric + isotonic)4 wk (Tx only, no F/U)Primary: Swallowing functionVDS: both groups P < 0.05 ↑PAS: both groups P < 0.05 ↑FOIS: both groups P < 0.05 ↑No between-group differenceSecondary: EngagementMotivation: gbCTAR > HLE, P < 0.05Interest/enjoyment: P < 0.05Physical demand: gbCTAR < HLE, P < 0.001Dropout: 19.6% (9/46)Khan (2024)Australia/CanadaFeasibility study(system development)Community/homeN = 12Male: 66%Age: 17.5–23 yHealthy young adults(system validation)No neurological conditionsNo swallowing disordersTMIS systemTongue–machine interactionIR-based tongue tracking for computer controlCursor/game control via tongueNintendo Wii remote IR cameraReflective tape on silicone tongue deviceIR LED arrayFreeTrack softwareDwell Clicker2Computer + displayVisual cursor control based on tongue positionGame software interactionAlarm system activationText processor controlTongue acts as computer pointerSystem testing onlyNo fixed protocolPrecision and ROM assessmentComparison systems:Finger touchscreenTongue capacitive touchscreenNA (single session testing)System performance:Horizontal FOV: 42°Vertical FOV: 33°Cursor efficiency: 36.5% (vs finger baseline)33.2% > tongue capacitiveClick accuracy: 70%ITR: 130 (6 commands)Cost: 100% stretchabilityBluetooth wireless (BioRadio)Threshold classification algorithmBandpass 30–150 Hz, RMS smoothingVisual biofeedback: ‘jumping ball’ gameBall jumps between plates on successful swallow detectionReal-time sEMG waveform displaySubmental electrode placementSingle session testingLiquid swallowing: 20 mL water every 3 s, 10 trialsDry swallowing during gameConventional rigid electrodes (MVAP II snap electrodes with gel)NA (acute testing only)Swallow detection:Liquid: 100% accuracy (both types)False positive: 3% vs 5% (rigid)Game performance:Success: 94.7 ± 4.4% vs 96.2 ± 1.0%False positive: 16.8% vs 22.3%Motion artefact: ~6× lower (skin-like)SNR: 21.4 ± 0.6 dB (equivalent)Acceptability: 6.5/7 vs 4/7 (rigid)Zhang (2024)China/Hong KongMixed-methods (feasibility)Rehabilitation centreN = 19(11 patients + 8 HCPs)Patients: 81.8% maleHCPs: 100% femalePost-stroke dysphagia (inpatient)Onset: 41.36 ± 29.48 daysAge: 60.18 ± 10.91 yGUSS severity:Mild–moderate 54.5%, moderate–severe 36.4%Video game swallowing systemMediaPipe AI + video game3 mini-games: Lip (rabbit), Tongue (maze), Jaw (bird + ball squeeze)MediaPipe face mesh algorithm468 facial landmarks, 3D real-time trackingRGB camera + computer display + headsetReal-time visual feedback via game character animationsAudio feedbackOral exercises (lip, tongue, jaw) through game control1-wk familiarisation(before qualitative interview)Non-standardised user experienceNA (qualitative design)NA (cross-sectional)Technology acceptance (TAM):Perceived usefulness:Rehabilitation benefits valuedIncreased motivationPerceived ease of use:Visual feedback appreciatedDigital literacy challenges (elderly)Intention to use:High willingnessBackground supervision neededAccuracy improvements suggestedKuramoto (2016)JapanDevice validation(technical feasibility)University laboratoryN = 11Male: 45.5%Age: 71.4 y (mean)Dysphagia patientsReferred for VFSS examinationAble to swallow 3 mL thickened bariumSwallowscope 2.0Smartphone-based wearable swallowing monitorLED visual feedback + remote display for caregiversNeck-mounted contact microphone6-axis IMU (accelerometer + gyroscope)LED visualisation unitAndroid smartphoneTemplate matching algorithm for swallow soundIMU-based posture detectionLED visual feedback (colour-coded status)Remote wireless display for caregiversReal-time monitoring interfaceSwallow sound + neck posture monitoringSingle VFSS sessionSwallowing 3 mL thickened liquid barium (25% mixture)Simultaneous VFSS imagingVFSS as gold standard referenceNA (validation study)Swallow detection:Accuracy: 75.9% (thickened liquid)Previous: 79.3% (thin liquid)Posture tracking:Mean error: 14.6° (SD 1.3)Bolus discrimination:Swallow duration differs by consistency (P < 0.05)Potential for nursing home mealtime assistanceKim (2020)South KoreaMixed-methods pilot(feasibility)Community senior centresN = 11Female: 90.9%Age: 75.7 ± 3.93 y (67–83)Community-dwelling elderlyAt-risk for dysphagia (presbyphagia)No diagnosed dysphagiaPreventive intervention target365 Healthy Swallowing Coach AppmHealth tablet application3 exercises: EPS, EPG, ETR (effortful swallowing exercises)Galaxy Tab A SM-P580 (with S-pen)App-based exercise guidanceVideo demonstrationAutomated remindersTraining record trackingVisual: exercise videos, animated guidesAudio instructionsProgress tracking displayEPS, EPG, ETR exercises2 sets × 10 reps = 60 reps/exercise/day8 wk5 days/wk3 sessions/day (AM/PM/EVE)Total: 120 sessions~15–20 min/sessionNA (single-arm feasibility)8 wk (post-intervention only)Usability (SUS):High-potential: ‘acceptable’ rangeLow-potential: ‘marginal’ rangeSelf-efficacy (mCSES):High-potential: 54.0 → 65.5, P = 0.04 ↑Low-potential: NSAdherence: 72.7% (8/11) optimal doseKey finding: education >10 y + tech experience predicts successAslan (2024)TurkeyFeasibility study(BCI classification)Research laboratory(Erciyes University)N = 30Male: 50%Healthy adultsHealthy volunteersNo neurological/psychiatric disordersNo dysphagiaEEG-based BCI systemBrain–computer interface for tongue movement detectionMotor imagery paradigmg.tec Nautilus EEG (20 channels, 10–20 system)ML: KNN, SVM, Decision Tree, Random Forest, AdaBoost, BaggingDL: CNN + CWT scalogram (256 × 256 × 3 RGB)Signal: EMD + 0.5–40 Hz bandpass + ICA30 time/frequency featuresNA (classification study)No real-time feedback testedMotor imagery: rest vs imagined tongue protrusion15 trials × 9 s eachSingle recording sessionEEG during rest and imagination phasesNo rehabilitation interventionRest phase as comparison (within-subject)NA (single session)Classification accuracy (rest vs imagination):Traditional ML:KNN: 79.4%SVM: 63.4%Ensemble methods:AdaBoost/Bagging/RF: 99.8%Deep learning:CNN + CWT: 83%Ensemble methods outperformed othersPotential for BCI-based dysphagia rehabilitationSrp (2024)Multi-centreMulti-centre prospectiveproof-of-conceptFour tertiary PD centres(Czech Republic, Spain, Canada, Portugal)N = 631-wk group: N = 3524-wk group: N = 28Parkinson’s diseaseIPD diagnosis (UK Brain Bank criteria)No prior EMST experienceSpiroGym mHealth appSmartphone app + EMST150 expiratory trainerMicrophone-connected trainer for sound monitoringReal-time sound level detection via microphoneTraining data recording and storageVisual curve displaySmartphone + EMST150 deviceVisual biofeedback:Curve display showing sound level during expirationHistorical data reviewTraining diary5 sets × 5 forced expirations1-wk: 5×/day × 5/wk × 1 wk24-wk: intensive 8 wk (5 × 5, 5/wk) + maintenance 16 wk (5 × 5, 2/wk)Fortnightly resistance adjustmentPre–post comparison (no control group)24 wk (for 24-wk group)Primary: SEHEPS1-wk: 48.0 → 55.5, d = 0.48, P = 0.02 ↑24-wk: 43.5 → 56.0, d = 0.57, P = 0.002 ↑Secondary:SUS: 80–85 (A − /A+ grade)Adherence: 100% (8-wk), 50% >800 (maintenance)No adverse eventsKim (2019)USASystem development+ pilot validationPurdue UniversityN = 30 (validation)N = 1 (PD pilot)Healthy: 26–79 yPD with oropharyngeal dysphagia (pilot)Pharyngeal constriction insufficiencyModerate vallecular residueFlexible submental sensor patchStretchable sEMG + strain gauge + wirelessRemote monitoring capabilityFlexible polyimide substrate (13 μm)Gold-plated sEMG electrodes (5 mm)Piezoresistive strain gaugeBluetooth HC-0625-bit ADCSlope-based swallow detectionReal-time visual biofeedback:Circular diagram showing normalised sEMG amplitude (upper) and burst duration (lower)Remote data transmission to clinicianSingle validation sessionSwallows: 5 mL/10 mL thin liquid, 5 cc puddingConditions: regular, effortful, MendelsohnCommercial electrodes (Red Dot, 3 M)NA (validation only)System validation:Signal correlation >0.95Response delay <20 msEffortful swallow: >20% maxMendelsohn: amplitude +200%, duration +270%Cost: ~$13.92 (patch) + $52 (unit)Reusable 5–10 timesComfort: acceptable 26–79 yBattel (2023)ItalyQuasi-experimentalwithin-subject feasibilityAcute hospitalEnroled: N = 12Completed: N = 10Male: 66.7%Age: 69.1 ± 6.24 yIdiopathic Parkinson’s disease7.25 ± 2.6 y since diagnosisDysphagia confirmed by FEESPAS > 2; FOIS-It >4Hoehn and Yahr stage 3 (mean)NeuroTrac MyoPlus ProsEMG biofeedback systemGame-based visual feedbackNeuroTrac MyoPlus Pro device3 surface electrodes (2 submental + 1 reference)Two programmes: wave display + plane gameDell laptopReal-time visual biofeedback:‘Open display’ – activation wave‘Plane game’ – gamified swallowingDelayed verbal feedback (3–4 s) per motor learning principles60 min × 5/wk × 4 wkTotal: 20 hReps increased weekly: 18 → 271 h after PD medicationWithin-subject pre–post (2 pre + 2 post assessments)16 wk (4 wk Tx + 12 wk F/U)Primary: FEESOral intake: 95% CI 4.70–5.50, P < 0.05 ↑Saliva residue: 95% CI 2.14–3.15, P < 0.05 ↑Solid residue: 95% CI 2.4–3.5, P < 0.05 ↑Maintained at 3 monthsSecondary:PAS improved all bolus typesUnexpected: voice and attention improvedAdherence: 83.3%Wang (2023)ChinaPragmatic single-blind RCTHospital → home transitionN = 109IG = 51, CG = 50Age: 66–68 yMale: 68–74%First ischaemic stroke≥1 wk post-strokeDysphagia confirmed (WST + SLP)FOIS 1–6Internet-literateWeChat mini programmeDigital coaching platformmHealth + video coaching + social featuresWeChat mini programme:Instructional videosExercise remindersCompletion recordingNon-adherence queryPeer support forumSmartphone with WeChatVideo-based exercise instructionCoach messaging via WeChatPeer support forumAutomated remindersExercises: effortful swallow, pitch glide, ice stimulationPhase 1: 20 min education in hospitalPhase 2: 6 wk post-dischargeWeekly video consultationsDaily exercisesCoach: ≥5 y experience nurseStandard care:Swallowing informationAdvice at discharge1-month appointmentNo video/platform6 wk (Tx period)Primary: FOIS3 wk: 25% vs 10% FOIS-7, P = 0.036 ↑6 wk: 35% vs 20% FOIS-7, P = 0.083 (NS)Secondary:SWAL-QOL: P = 0.032 ↑Pneumonia: 4% vs 14%, NSAdherence: 37% vs 12%, P = 0.003 ↑Shieh (2023)TaiwanSystem development+ repeated measuresLaboratory + institutional care(Chang Gung University)Enroled: N = 33Completed: N = 30Age: 28.6 ± 7.1 y (21–50 y)Healthy volunteersNo dysphagia signsSystem validation populationMulti-sensor respiratory–swallow telecare systemWearable + real-time monitoring + automated warningSubmental sEMG electrodesNasal airflow cannulaFSR for thyroid movementPortable signal collectorBluetooth transmissionSlope-based characteristic point detectionAutomated colour-coded warningsReal-time smartphone monitoringColour-coded warning:Red = level 1 high riskYellow = level 2 moderateGreen = safeRemote caregiver transmissionSingle validation session4 trunk angles: 5°, 15°, 30°, 45°3 food types × 3 trials = 36 swallowsBIOPAC MP150 as reference standardNA (single session)Primary: Trunk angle effectsSAD: P = 0.045 ↑TET: P = 0.037 ↑5° vs 45°: significantSecondary:Warning triggers at 45°:10 mL water: 4 yellow, 3 redPudding: 8 yellow, 12 redNo warnings at 5°–30°Suitable for nursing facility monitoringKrekeler (2023)USARCTAcademic medical centre + VA hospitalN = 19IG = 9, CG = 10Age: 69.33 yMale: 84.2%Post-stroke dysphagiaIschaemic or haemorrhagicWithin 6 months of strokePAS ≥ 3 or residue ≥2SwallowSTRONG® devicePortable tongue pressure biofeedback + tablet interfaceHandheld device4-sensor array mouthpiece (custom-moulded)Digital tablet displayReal-time tongue pressure trackingAutomated exercise completion monitoringReal-time visual biofeedback:Tablet screen showing pressure vs targetProgressive resistance:Wk 1: 60% 1RMWk 2–12: 80% 1RM20 reps × 3/day × 3/wk × 12 wk(10 anterior + 10 posterior)Total: ~108 sessionsFortnightly 1RM reassessment via telephoneUsual care:Compensatory treatmentsThickened liquidsPostural strategiesNon-tongue exercises12 wk (assessments at 0, 8, 12 wk)Primary: MIPNo significant between-group differenceLarge effect: anterior d = 0.95, posterior d = 0.96Secondary:FOIS: P = 0.04 ↑ (8 wk)Vallecular residue: d = 1.2PAS: NSSWAL-QOL: NSAttrition: 41%Kim (2020)South KoreaLongitudinal observational(single-arm)Community senior welfare centresEnroled: N = 11Completed: N = 8Age: 75.73 ± 4.96 yMale: 9.1%Community elderlyAge-related swallowing complaintsNo clinical diagnosisNo neurological diseaseMMSE normal365HealthySwallowingCoach AppTablet-based mHealth + IOPI measurementSamsung Galaxy Tab AIOPI for tongue pressureAutomated training data recordingReal-time pitch graphCompliance monitoring with calendarVisual biofeedback:Mirror for self-observationReal-time pitch graph (red = demo, blue = user)Video demonstrationsColour-coded calendarASSET protocol: EPS, EPG, ETR20 reps × 3/day × 5/wk × 8 wkTotal: 120 sessions~30 min/sessionFortnightly face-to-face mediationNA (single-arm feasibility)20 wk (8 wk Tx + 12 wk F/U)Primary: Swallowing tongue pressurePre 17.5 → Post 26.5 kPaz = −1.992, P = 0.046 ↑Secondary:Isometric pressure: NS12-wk maintenance: not sustainedAdherence: 72.7% completion83–100% compliance (completers)Human mediation appears necessaryHou (2024)ChinaThree-arm RCTDepartment of NeurologyN = 90CG = 30, EG1 = 30, EG2 = 30Age: 63–64 yMale: 76.7%Early post-stroke dysphagiaFirst stroke (CT/MRI confirmed)48 h to 2 wk post-strokeWater swallow test grade 3–5Age 55–75 yVitalStim® Plus 5923-3sEMG biofeedback + gamified interface + tDCSVitalStim® Plus systemsEMG electrodes (submental)Built-in computer for gametDCS stimulatorThreshold-based game controlAdjustable threshold per patientVisual biofeedback:‘Rabbit Climbing Mountain’ gameMendelsohn triggers rabbit ascentRabbit eats carrot on successEG1: sEMG-BF (50–150% threshold)EG2: EG1 + game20 min/day × 7 days/wk × 1–2 wkTotal: 7–14 sessionsCG: routine + tDCSEG1: CG + sEMG-BFEG2: EG1 + gameCG: routine swallowing therapy + tDCSEG1: CG + sEMG-BF1–2 wk (Tx only, no F/U)Primary: Clinical efficacy rateEG2 90% > EG1 76.7% > CG 60%P < 0.05 ↑Secondary:sEMG peak: CG 22 μV, EG1 30 μV, EG2 44 μVSwallowing duration: EG2 1.09 s < EG1 1.32 s < CG 1.52 sAll outcomes: EG2 > EG1 > CG, P < 0.05Adherence: 100%Dose–response relationship confirmedXu (2025)USAQuasi-experimental studyNorthwestern Memorial HospitalN = 20 (PD)+ N = 58 (healthy, algorithm)Age: 71.7 yPD duration: 5 yParkinson’s diseaseMild (60%) to moderate (40%) oropharyngeal dysphagiaSLP-confirmedADAM sensorAI-driven haptic biofeedback wearableSkin-mounted mechano-acoustic sensorCNN for swallow detection:2 1D conv layers (32 filters)Batch normalisation + dense layers3-axis accelerometer (1,600 Hz)ERM vibration motorBluetooth + edge computingHaptic biofeedback:ERM motor delivers gentle vibration when swallow frequency below thresholdVibration profile selected from 19 options tested on n = 20 healthySingle session crossover:15 min baseline → 15 min biofeedback or monitoring → 15 min crossoverTotal: ~45–60 minWithin-subject: ADAM monitoring only (no haptic)NA (single session feasibility)Primary: Swallow frequencyBaseline 0.77 → BF 1.10 swallows/min45% increase, P < 0.01 ↑All 20 patients improvedSecondary:Algorithm sensitivity: 95% vs SLP (r = 0.92)No skin adverse eventsFocus groups: positive on comfortFDA Drug Development Tool acceptanceLi (2016)TaiwanCase–control study(matched pairs)Academic tertiary hospitalN = 20IG = 10, CG = 10Age: 65–70 yMale: 55%Post-stroke dysphagia2–72 months post-strokeFOIS ≤ 3 (tube-dependent)100% NG tube insertionGame-based swallowing biofeedback‘Frog Catching Mosquito’Accelerometer-based with 3D animationMicro 3-axis accelerometer on thyroid cartilageThreshold-based game controlAdjustable threshold per patientSubmental ultrasound (Broadsound BS3C673) for outcomeVisual biofeedback:3D ‘Frog Catching Mosquito’ animationMosquito flies 15 s → effortful swallow (1 s hold) → frog catches mosquitoReal-time laryngeal acceleration display60 min × 3/wk × 16 sessions(30 min traditional + 30 min laryngeal exercise)Total: ~5–6 wkTraditional swallowing therapy + laryngeal exercise without biofeedback (verbal/gestural cues only)5–6 wk (Tx only, no F/U)Primary: Hyoid displacementIG: 11.37 → 14.45 mm, P = 0.002 ↑CG: 12.84 → 13.35 mm, NSBetween-group: P = 0.007 ↑Secondary:FOIS: IG 1.9 → 5.1 (P = 0.004), CG NSNG tube removal: 80% vs 20%OR = 6.00, P = 0.009Adherence: 100%Jansen (2025)NetherlandsNon-controlled prospectivefeasibility14-bed mixed ICUGelre HospitalsN = 20Age: 69.4 ± 8.2 yMale: 65%ICU stay: 52.0 ± 21.9 daysICU-acquired dysphagia with tracheostomyProlonged intubationICU-acquired weaknessFEES-confirmed (Murray ≥2 or PAS > 2)Rephagia Biofeedback (SilverFit)‘Kangaroo Game’sEMG biofeedback for ICU swallowingsEMG sensors on submental regionBluetooth to laptopIndividual calibration (3 swallows)Threshold-based game controlAutomatic swallow trackingVisual biofeedback:‘Kangaroo’ jumps when sEMG exceeds threshold10–25 swallow repetitions per sessionICU nurse supervised~20 min/day × 7/wkVariable duration (until decannulation or 3 wk)Median: 4 sessions (IQR 3–6.2)NA (single-arm feasibility)4 wk (1 month post-training chart review)Primary: Feasibility100% completion100% decannulationNo safety eventsSecondary:sEMG: 52 → 57 μV, NS89% understood importance50% motivated2 deaths (unrelated)Note: sEMG did not correlate with clinical improvementAlyanak (2025)TurkeyProspective assessor-blind RCTRehabilitation hospitalN = 33IG = 17, CG = 16Age: 61.93 ± 11.88 yMale: 66.7%Post-stroke dysphagia90.9% ischaemic strokeMedian 5 months post-strokeClinically confirmedVitalstim PlusRose game + Rabbit gamesEMG biofeedback + gamificationVitalstim Plus deviceBipolar sEMG electrodes (submental)Bandpass 100–370 Hz50 Hz notch filterRMS 0.2–2,000 μVComputer display + BluetoothVisual + auditory biofeedback:Rose game: rose fades when threshold exceeded (effortful swallow)Rabbit game: rabbit climbs hill (Mendelsohn)Threshold gradually increased50 min × 5/wk × 3 wk(30 min EMG-BF + 15 oral motor + 5 thermal–tactile)Total: 15 sessionsSame exercises with verbal feedback only (‘hold longer’, ‘squeeze harder’) + oral motor + thermal–tactile3 wk (post-treatment only)Primary: FOISIG: within-group P = 0.038 ↑CG: NSBetween-group: NSSecondary:PAS-liquid: IG P = 0.026 ↑PAS-semisolid: P = 0.031 (between) ↑DOSS: P = 0.004 (between) ↑DHI: all subscales P < 0.05 ↑Bahia (2025)USAFeasibility study (PDSA framework)Shirley Ryan AbilityLabN = 22 (survey)N = 2 (pilot patients)Patient 1: 56 y, FPatient 2: 33 y, MPost-stroke dysphagia (subacute)Moderate–severeMASA + MBSS confirmedAble to swallow ≥5 mL liquidEAT-STARS with Mobili-T sEMGSkill-based swallowing trainingMotor learning principlesMobili-T sEMG deviceiPad displayAuto-calibration (110% peak from 5 baseline swallows)10 s timing display with dashed effort line + 5 s arrowReal-time visual sEMG biofeedback:Effort level + timing displayModules: Learn → Identify (card recognition) → Practice (10 swallows) → Challenge → Combo80% accuracy to progressUp to 8 sessions over 2 wk4–5 sessions/wkPatient 1: 6 sessionsPatient 2: 8 sessions (incomplete)NA (within-subject feasibility)2 wk (post-treatment only)Primary: Acceptability/feasibilityAIM: 90%IAM: 87.5%FIM: 83.4%All exceeded 80% thresholdEarly efficacy:MASA improved both patientsMBSImP improved both patientsEAT-10, VAS decreasedMotor learning principles successfully incorporatedKantarcigil (2025)USAQualitative studySemi-structured interviewsRemote/home-basedN = 12(10 PD patients + 2 caregivers)Parkinson’s diseasePotential swallowing concernsEstablished PD diagnosisNovel wearable sensor+ iOS applicationSwallowing and breathing monitoringWearable neck sensor with adhesiveCustomised iOS appSwallowing/breathing detection algorithmsLongitudinal tracking functionalitySensor kit shipped to homesVisual feedback via iOS appData visualisationParticipant-suggested: add reminder vibrations to prompt swallowingQualitative interview sessionSensor kit exploration → semi-structured interviewNA (qualitative design)NA (cross-sectional)5 major themes:1. Sensor/adhesive: comfort, size, placement2. Public wearing: social acceptability3. App features: interface usability4. Recommendations: add vibration reminders5. Benefits/drawbacks: tracking valued, sensory distraction concernsKey insight: affordability and real-life usability concernsWang (2026)ChinaRCTHospitalN = 30IG = 15, CG = 15Age: ~60 yMale: 86.7%Post-stroke adults with dysphagiaVFSS confirmedE-SW-E pattern <40%Bimodal signal biofeedback systemRespiratory–swallow coordination trainingRespiratory + sEMG monitoringRespiratory signal monitoringNasal airflow sensor (AFM3000-200)sEMG sensorsAccelerometer (ADXL357)Visual display systemReal-time RSC pattern detectionCoordination pattern analysisReal-time visual biofeedback:Respiratory–swallow coordination patterns displayedTarget: optimal EXP-SW-EXP patternThree-phase training protocol: Recognise → Learn → Master30 min/day × 6/wk × 2 wkTotal: 12 sessions+ conventional rehabilitationConventional dysphagia rehabilitation only6 wk (2 wk Tx + 4 wk F/U)Primary: FOIS + PASFOIS: both improved, IG > CG (P < 0.001) ↑ ≥2-point FOIS increase: IG > CG (P < 0.01) ↑ PAS ≥ 6 (aspiration): IG < CG (P < 0.001) ↓ Secondary:Optimal RSC pattern: IG > CG (P < 0.001) ↑ Maintained at 1-month F/UStarmer (2023)USAMulti-centre RCTStanford + Massachusetts General + Fox ChaseN = 91IG = 44, CG = 47Age: ~60 yMale: ~86%HNCAt-risk for radiation-induced dysphagiaProphylactic interventionHNC Virtual Coach appmHealth for prophylactic swallowing exerciseiOS application (Vibrent Health)Smartphone/tablet with iOS appVideo-guided exercises (led by Heather Starmer)Daily challenge trackingAutomated remindersSymptom-based intensity adjustmentVisual feedback:Video demonstrationsExercises: jaw stretches, effortful swallow, Mendelsohn, MasakoProgress tracking with daily challengesNutrition tutorialsEmpowerment content7 wk during radiationDaily exercises (self-paced)Total: ~49 sessionsStandard clinical care + paper exercise logs (same exercise protocol)10–14 wk (7 wk Tx + 2–3 month F/U)Primary: Multiple measuresAdherence: app group tended betterMDADI (QOL): adherence associated with better scoresPhysiological function: NS at 2–3 monthsKey finding: app improved adherence and QOL but not physiological outcomesNordio (2022)ItalyPilot RCTIRCCS San Camillo Hospital, VeniceN = 17IG = 9, CG = 7 (1 dropout)Post-stroke adultsAge: 70.12 ± 10.34 yMale: 75%Post-stroke dysphagiaFOIS ≤ 5 (restricted intake)FEES-confirmedSubacute (>6 wk post-stroke)Rehaingest sEMG biofeedbackProComp Infiniti system(Thought Technology)ProComp Infiniti encodersEMG electrodes (submental)BioGraph Infiniti softwareVisual feedback generationThreshold-based progressive targetsVisual biofeedback:Real-time sEMG waveform displaySubmental muscle activationExercises: effortful swallowing, Mendelsohn manoeuvre, supraglottic swallow, MasakoProgressive targets by therapist40 min training × 5/wk × 5 wk (with rest periods, 1 h total/day)Total: 25 sessionsStandard swallowing rehabilitation (same exercises with verbal feedback only)12 wk (4 wk Tx + 8 wk F/U)Primary: FOISImproved in all patients regardless of groupSecondary:P-Score (pharyngeal residue): IG betterPAS (aspiration risk): IG betterEffects stable at 2-month F/UAdherence: 94% (16/17)1 dropout (somnolence)1RM one-repetition maximum, ADAM Automated Detection and Assessment of Movement, ADC analogue-to-digital converter, AI artificial intelligence, AI-VG artificial intelligence video game, AIM Acceptability of Intervention Measure, ASSET Assessment of Swallowing Exercises in the Treatment, BCI brain–computer interface, BF biofeedback, CG control group, CI confidence interval, CNN convolutional neural network, CT computed tomography, CTAR chin tuck against resistance, CWT continuous wavelet transform, DDK diadochokinetic, DHI Dysphagia Handicap Index, DL deep learning, DOSS Dysphagia Outcome and Severity Scale, DRACE Dysphagia Risk Assessment for the Community-dwelling Elderly, DSS decision support system, DSRS Dysphagia Severity Rating Scale, EAT-10 Eating Assessment Tool-10, EAT-STARS Eating and Swallowing Training for Accurate, Reliable and Standardised skill-building, EEG electroencephalography, EG experimental group, EMD empirical mode decomposition, EMST expiratory muscle strength training, EPS effortful pitch swallow, EPG effortful pitch glide, ERM eccentric rotating mass, ERT ensemble of regression trees, ETR effortful tongue retraction, EXP-SW-EXP expiration–swallow–expiration, F female, FDA US Food and Drug Administration, FEES fibreoptic endoscopic evaluation of swallowing, FIM Feasibility of Intervention Measure, FOIS Functional Oral Intake Scale, FOV field of view, FR-VG facial recognition video game, FSR force sensing resistor, FSS2 Flow Short Scale-2, F/U follow-up, gbCTAR game-based chin tuck against resistance, GOHAI General Oral Health Assessment Index, GUSS Gugging Swallowing Screen, H&Y Hoehn and Yahr, HCPs healthcare professionals, HLE head-lift exercise, HNC head and neck cancer, HRM heart rate monitor, HRV heart rate variability, IAM Intervention Appropriateness Measure, ICA independent component analysis, ICU intensive care unit, IG intervention group, IMU inertial measurement unit, IOPI Iowa Oral Performance Instrument, IPD idiopathic Parkinson's disease, IQR interquartile range, IR infrared, ITR information transfer rate, KNN k-nearest neighbours, LED light-emitting diode, M male, MASA Mann Assessment of Swallowing Ability, MBSS modified barium swallow study, MBSImP Modified Barium Swallow Impairment Profile, mCSES modified Computer Self-Efficacy Scale, MDADI MD Anderson Dysphagia Inventory, mHealth mobile health, MIP maximum isometric pressure, ML machine learning, MMSE Mini-Mental State Examination, MNA-SF Mini Nutritional Assessment-Short Form, MRI magnetic resonance imaging, NA not applicable, NASA-TLX NASA Task Load Index, NG nasogastric, NR not reported, NS not significant, OHIP-14 Oral Health Impact Profile-14, OR odds ratio, OWL Web Ontology Language, PAS Penetration–Aspiration Scale, PD Parkinson's disease, PDSA Plan-Do-Study-Act, QOL quality of life, RCT randomised controlled trial, RF random forest, RGB red green blue, RMS root mean square, ROM range of motion, RSC respiratory–swallow coordination, SAD swallow apnoea duration, SD standard deviation, SEHEPS Self-Efficacy for Home Exercise Program Scale, sEMG surface electromyography, SLP speech-language pathologist, SMS short message service, SNR signal-to-noise ratio, SSA Standardised Swallowing Assessment, SUS System Usability Scale, SVM support vector machine, SWAL-QOL Swallowing Quality of Life, SWRL Semantic Web Rule Language, TAM Technology Acceptance Model, tDCS transcranial direct current stimulation, TET total eating time, TMIS tongue–machine interaction system, Tx treatment, VA Veterans Affairs, VAS visual analogue scale, VDS Videofluoroscopic Dysphagia Scale, VFSS videofluoroscopic swallowing study, VR virtual reality, WST water swallowing test, wk week(s), y year(s).↑, statistically significant improvement; ↓, statistically significant deterioration; NS, not significant. Study characteristics, intervention technology, and outcomes Single-arm randomised pilot (feasibility) Home-based N = 32 Male: 21.9% Healthy middle-aged/elderly Pre-clinical (prevention) EAT-10 screened No swallowing disorders Tongue Training App AI mobile exergame Flow-based gameplay 3 speed levels (slow/medium/fast) TeachableMachine (Google) 1,200+ training images Keras/TensorFlow Lite + Unity Android smartphone + Garmin HRM Visual + audio feedback Game prompts (‘great’, ‘nice’) Music-synchronised rhythm Adaptive speed progression 30 min single session (feasibility testing) Within-subject speed comparison Primary: Engagement FSS2: P = 0.001 ↑ NASA-TLX: P < 0.001 Flow Distance: P < 0.001 Cronbach α = 0.802 Secondary: HRV measures: 25–58% significant K-means: 3 groups identified Adherence: 100% Medium speed optimal Jung (2024) South Korea RCT (3-arm) Home-based Enroled: N = 90 Completed: N = 76 App: 25, Non-app: 25, Control: 26 Male: 11.5–16% Community-dwelling elderly Age: 65–85 y Healthy population Smartphone exercise app Combined oral + whole-body exercise 4 categories: oral health, expert hygiene, oral exercises, systemic exercises Structured exercise guidance application SMS reminder system Smartphone with app Visual exercise demonstrations SMS reminders for adherence Oral + systemic exercise guidance Self-directed daily practice App: 60 min × 7/wk × 5 wk Non-app: 60 min × 2/wk × 5 wk Non-app: same exercises with supervised training 2×/wk Control: no exercise Primary: Tongue pressure (IOPI) Anterior: +4.48 kPa, P < 0.05 ↑ Posterior: +3.88 kPa, NS Cheek: +2.76 kPa, NS Secondary: Saliva flow: +0.71 g/min, P < 0.05 ↑ OHIP-14, GOHAI: NS Dropout: 8.9% (8/90) Benfield (2023) UK RCT (feasibility) Hospital (acute stroke unit) N = 27 IG = 12, CG = 15 Male: IG 50%, CG 13% Acute stroke with dysphagia Within 4 wk of onset Age: 73.3 ± 11.0 y Medically stable patients BiSSkiT system sEMG biofeedback software Strength training: 70% max amplitude Skill training: amplitude + timing targets within 30 s window NeuroTrac Simplex sEMG device Triode electrodes (submental) BiSSkiT software Computer + display Real-time visual sEMG waveform feedback Target amplitude thresholds Timing windows for skill training Swallowing exercises with biofeedback 45 min × 5/wk × 2 wk Total: 10 sessions Supervised hospital setting Feasibility: 84.6% completed >80% sessions 92% comfortable/very comfortable 75% appropriate frequency Clinical (within 2 wk): FOIS: +2 points ↑ DSRS: +3 points ↑ PAS 5 mL: +1 point ↑ At 90 days: no group differences Park (2019) South Korea RCT (assessor-blind) Hospital Enroled: N = 46 Completed: N = 37 IG = 20, CG = 17 Male: IG 65%, CG 50% Post-stroke dysphagia VFSS-confirmed Age: IG 60.95 ± 11.19 y, CG 59.45 ± 9.34 y gbCTAR system ‘Eating Little Fish’ game Game-based CTAR with elastic resistance bar LES 100 device with elastic resistance bar Motion sensing game software Display monitor for game Real-time visual game feedback ‘Eating Little Fish’ metaphor Adjustable resistance levels CTAR exercises with gamification 30 min × 5/wk × 4 wk Total: 20 sessions Supervised training Head-lift exercise (HLE) Conventional Shaker exercise (isometric + isotonic) Primary: Swallowing function VDS: both groups P < 0.05 ↑ PAS: both groups P < 0.05 ↑ FOIS: both groups P < 0.05 ↑ No between-group difference Secondary: Engagement Motivation: gbCTAR > HLE, P < 0.05 Interest/enjoyment: P < 0.05 Physical demand: gbCTAR < HLE, P < 0.001 Dropout: 19.6% (9/46) Khan (2024) Australia/Canada Feasibility study (system development) Community/home N = 12 Male: 66% Age: 17.5–23 y Healthy young adults (system validation) No neurological conditions No swallowing disorders TMIS system Tongue–machine interaction IR-based tongue tracking for computer control Cursor/game control via tongue Nintendo Wii remote IR camera Reflective tape on silicone tongue device IR LED array FreeTrack software Dwell Clicker2 Computer + display Visual cursor control based on tongue position Game software interaction Alarm system activation Text processor control Tongue acts as computer pointer System testing only No fixed protocol Precision and ROM assessment Comparison systems: Finger touchscreen Tongue capacitive touchscreen System performance: Horizontal FOV: 42° Vertical FOV: 33° Cursor efficiency: 36.5% (vs finger baseline) 33.2% > tongue capacitive Click accuracy: 70% ITR: 130 (6 commands) Cost: 100% stretchability Bluetooth wireless (BioRadio) Threshold classification algorithm Bandpass 30–150 Hz, RMS smoothing Visual biofeedback: ‘jumping ball’ game Ball jumps between plates on successful swallow detection Real-time sEMG waveform display Submental electrode placement Single session testing Liquid swallowing: 20 mL water every 3 s, 10 trials Dry swallowing during game Swallow detection: Liquid: 100% accuracy (both types) False positive: 3% vs 5% (rigid) Game performance: Success: 94.7 ± 4.4% vs 96.2 ± 1.0% False positive: 16.8% vs 22.3% Motion artefact: ~6× lower (skin-like) SNR: 21.4 ± 0.6 dB (equivalent) Acceptability: 6.5/7 vs 4/7 (rigid) Zhang (2024) China/Hong Kong Mixed-methods (feasibility) Rehabilitation centre N = 19 (11 patients + 8 HCPs) Patients: 81.8% male HCPs: 100% female Post-stroke dysphagia (inpatient) Onset: 41.36 ± 29.48 days Age: 60.18 ± 10.91 y GUSS severity: Mild–moderate 54.5%, moderate–severe 36.4% Video game swallowing system MediaPipe AI + video game 3 mini-games: Lip (rabbit), Tongue (maze), Jaw (bird + ball squeeze) MediaPipe face mesh algorithm 468 facial landmarks, 3D real-time tracking RGB camera + computer display + headset Real-time visual feedback via game character animations Audio feedback Oral exercises (lip, tongue, jaw) through game control 1-wk familiarisation (before qualitative interview) Non-standardised user experience Technology acceptance (TAM): Perceived usefulness: Rehabilitation benefits valued Increased motivation Perceived ease of use: Visual feedback appreciated Digital literacy challenges (elderly) Intention to use: High willingness Background supervision needed Accuracy improvements suggested Kuramoto (2016) Japan Device validation (technical feasibility) University laboratory N = 11 Male: 45.5% Age: 71.4 y (mean) Dysphagia patients Referred for VFSS examination Able to swallow 3 mL thickened barium Swallowscope 2.0 Smartphone-based wearable swallowing monitor LED visual feedback + remote display for caregivers Neck-mounted contact microphone 6-axis IMU (accelerometer + gyroscope) LED visualisation unit Android smartphone Template matching algorithm for swallow sound IMU-based posture detection LED visual feedback (colour-coded status) Remote wireless display for caregivers Real-time monitoring interface Swallow sound + neck posture monitoring Single VFSS session Swallowing 3 mL thickened liquid barium (25% mixture) Simultaneous VFSS imaging Swallow detection: Accuracy: 75.9% (thickened liquid) Previous: 79.3% (thin liquid) Posture tracking: Mean error: 14.6° (SD 1.3) Bolus discrimination: Swallow duration differs by consistency (P < 0.05) Potential for nursing home mealtime assistance Kim (2020) South Korea Mixed-methods pilot (feasibility) Community senior centres N = 11 Female: 90.9% Age: 75.7 ± 3.93 y (67–83) Community-dwelling elderly At-risk for dysphagia (presbyphagia) No diagnosed dysphagia Preventive intervention target 365 Healthy Swallowing Coach App mHealth tablet application 3 exercises: EPS, EPG, ETR (effortful swallowing exercises) Galaxy Tab A SM-P580 (with S-pen) App-based exercise guidance Video demonstration Automated reminders Training record tracking Visual: exercise videos, animated guides Audio instructions Progress tracking display EPS, EPG, ETR exercises 2 sets × 10 reps = 60 reps/exercise/day 8 wk 5 days/wk 3 sessions/day (AM/PM/EVE) Total: 120 sessions ~15–20 min/session Usability (SUS): High-potential: ‘acceptable’ range Low-potential: ‘marginal’ range Self-efficacy (mCSES): High-potential: 54.0 → 65.5, P = 0.04 ↑ Low-potential: NS Adherence: 72.7% (8/11) optimal dose Key finding: education >10 y + tech experience predicts success Aslan (2024) Turkey Feasibility study (BCI classification) Research laboratory (Erciyes University) N = 30 Male: 50% Healthy adults Healthy volunteers No neurological/psychiatric disorders No dysphagia EEG-based BCI system Brain–computer interface for tongue movement detection Motor imagery paradigm g.tec Nautilus EEG (20 channels, 10–20 system) ML: KNN, SVM, Decision Tree, Random Forest, AdaBoost, Bagging DL: CNN + CWT scalogram (256 × 256 × 3 RGB) Signal: EMD + 0.5–40 Hz bandpass + ICA 30 time/frequency features NA (classification study) No real-time feedback tested Motor imagery: rest vs imagined tongue protrusion 15 trials × 9 s each Single recording session EEG during rest and imagination phases No rehabilitation intervention Classification accuracy (rest vs imagination): Traditional ML: KNN: 79.4% SVM: 63.4% Ensemble methods: AdaBoost/Bagging/RF: 99.8% Deep learning: CNN + CWT: 83% Ensemble methods outperformed others Potential for BCI-based dysphagia rehabilitation Srp (2024) Multi-centre Multi-centre prospective proof-of-concept Four tertiary PD centres (Czech Republic, Spain, Canada, Portugal) N = 63 1-wk group: N = 35 24-wk group: N = 28 Parkinson’s disease IPD diagnosis (UK Brain Bank criteria) No prior EMST experience SpiroGym mHealth app Smartphone app + EMST150 expiratory trainer Microphone-connected trainer for sound monitoring Real-time sound level detection via microphone Training data recording and storage Visual curve display Smartphone + EMST150 device Visual biofeedback: Curve display showing sound level during expiration Historical data review Training diary 5 sets × 5 forced expirations 1-wk: 5×/day × 5/wk × 1 wk 24-wk: intensive 8 wk (5 × 5, 5/wk) + maintenance 16 wk (5 × 5, 2/wk) Fortnightly resistance adjustment Primary: SEHEPS 1-wk: 48.0 → 55.5, d = 0.48, P = 0.02 ↑ 24-wk: 43.5 → 56.0, d = 0.57, P = 0.002 ↑ Secondary: SUS: 80–85 (A − /A+ grade) Adherence: 100% (8-wk), 50% >800 (maintenance) No adverse events Kim (2019) USA System development + pilot validation Purdue University N = 30 (validation) N = 1 (PD pilot) Healthy: 26–79 y PD with oropharyngeal dysphagia (pilot) Pharyngeal constriction insufficiency Moderate vallecular residue Flexible submental sensor patch Stretchable sEMG + strain gauge + wireless Remote monitoring capability Flexible polyimide substrate (13 μm) Gold-plated sEMG electrodes (5 mm) Piezoresistive strain gauge Bluetooth HC-06 25-bit ADC Slope-based swallow detection Real-time visual biofeedback: Circular diagram showing normalised sEMG amplitude (upper) and burst duration (lower) Remote data transmission to clinician Single validation session Swallows: 5 mL/10 mL thin liquid, 5 cc pudding Conditions: regular, effortful, Mendelsohn System validation: Signal correlation >0.95 Response delay <20 ms Effortful swallow: >20% max Mendelsohn: amplitude +200%, duration +270% Cost: ~$13.92 (patch) + $52 (unit) Reusable 5–10 times Comfort: acceptable 26–79 y Battel (2023) Italy Quasi-experimental within-subject feasibility Acute hospital Enroled: N = 12 Completed: N = 10 Male: 66.7% Age: 69.1 ± 6.24 y Idiopathic Parkinson’s disease 7.25 ± 2.6 y since diagnosis Dysphagia confirmed by FEES PAS > 2; FOIS-It >4 Hoehn and Yahr stage 3 (mean) NeuroTrac MyoPlus Pro sEMG biofeedback system Game-based visual feedback NeuroTrac MyoPlus Pro device 3 surface electrodes (2 submental + 1 reference) Two programmes: wave display + plane game Dell laptop Real-time visual biofeedback: ‘Open display’ – activation wave ‘Plane game’ – gamified swallowing Delayed verbal feedback (3–4 s) per motor learning principles 60 min × 5/wk × 4 wk Total: 20 h Reps increased weekly: 18 → 27 1 h after PD medication Primary: FEES Oral intake: 95% CI 4.70–5.50, P < 0.05 ↑ Saliva residue: 95% CI 2.14–3.15, P < 0.05 ↑ Solid residue: 95% CI 2.4–3.5, P < 0.05 ↑ Maintained at 3 months Secondary: PAS improved all bolus types Unexpected: voice and attention improved Adherence: 83.3% Wang (2023) China Pragmatic single-blind RCT Hospital → home transition N = 109 IG = 51, CG = 50 Age: 66–68 y Male: 68–74% First ischaemic stroke ≥1 wk post-stroke Dysphagia confirmed (WST + SLP) FOIS 1–6 Internet-literate WeChat mini programme Digital coaching platform mHealth + video coaching + social features WeChat mini programme: Instructional videos Exercise reminders Completion recording Non-adherence query Peer support forum Smartphone with WeChat Video-based exercise instruction Coach messaging via WeChat Peer support forum Automated reminders Exercises: effortful swallow, pitch glide, ice stimulation Phase 1: 20 min education in hospital Phase 2: 6 wk post-discharge Weekly video consultations Daily exercises Coach: ≥5 y experience nurse Standard care: Swallowing information Advice at discharge 1-month appointment No video/platform Primary: FOIS 3 wk: 25% vs 10% FOIS-7, P = 0.036 ↑ 6 wk: 35% vs 20% FOIS-7, P = 0.083 (NS) Secondary: SWAL-QOL: P = 0.032 ↑ Pneumonia: 4% vs 14%, NS Adherence: 37% vs 12%, P = 0.003 ↑ Shieh (2023) Taiwan System development + repeated measures Laboratory + institutional care (Chang Gung University) Enroled: N = 33 Completed: N = 30 Age: 28.6 ± 7.1 y (21–50 y) Healthy volunteers No dysphagia signs System validation population Multi-sensor respiratory–swallow telecare system Wearable + real-time monitoring + automated warning Submental sEMG electrodes Nasal airflow cannula FSR for thyroid movement Portable signal collector Bluetooth transmission Slope-based characteristic point detection Automated colour-coded warnings Real-time smartphone monitoring Colour-coded warning: Red = level 1 high risk Yellow = level 2 moderate Green = safe Remote caregiver transmission Single validation session 4 trunk angles: 5°, 15°, 30°, 45° 3 food types × 3 trials = 36 swallows Primary: Trunk angle effects SAD: P = 0.045 ↑ TET: P = 0.037 ↑ 5° vs 45°: significant Secondary: Warning triggers at 45°: 10 mL water: 4 yellow, 3 red Pudding: 8 yellow, 12 red No warnings at 5°–30° Suitable for nursing facility monitoring Krekeler (2023) USA RCT Academic medical centre + VA hospital N = 19 IG = 9, CG = 10 Age: 69.33 y Male: 84.2% Post-stroke dysphagia Ischaemic or haemorrhagic Within 6 months of stroke PAS ≥ 3 or residue ≥2 SwallowSTRONG® device Portable tongue pressure biofeedback + tablet interface Handheld device 4-sensor array mouthpiece (custom-moulded) Digital tablet display Real-time tongue pressure tracking Automated exercise completion monitoring Real-time visual biofeedback: Tablet screen showing pressure vs target Progressive resistance: Wk 1: 60% 1RM Wk 2–12: 80% 1RM 20 reps × 3/day × 3/wk × 12 wk (10 anterior + 10 posterior) Total: ~108 sessions Fortnightly 1RM reassessment via telephone Usual care: Compensatory treatments Thickened liquids Postural strategies Non-tongue exercises Primary: MIP No significant between-group difference Large effect: anterior d = 0.95, posterior d = 0.96 Secondary: FOIS: P = 0.04 ↑ (8 wk) Vallecular residue: d = 1.2 PAS: NS SWAL-QOL: NS Attrition: 41% Kim (2020) South Korea Longitudinal observational (single-arm) Community senior welfare centres Enroled: N = 11 Completed: N = 8 Age: 75.73 ± 4.96 y Male: 9.1% Community elderly Age-related swallowing complaints No clinical diagnosis No neurological disease MMSE normal 365HealthySwallowingCoach App Tablet-based mHealth + IOPI measurement Samsung Galaxy Tab A IOPI for tongue pressure Automated training data recording Real-time pitch graph Compliance monitoring with calendar Visual biofeedback: Mirror for self-observation Real-time pitch graph (red = demo, blue = user) Video demonstrations Colour-coded calendar ASSET protocol: EPS, EPG, ETR 20 reps × 3/day × 5/wk × 8 wk Total: 120 sessions ~30 min/session Fortnightly face-to-face mediation Primary: Swallowing tongue pressure Pre 17.5 → Post 26.5 kPa z = −1.992, P = 0.046 ↑ Secondary: Isometric pressure: NS 12-wk maintenance: not sustained Adherence: 72.7% completion 83–100% compliance (completers) Human mediation appears necessary Hou (2024) China Three-arm RCT Department of Neurology N = 90 CG = 30, EG1 = 30, EG2 = 30 Age: 63–64 y Male: 76.7% Early post-stroke dysphagia First stroke (CT/MRI confirmed) 48 h to 2 wk post-stroke Water swallow test grade 3–5 Age 55–75 y VitalStim® Plus 5923-3 sEMG biofeedback + gamified interface + tDCS VitalStim® Plus system sEMG electrodes (submental) Built-in computer for game tDCS stimulator Threshold-based game control Adjustable threshold per patient Visual biofeedback: ‘Rabbit Climbing Mountain’ game Mendelsohn triggers rabbit ascent Rabbit eats carrot on success EG1: sEMG-BF (50–150% threshold) EG2: EG1 + game 20 min/day × 7 days/wk × 1–2 wk Total: 7–14 sessions CG: routine + tDCS EG1: CG + sEMG-BF EG2: EG1 + game CG: routine swallowing therapy + tDCS EG1: CG + sEMG-BF Primary: Clinical efficacy rate EG2 90% > EG1 76.7% > CG 60% P < 0.05 ↑ Secondary: sEMG peak: CG 22 μV, EG1 30 μV, EG2 44 μV Swallowing duration: EG2 1.09 s < EG1 1.32 s < CG 1.52 s All outcomes: EG2 > EG1 > CG, P < 0.05 Adherence: 100% Dose–response relationship confirmed Xu (2025) USA Quasi-experimental study Northwestern Memorial Hospital N = 20 (PD) + N = 58 (healthy, algorithm) Age: 71.7 y PD duration: 5 y Parkinson’s disease Mild (60%) to moderate (40%) oropharyngeal dysphagia SLP-confirmed ADAM sensor AI-driven haptic biofeedback wearable Skin-mounted mechano-acoustic sensor CNN for swallow detection: 2 1D conv layers (32 filters) Batch normalisation + dense layers 3-axis accelerometer (1,600 Hz) ERM vibration motor Bluetooth + edge computing Haptic biofeedback: ERM motor delivers gentle vibration when swallow frequency below threshold Vibration profile selected from 19 options tested on n = 20 healthy Single session crossover: 15 min baseline → 15 min biofeedback or monitoring → 15 min crossover Total: ~45–60 min Primary: Swallow frequency Baseline 0.77 → BF 1.10 swallows/min 45% increase, P < 0.01 ↑ All 20 patients improved Secondary: Algorithm sensitivity: 95% vs SLP (r = 0.92) No skin adverse events Focus groups: positive on comfort FDA Drug Development Tool acceptance Li (2016) Taiwan Case–control study (matched pairs) Academic tertiary hospital N = 20 IG = 10, CG = 10 Age: 65–70 y Male: 55% Post-stroke dysphagia 2–72 months post-stroke FOIS ≤ 3 (tube-dependent) 100% NG tube insertion Game-based swallowing biofeedback ‘Frog Catching Mosquito’ Accelerometer-based with 3D animation Micro 3-axis accelerometer on thyroid cartilage Threshold-based game control Adjustable threshold per patient Submental ultrasound (Broadsound BS3C673) for outcome Visual biofeedback: 3D ‘Frog Catching Mosquito’ animation Mosquito flies 15 s → effortful swallow (1 s hold) → frog catches mosquito Real-time laryngeal acceleration display 60 min × 3/wk × 16 sessions (30 min traditional + 30 min laryngeal exercise) Total: ~5–6 wk Primary: Hyoid displacement IG: 11.37 → 14.45 mm, P = 0.002 ↑ CG: 12.84 → 13.35 mm, NS Between-group: P = 0.007 ↑ Secondary: FOIS: IG 1.9 → 5.1 (P = 0.004), CG NS NG tube removal: 80% vs 20% OR = 6.00, P = 0.009 Adherence: 100% Jansen (2025) Netherlands Non-controlled prospective feasibility 14-bed mixed ICU Gelre Hospitals N = 20 Age: 69.4 ± 8.2 y Male: 65% ICU stay: 52.0 ± 21.9 days ICU-acquired dysphagia with tracheostomy Prolonged intubation ICU-acquired weakness FEES-confirmed (Murray ≥2 or PAS > 2) Rephagia Biofeedback (SilverFit) ‘Kangaroo Game’ sEMG biofeedback for ICU swallowing sEMG sensors on submental region Bluetooth to laptop Individual calibration (3 swallows) Threshold-based game control Automatic swallow tracking Visual biofeedback: ‘Kangaroo’ jumps when sEMG exceeds threshold 10–25 swallow repetitions per session ICU nurse supervised ~20 min/day × 7/wk Variable duration (until decannulation or 3 wk) Median: 4 sessions (IQR 3–6.2) Primary: Feasibility 100% completion 100% decannulation No safety events Secondary: sEMG: 52 → 57 μV, NS 89% understood importance 50% motivated 2 deaths (unrelated) Note: sEMG did not correlate with clinical improvement Alyanak (2025) Turkey Prospective assessor-blind RCT Rehabilitation hospital N = 33 IG = 17, CG = 16 Age: 61.93 ± 11.88 y Male: 66.7% Post-stroke dysphagia 90.9% ischaemic stroke Median 5 months post-stroke Clinically confirmed Vitalstim Plus Rose game + Rabbit game sEMG biofeedback + gamification Vitalstim Plus device Bipolar sEMG electrodes (submental) Bandpass 100–370 Hz 50 Hz notch filter RMS 0.2–2,000 μV Computer display + Bluetooth Visual + auditory biofeedback: Rose game: rose fades when threshold exceeded (effortful swallow) Rabbit game: rabbit climbs hill (Mendelsohn) Threshold gradually increased 50 min × 5/wk × 3 wk (30 min EMG-BF + 15 oral motor + 5 thermal–tactile) Total: 15 sessions Primary: FOIS IG: within-group P = 0.038 ↑ CG: NS Between-group: NS Secondary: PAS-liquid: IG P = 0.026 ↑ PAS-semisolid: P = 0.031 (between) ↑ DOSS: P = 0.004 (between) ↑ DHI: all subscales P < 0.05 ↑ Bahia (2025) USA Feasibility study (PDSA framework) Shirley Ryan AbilityLab N = 22 (survey) N = 2 (pilot patients) Patient 1: 56 y, F Patient 2: 33 y, M Post-stroke dysphagia (subacute) Moderate–severe MASA + MBSS confirmed Able to swallow ≥5 mL liquid EAT-STARS with Mobili-T sEMG Skill-based swallowing training Motor learning principles Mobili-T sEMG device iPad display Auto-calibration (110% peak from 5 baseline swallows) 10 s timing display with dashed effort line + 5 s arrow Real-time visual sEMG biofeedback: Effort level + timing display Modules: Learn → Identify (card recognition) → Practice (10 swallows) → Challenge → Combo 80% accuracy to progress Up to 8 sessions over 2 wk 4–5 sessions/wk Patient 1: 6 sessions Patient 2: 8 sessions (incomplete) Primary: Acceptability/feasibility AIM: 90% IAM: 87.5% FIM: 83.4% All exceeded 80% threshold Early efficacy: MASA improved both patients MBSImP improved both patients EAT-10, VAS decreased Motor learning principles successfully incorporated Kantarcigil (2025) USA Qualitative study Semi-structured interviews Remote/home-based N = 12 (10 PD patients + 2 caregivers) Parkinson’s disease Potential swallowing concerns Established PD diagnosis Novel wearable sensor + iOS application Swallowing and breathing monitoring Wearable neck sensor with adhesive Customised iOS app Swallowing/breathing detection algorithms Longitudinal tracking functionality Sensor kit shipped to homes Visual feedback via iOS app Data visualisation Participant-suggested: add reminder vibrations to prompt swallowing Qualitative interview session Sensor kit exploration → semi-structured interview 5 major themes: 1. Sensor/adhesive: comfort, size, placement 2. Public wearing: social acceptability 3. App features: interface usability 4. Recommendations: add vibration reminders 5. Benefits/drawbacks: tracking valued, sensory distraction concerns Key insight: affordability and real-life usability concerns Wang (2026) China RCT Hospital N = 30 IG = 15, CG = 15 Age: ~60 y Male: 86.7% Post-stroke adults with dysphagia VFSS confirmed E-SW-E pattern <40% Bimodal signal biofeedback system Respiratory–swallow coordination training Respiratory + sEMG monitoring Respiratory signal monitoring Nasal airflow sensor (AFM3000-200) sEMG sensors Accelerometer (ADXL357) Visual display system Real-time RSC pattern detection Coordination pattern analysis Real-time visual biofeedback: Respiratory–swallow coordination patterns displayed Target: optimal EXP-SW-EXP pattern Three-phase training protocol: Recognise → Learn → Master 30 min/day × 6/wk × 2 wk Total: 12 sessions + conventional rehabilitation Primary: FOIS + PAS FOIS: both improved, IG > CG (P < 0.001) ↑ ≥2-point FOIS increase: IG > CG (P < 0.01) ↑ PAS ≥ 6 (aspiration): IG < CG (P < 0.001) ↓ Secondary: Optimal RSC pattern: IG > CG (P < 0.001) ↑ Maintained at 1-month F/U Starmer (2023) USA Multi-centre RCT Stanford + Massachusetts General + Fox Chase N = 91 IG = 44, CG = 47 Age: ~60 y Male: ~86% HNC At-risk for radiation-induced dysphagia Prophylactic intervention HNC Virtual Coach app mHealth for prophylactic swallowing exercise iOS application (Vibrent Health) Smartphone/tablet with iOS app Video-guided exercises (led by Heather Starmer) Daily challenge tracking Automated reminders Symptom-based intensity adjustment Visual feedback: Video demonstrations Exercises: jaw stretches, effortful swallow, Mendelsohn, Masako Progress tracking with daily challenges Nutrition tutorials Empowerment content 7 wk during radiation Daily exercises (self-paced) Total: ~49 sessions Primary: Multiple measures Adherence: app group tended better MDADI (QOL): adherence associated with better scores Physiological function: NS at 2–3 months Key finding: app improved adherence and QOL but not physiological outcomes Nordio (2022) Italy Pilot RCT IRCCS San Camillo Hospital, Venice N = 17 IG = 9, CG = 7 (1 dropout) Post-stroke adults Age: 70.12 ± 10.34 y Male: 75% Post-stroke dysphagia FOIS ≤ 5 (restricted intake) FEES-confirmed Subacute (>6 wk post-stroke) Rehaingest sEMG biofeedback ProComp Infiniti system (Thought Technology) ProComp Infiniti encoder sEMG electrodes (submental) BioGraph Infiniti software Visual feedback generation Threshold-based progressive targets Visual biofeedback: Real-time sEMG waveform display Submental muscle activation Exercises: effortful swallowing, Mendelsohn manoeuvre, supraglottic swallow, Masako Progressive targets by therapist 40 min training × 5/wk × 5 wk (with rest periods, 1 h total/day) Total: 25 sessions Primary: FOIS Improved in all patients regardless of group Secondary: P-Score (pharyngeal residue): IG better PAS (aspiration risk): IG better Effects stable at 2-month F/U Adherence: 94% (16/17) 1 dropout (somnolence) 1RM one-repetition maximum, ADAM Automated Detection and Assessment of Movement, ADC analogue-to-digital converter, AI artificial intelligence, AI-VG artificial intelligence video game, AIM Acceptability of Intervention Measure, ASSET Assessment of Swallowing Exercises in the Treatment, BCI brain–computer interface, BF biofeedback, CG control group, CI confidence interval, CNN convolutional neural network, CT computed tomography, CTAR chin tuck against resistance, CWT continuous wavelet transform, DDK diadochokinetic, DHI Dysphagia Handicap Index, DL deep learning, DOSS Dysphagia Outcome and Severity Scale, DRACE Dysphagia Risk Assessment for the Community-dwelling Elderly, DSS decision support system, DSRS Dysphagia Severity Rating Scale, EAT-10 Eating Assessment Tool-10, EAT-STARS Eating and Swallowing Training for Accurate, Reliable and Standardised skill-building, EEG electroencephalography, EG experimental group, EMD empirical mode decomposition, EMST expiratory muscle strength training, EPS effortful pitch swallow, EPG effortful pitch glide, ERM eccentric rotating mass, ERT ensemble of regression trees, ETR effortful tongue retraction, EXP-SW-EXP expiration–swallow–expiration, F female, FDA US Food and Drug Administration, FEES fibreoptic endoscopic evaluation of swallowing, FIM Feasibility of Intervention Measure, FOIS Functional Oral Intake Scale, FOV field of view, FR-VG facial recognition video game, FSR force sensing resistor, FSS2 Flow Short Scale-2, F/U follow-up, gbCTAR game-based chin tuck against resistance, GOHAI General Oral Health Assessment Index, GUSS Gugging Swallowing Screen, H&Y Hoehn and Yahr, HCPs healthcare professionals, HLE head-lift exercise, HNC head and neck cancer, HRM heart rate monitor, HRV heart rate variability, IAM Intervention Appropriateness Measure, ICA independent component analysis, ICU intensive care unit, IG intervention group, IMU inertial measurement unit, IOPI Iowa Oral Performance Instrument, IPD idiopathic Parkinson's disease, IQR interquartile range, IR infrared, ITR information transfer rate, KNN k-nearest neighbours, LED light-emitting diode, M male, MASA Mann Assessment of Swallowing Ability, MBSS modified barium swallow study, MBSImP Modified Barium Swallow Impairment Profile, mCSES modified Computer Self-Efficacy Scale, MDADI MD Anderson Dysphagia Inventory, mHealth mobile health, MIP maximum isometric pressure, ML machine learning, MMSE Mini-Mental State Examination, MNA-SF Mini Nutritional Assessment-Short Form, MRI magnetic resonance imaging, NA not applicable, NASA-TLX NASA Task Load Index, NG nasogastric, NR not reported, NS not significant, OHIP-14 Oral Health Impact Profile-14, OR odds ratio, OWL Web Ontology Language, PAS Penetration–Aspiration Scale, PD Parkinson's disease, PDSA Plan-Do-Study-Act, QOL quality of life, RCT randomised controlled trial, RF random forest, RGB red green blue, RMS root mean square, ROM range of motion, RSC respiratory–swallow coordination, SAD swallow apnoea duration, SD standard deviation, SEHEPS Self-Efficacy for Home Exercise Program Scale, sEMG surface electromyography, SLP speech-language pathologist, SMS short message service, SNR signal-to-noise ratio, SSA Standardised Swallowing Assessment, SUS System Usability Scale, SVM support vector machine, SWAL-QOL Swallowing Quality of Life, SWRL Semantic Web Rule Language, TAM Technology Acceptance Model, tDCS transcranial direct current stimulation, TET total eating time, TMIS tongue–machine interaction system, Tx treatment, VA Veterans Affairs, VAS visual analogue scale, VDS Videofluoroscopic Dysphagia Scale, VFSS videofluoroscopic swallowing study, VR virtual reality, WST water swallowing test, wk week(s), y year(s). ↑, statistically significant improvement; ↓, statistically significant deterioration; NS, not significant. Twenty-seven studies (87.1%) recruited independent samples. Four studies had potential sample overlap: two by Zhang et al. validated the same facial recognition system across different settings41,42, and two by Kim H et al. employed the same application with identical sample sizes (n = 11), raising the possibility of shared participants43,44. Findings were analysed at the study level rather than by pooling participant data. Population characteristics Among the 26 studies reporting clinical populations, a total of 1,012 participants were enroled. The distribution by aetiology was as follows: post-stroke dysphagia, 13 studies (41.9%); Parkinson’s disease, five (16.1%); age-related swallowing decline, four (12.9%); HNC, one; intensive care unit (ICU)-acquired dysphagia, one; neurogenic dysphagia, one; unclassified dysphagia, one; and healthy volunteers, five (16.1%). Participant age ranged from young adults (17.5–35 years) in validation studies to older populations (mean age 65–73 years) in clinical intervention trials. Sex distribution varied by aetiology and clinical setting, with the highest proportion of male participants observed in stroke studies (up to 84.2%) and the lowest in studies of community-dwelling older adults (9.1%). Technology systems and intervention characteristics The 31 included studies employed five principal categories of AI-enhanced technologies. sEMG biofeedback systems (n = 11, 35.5%)29,45–54 captured muscular activation patterns during swallowing via submental electrode placement; AI algorithms provided threshold-based detection, real-time signal processing, and visual feedback, technical complexity ranged from simple threshold-triggered devices to multi-sensor flexible electrode arrays. AI-driven computer vision and facial recognition technologies (n = 5, 16.1%)41,42,55–57 employed contactless, real-time tracking of facial or lingual movements; principal approaches included MediaPipe 468-point three-dimensional facial landmark detection, ensemble of regression trees (ERT) 68-point detection, and Google Teachable Machine-based image classification. Mobile health (mHealth) application platforms (n = 10, 32.3%)43,44,56–63 delivered interventions via smartphone or tablet applications, with functionality encompassing exercise video guidance, real-time feedback, progress tracking, and automated reminders. Machine learning and deep learning algorithms (n = 7, 22.6%)30,41,42,55–57,64 were deployed across a broad range of applications, from electroencephalography (EEG) signal classification for brain–computer interfaces to convolutional neural network (CNN)-based wearable swallowing detection. Three studies applied pre-trained or custom algorithms within clinical interventions41,42,57; the remaining four were confined to algorithm validation or acceptance testing in non-clinical or single-session settings. Other technologies included an ontology-based clinical decision support system (n = 1)65, an infra-red lingual motion tracking system (n = 1)66, and a lingual pressure biofeedback device (n = 1)67. Eleven studies (35.5%)31,41,42,47,49,50,52,54–56,68 incorporated gamification elements—visual rewards and real-time feedback—into rehabilitation training across multiple sensing modalities. These technologies employed diverse feedback modalities. Visual biofeedback was adopted in 30 studies (96.8%), delivered through real-time waveform displays, animated game interfaces, progress dashboards, and LED colour-coded status indicators. Auditory feedback supplemented visual modalities in six studies (19.4%). One study (3.2%) implemented haptic feedback. Multimodal feedback (≥2 modalities) was employed in seven studies (22.6%), predominantly in visual–auditory combinations. Intervention dose parameters differed considerably between clinical settings. In acute rehabilitation, sessions typically lasted 45–60 min, were delivered 5–7 sessions per week (high-intensity), and comprised 10–20 total sessions. In community and home-based settings, sessions lasted 15–30 min, were delivered 2–3 sessions per week (moderate frequency), and comprised 50–120 total sessions. Seven sEMG-based studies employed individualised progressive threshold adjustment, with initial thresholds set at 50–70% of baseline peak amplitude and increasing to 80–150%46,49–51,53,54,67. Human–AI collaboration modes Using the human–AI collaboration coding system (see ‘Methods’), the 31 studies were classified into five collaboration modes (Table 2). The distribution was as follows: Mode A, AI-augmented direct supervision (n = 11, 35.5%); Mode B, supervised autonomous practice (n = 5, 16.1%); Mode C, hybrid periodic consultation (n = 7, 22.6%); Mode D, AI-driven clinical decision validation (n = 3, 9.7%); Mode E, autonomous AI operation (n = 5, 16.1%). Table 2Human–AI collaboration role matrixStudyHuman Roles (H1–H6)AI Roles (A1–A6)Collaboration ModeAI:Human RatioZhang (2025)H1: System demonstration, patient education, home equipment installationH2: GUSS, SSA, FOIS, MNA-SF, SWAL-QOL at baseline and endpointsH3: Remote asynchronous weekly monitoringH4: Training parameters as neededA1: 468-point facial landmarks (MediaPipe), 3D position trackingA2: Exercise completion, repetition counting, movement qualityA3: Real-time visual game control, performance displayA4: Automated game-based rewardsMode B70:30Spoladore (2024)H1: Ontology development, clinical knowledge encodingH5: Dietary supervision, exception handlingH6: Recipe approval, safety verification, expert consensusA1: SWRL rule inference, severity classificationA3: Recipe list display, consistency warningsA5: Safe recipe identification, C1/C2 filteringA6: Patient profile storage, recipe databaseMode D79:21Zhang (2025)H1: Patient assessment, system calibrationH2: GUSS, FOIS evaluationH3: Direct in-person monitoring throughout sessionsH4: Real-time protocol modificationA1: Facial movement recognition (lip, tongue, chin)A2: Performance monitoring, quality assessmentA3: Game-based visual biofeedback, movement guidanceMode A50:50Chan (2024)H1: App installation, user training demonstrationH2: Baseline oral function, bite force, tongue pressureH3: Periodic fortnightly SLP consultationH4: Protocol modification as neededA1: 68-point ERT landmark detectionA2: Movement analysis, adherence monitoringA3: Visual demonstration, audio guidance, real-time accuracyA6: Progress logging, session trackingMode C72:28Su (2025)H1: Initial instruction, equipment set-up (HRM)H2: FSS2, NASA-TLX data collectionH3: Minimal (research monitoring only)A1: TeachableMachine tongue image recognition (1,200+ images)A2: Performance and engagement monitoringA3: Visual prompts, music-synchronised audioA4: Speed level adjustment, difficulty progressionMode E82:18Jung (2024)H1: Initial instruction for app group; supervised training for non-app (2×/wk)H2: Tongue pressure (IOPI), saliva flow, OHIP-14, GOHAIH3: Periodic check-ins for app group; direct supervision for non-appA3: Visual exercise demonstrations through app interfaceA6: SMS reminder scheduling; self-directed training support; adherence trackingMode C64:36Benfield (2023)H1: Electrode placement and calibration; system set-upH2: FOIS, DSRS, PAS, VFSS assessment at multiple time pointsH3: Direct supervision of all training sessionsH4: Target amplitude adjustment as neededA1: Real-time sEMG signal acquisition from submental musclesA2: Automated threshold monitoring; session data recordingA3: Visual biofeedback of sEMG waveforms; target amplitude displayMode A41:59Park (2019)H1: Device set-up; resistance adjustment; patient instructionH2: VDS, PAS, FOIS assessment via VFSSH3: Direct supervision during all training sessionsH4: Resistance level adjustment based on progressA1: Motion sensing of chin tuck movementsA2: Movement tracking; exercise completion monitoringA3: Real-time visual game feedback (‘Eating Little Fish’); gamificationMode A42:58Khan (2024)H1: System design and development; silicone device fittingH2: Precision testing; ROM assessment; performance metricsH3: Research monitoring only (no clinical supervision)A1: Real-time tongue position tracking via IR cameraA2: Cursor control translation; click activation detectionA3: Visual feedback (cursor movement); game interface responseA4: Tongue position mapping to screen coordinatesMode E88:12Lee (2017)H1: Electrode placement; individual threshold calibrationH2: Detection accuracy evaluation; acceptability interviewH3: Researcher supervision during testing sessionsA1: sEMG signal acquisition; bandpass filtering (30–150 Hz); RMS calculationA2: Threshold-based swallow detection (3-part algorithm)A3: Real-time visual biofeedback (‘jumping ball’ game); binary classificationMode A54:46Zhang (2024)H1: System demonstration; initial practice supervisionH2: Qualitative interview (TAM assessment)H3: Therapist verbal guidance during 1-wk familiarisationA1: Real-time facial landmark tracking (468 points, MediaPipe)A2: Movement detection for game controlA3: Visual feedback via game animations (rabbit, maze, bird); audio feedbackMode B60:40Kuramoto (2016)H1: Device placement; system set-up during VFSSH2: VFSS interpretation as gold standard; accuracy evaluationH3: Radiologist conducts VFSS; SLP interprets swallowing functionA1: Automatic swallow detection via sound template matching; IMU posture monitoringA2: Real-time bolus consistency estimation via duration analysisA3: LED visual feedback (colour-coded status); remote wireless displayMode D68:32Kim (2020)H1: Initial training on app use; troubleshootingH2: Usability assessment (SUS, mCSES); adherence monitoringH3: Weekly check-in calls (periodic remote support)A3: Automated exercise guidance via video demonstrations; animated guidesA6: Progress tracking; training record storage; automated remindersMode B60:40Aslan (2024)H1: EEG cap set-up; electrode placement (10–20 system)H2: EEG quality monitoring; data analysis; classification evaluationH3: Researcher instructions during motor imagery taskA1: EEG signal acquisition (20 channels); signal processing (EMD, bandpass, ICA)A2: Feature extraction (30 time/frequency domain features)A5: ML/DL classification of rest vs tongue imageryMode E88:12Srp (2024)H1: Initial device set-up, app installation trainingH2: SEHEPS assessment at time pointsH3: Fortnightly outpatient visits for resistance adjustmentH4: Progressive resistance modification based on performanceA1: Real-time sound level detection via microphoneA2: Training completion monitoring, adherence trackingA3: Visual feedback (sound curve display)A6: Automatic training data recording, training diaryMode C68:32Kim (2019)H1: Sensor patch placement, system calibrationH2: VFSS assessment, signal validationH3: Remote real-time data monitoringH5: Biofeedback data interpretation, treatment decisionsA1: Dual-differential sEMG recording, swallowing event detectionA2: Real-time strain/sEMG signal acquisitionA3: Circular diagram biofeedback displayA6: Wireless data transmission, event loggingMode B64:36Battel (2023)H1: Electrode placement, system set-upH2: FEES evaluation at pre/post/3-monthH3: Direct supervision of all biofeedback sessionsH4: Task difficulty adjustment, repetition progressionA1: Real-time sEMG signal acquisitionA2: Muscle activity detection, performance monitoringA3: Wave display + ‘plane game’ gamified feedbackA4: Game-based motivation mechanismMode A52:48Wang (2023)H1: Platform introduction, education sessionH2: Assessor-blind FOIS evaluationH3: Weekly video consultations with stroke coachH4: Individualised goal setting, problem-solvingA3: Video exercise demonstrations via appA4: Automated exercise remindersA6: Completion recording, non-adherence tracking, peer forum hostingMode C58:42Shieh (2023)H1: Sensor positioning, signal quality verificationH2: Monitoring for aspiration signs, BIOPAC validationH3: Caregiver receives alerts, adjusts feedingH5: Clinical response to warning alertsA1: Automatic swallowing event detection (sEMG, FSR, airflow)A2: Real-time respiratory–swallow coordination monitoringA3: Colour-coded warning display (red/yellow/green)A5: Unsafe pattern detection (piecemeal deglutition, prolonged apnoea)Mode D68:32Krekeler (2023)H1: Custom mouthpiece fitting, exercise trainingH2: MBSS assessment at 0, 8, 12 wkH3: Fortnightly telephone 1RM reassessmentH4: Progressive resistance adjustment (60% → 80%)A1: 4-sensor tongue pressure detectionA2: Target achievement trackingA3: Real-time visual biofeedback on tabletA6: Exercise completion loggingMode C44:56Kim (2020)H1: App installation, initial trainingH2: IOPI tongue pressure measurement at time pointsH3: Fortnightly face-to-face mediation sessionsH4: Technique verification and correctionA2: Automated compliance monitoringA3: Mirror function, real-time pitch graph, video demonstrationsA6: Colour-coded calendar progress tracking, training data recordingMode B56:44Hou (2024)H1: Electrode positioning, threshold calibrationH2: Water swallow test, FOIS, sEMG assessmentH3: Direct supervision of all hospital sessionsH4: Game difficulty adjustment based on progressA1: Real-time sEMG signal processingA2: Threshold-based swallow detectionA3: ‘Rabbit Climbing Mountain’ game feedbackA4: Adjustable threshold for personalised difficultyMode A50:50Xu (2025)H1: SLP places sensor on suprasternal notch; initial clinical evaluationH2: Determines eligibility; validates swallow events as gold standardH3: Focus group facilitation for user feedbackA1: CNN real-time swallow detection via edge computing (F1 = 0.89, sensitivity 95%)A2: Continuous swallow frequency monitoringA3: Haptic biofeedback via ERM vibration when below thresholdA4: Closed-loop operation without external devicesMode E82:18Li (2016)H1: Accelerometer placement on thyroid cartilage; threshold calibrationH2: Submental ultrasound for hyoid displacement; FOIS assessmentH3: Direct supervision of all sessionsH4: Acceleration threshold adjustment based on performanceA1: Accelerometer detection of laryngeal elevationA2: Real-time acceleration monitoring and performance trackingA3: Visual biofeedback via ‘Frog Catching Mosquito’ 3D animationA4: Adjustable threshold for personalised difficultyMode A48:52Jansen (2025)H1: SLP performs FEES; prescribes exercise parametersH2: Weekly SLP evaluation of progressH3: ICU nurses supervise daily Rephagia training (~20 min)H4: Exercise settings adjusted by SLP based on progressA1: sEMG signal acquisition from submental muscles; individual calibrationA2: Automatic tracking of swallow eventsA3: Visual biofeedback via ‘Kangaroo’ game — jumps when threshold exceededMode C60:40Alyanak (2025)H1: Electrode placement; skin preparation; threshold settingH2: VFSS assessment; FOIS, PAS, DOSS, DHI evaluationH3: Direct supervision of all 50 min sessionsH4: Progressive threshold adjustment based on performanceA1: Real-time sEMG signal processingA2: Threshold-based game control; progress trackingA3: Visual + auditory biofeedback via Rose and Rabbit gamesA4: Game difficulty progression signalsMode A44:56Bahia (2025)H1: Mobili-T sEMG set-up; guides patients through modulesH2: MASA, MBSS, FEES evaluations at multiple time pointsH3: Delivers programme; provides delayed verbal feedback (3–4 s post-swallow)H4: Adjusts task difficulty; progresses modules at 80% accuracyA1: sEMG acquisition from submental musclesA2: Auto-calibration (110% peak from 5 baseline swallows); timing displayA3: Real-time visual biofeedback of effort and timing on iPadA6: Module progression tracking (Learn → Identify → Practice → Challenge → Combo)Mode A48:52Kantarcigil (2025)H1: Study design; sensor kit development and shippingH2: Semi-structured interview; constant comparative analysisH6: Theme identification and interpretationA1: Swallowing and breathing event detection algorithmsA3: Visual feedback via iOS app interface; data visualisationA6: Longitudinal tracking functionality (design stage)A4: Participant-suggested: reminder vibrationsMode E79:21Wang (2026)H1: Sensor set-up; guides three-phase RSC protocolH2: VFSS assessments at baseline, 2 wk, 1 monthH3: Direct supervision of all training sessionsH4: Monitors patient progress towards optimal EXP-SW-EXP patternA1: Real-time respiratory and sEMG signal processingA2: Respiratory–swallow coordination pattern detection and analysisA3: Visual biofeedback showing RSC patternsA4: Target setting for optimal coordination patternMode A48:52Starmer (2023)H1: Baseline education and exercise instruction; prescribes protocolH2: MBSS, MDADI, FOIS, PSS-HN assessments at follow-upH3: Standard clinical care (periodic contact)A3: Automated exercise guidance via video demonstrationsA4: Exercise intensity adjustment based on symptomsA6: Daily challenge tracking; automated reminders; progress data storageMode C68:32Nordio (2022)H1: sEMG electrode placement (~15 min set-up); system calibrationH2: FEES assessments pre/post and 2-month follow-up; FOIS, PAS, P-ScoreH3: Direct supervision of all 45 min sessionsH4: Sets and adjusts sEMG target thresholds; provides verbal coachingA1: sEMG signal acquisition and amplification from submental musclesA2: Objective measurement of exercise performance and intensityA3: Real-time visual biofeedback showing sEMG waveformA6: sEMG amplitude data storageMode A38:62Human role codes: H1, initial set-up and training; H2, clinical assessment; H3, supervision and monitoring; H4, protocol adjustment; H5, clinical decision-making; H6, validation and verification. AI role codes: A1, signal detection and processing; A2, performance tracking and monitoring; A3, real-time feedback delivery; A4, adaptive parameter adjustment; A5, decision support and classification; A6, data management and storage. Collaboration modes: Mode A, AI-augmented direct (clinician present throughout, AI enhances clinical delivery); Mode B, supervised autonomous (AI delivers intervention autonomously, clinician reviews periodically); Mode C, hybrid periodic (patient uses AI-driven system independently, with periodic clinician consultations); Mode D, AI-driven validation (AI generates recommendations or detects events, clinician validates outputs); Mode E, fully autonomous (AI operates independently, clinician involved only for initial set-up and outcome evaluation). 1RM one-repetition maximum, ADAM Automated Detection and Assessment of Movement, AI-VG artificial intelligence video game, BCI brain–computer interface, BiSSkiT Biofeedback in Strength and Skill Training, CNN convolutional neural network, CTAR chin tuck against resistance, DDK diadochokinetic, DHI Dysphagia Handicap Index, DL deep learning, DOSS Dysphagia Outcome and Severity Scale, DRACE Dysphagia Risk Assessment for the Community-dwelling Elderly, DSS decision support system, DSRS Dysphagia Severity Rating Scale, EAT-10 Eating Assessment Tool-10, EAT-STARS Eating and Swallowing Training for Accurate, Reliable and Standardised skill-building, EEG electroencephalography, EMD empirical mode decomposition, EMST expiratory muscle strength training, EPS effortful pitch swallow, EPG effortful pitch glide, ERM eccentric rotating mass, ERT ensemble of regression trees, ETR effortful tongue retraction, EXP-SW-EXP expiration–swallow–expiration, FEES fibreoptic endoscopic evaluation of swallowing, FOIS Functional Oral Intake Scale, FR-VG facial recognition video game, FSR force sensing resistor, FSS2 Flow Short Scale-2, gbCTAR game-based chin tuck against resistance, GOHAI General Oral Health Assessment Index, GUSS Gugging Swallowing Screen, HNC head and neck cancer, HRM heart rate monitor, HRV heart rate variability, ICA independent component analysis, ICU intensive care unit, IMU inertial measurement unit, IOPI Iowa Oral Performance Instrument, IR infrared, LED light-emitting diode, MASA Mann Assessment of Swallowing Ability, MBSS modified barium swallow study, MBSImP Modified Barium Swallow Impairment Profile, mCSES modified Computer Self-Efficacy Scale, MDADI MD Anderson Dysphagia Inventory, mHealth mobile health, ML machine learning, MNA-SF Mini Nutritional Assessment-Short Form, NASA-TLX NASA Task Load Index, OHIP-14 Oral Health Impact Profile-14, OWL Web Ontology Language, PAS Penetration–Aspiration Scale, PSS-HN Performance Status Scale for Head and Neck cancer, RMS root mean square, ROM range of motion, RSC respiratory–swallow coordination, SEHEPS Self-Efficacy for Home Exercise Program Scale, sEMG surface electromyography, SLP speech-language pathologist, SMS short message service, SSA Standardised Swallowing Assessment, SUS System Usability Scale, SWAL-QOL Swallowing Quality of Life, SWRL Semantic Web Rule Language, TAM Technology Acceptance Model, tDCS transcranial direct current stimulation, TMIS tongue–machine interaction system, VDS Videofluoroscopic Dysphagia Scale, VFSS videofluoroscopic swallowing study, wk week(s).The AI:Human ratio represents the estimated proportional distribution of task responsibilities between AI-driven and human-delivered components, derived from the reported intervention protocols. Ratios were assigned based on the relative contribution of automated versus clinician-directed activities during the intervention period and do not reflect set-up or assessment phases. Human–AI collaboration role matrix H1: System demonstration, patient education, home equipment installation H2: GUSS, SSA, FOIS, MNA-SF, SWAL-QOL at baseline and endpoints H3: Remote asynchronous weekly monitoring H4: Training parameters as needed A1: 468-point facial landmarks (MediaPipe), 3D position tracking A2: Exercise completion, repetition counting, movement quality A3: Real-time visual game control, performance display A4: Automated game-based rewards H1: Ontology development, clinical knowledge encoding H5: Dietary supervision, exception handling H6: Recipe approval, safety verification, expert consensus A1: SWRL rule inference, severity classification A3: Recipe list display, consistency warnings A5: Safe recipe identification, C1/C2 filtering A6: Patient profile storage, recipe database H1: Patient assessment, system calibration H2: GUSS, FOIS evaluation H3: Direct in-person monitoring throughout sessions H4: Real-time protocol modification A1: Facial movement recognition (lip, tongue, chin) A2: Performance monitoring, quality assessment A3: Game-based visual biofeedback, movement guidance H1: App installation, user training demonstration H2: Baseline oral function, bite force, tongue pressure H3: Periodic fortnightly SLP consultation H4: Protocol modification as needed A1: 68-point ERT landmark detection A2: Movement analysis, adherence monitoring A3: Visual demonstration, audio guidance, real-time accuracy A6: Progress logging, session tracking H1: Initial instruction, equipment set-up (HRM) H2: FSS2, NASA-TLX data collection H3: Minimal (research monitoring only) A1: TeachableMachine tongue image recognition (1,200+ images) A2: Performance and engagement monitoring A3: Visual prompts, music-synchronised audio A4: Speed level adjustment, difficulty progression H1: Initial instruction for app group; supervised training for non-app (2×/wk) H2: Tongue pressure (IOPI), saliva flow, OHIP-14, GOHAI H3: Periodic check-ins for app group; direct supervision for non-app A3: Visual exercise demonstrations through app interface A6: SMS reminder scheduling; self-directed training support; adherence tracking H1: Electrode placement and calibration; system set-up H2: FOIS, DSRS, PAS, VFSS assessment at multiple time points H3: Direct supervision of all training sessions H4: Target amplitude adjustment as needed A1: Real-time sEMG signal acquisition from submental muscles A2: Automated threshold monitoring; session data recording A3: Visual biofeedback of sEMG waveforms; target amplitude display H1: Device set-up; resistance adjustment; patient instruction H2: VDS, PAS, FOIS assessment via VFSS H3: Direct supervision during all training sessions H4: Resistance level adjustment based on progress A1: Motion sensing of chin tuck movements A2: Movement tracking; exercise completion monitoring A3: Real-time visual game feedback (‘Eating Little Fish’); gamification H1: System design and development; silicone device fitting H2: Precision testing; ROM assessment; performance metrics H3: Research monitoring only (no clinical supervision) A1: Real-time tongue position tracking via IR camera A2: Cursor control translation; click activation detection A3: Visual feedback (cursor movement); game interface response A4: Tongue position mapping to screen coordinates H1: Electrode placement; individual threshold calibration H2: Detection accuracy evaluation; acceptability interview H3: Researcher supervision during testing sessions A1: sEMG signal acquisition; bandpass filtering (30–150 Hz); RMS calculation A2: Threshold-based swallow detection (3-part algorithm) A3: Real-time visual biofeedback (‘jumping ball’ game); binary classification H1: System demonstration; initial practice supervision H2: Qualitative interview (TAM assessment) H3: Therapist verbal guidance during 1-wk familiarisation A1: Real-time facial landmark tracking (468 points, MediaPipe) A2: Movement detection for game control A3: Visual feedback via game animations (rabbit, maze, bird); audio feedback H1: Device placement; system set-up during VFSS H2: VFSS interpretation as gold standard; accuracy evaluation H3: Radiologist conducts VFSS; SLP interprets swallowing function A1: Automatic swallow detection via sound template matching; IMU posture monitoring A2: Real-time bolus consistency estimation via duration analysis A3: LED visual feedback (colour-coded status); remote wireless display H1: Initial training on app use; troubleshooting H2: Usability assessment (SUS, mCSES); adherence monitoring H3: Weekly check-in calls (periodic remote support) A3: Automated exercise guidance via video demonstrations; animated guides A6: Progress tracking; training record storage; automated reminders H1: EEG cap set-up; electrode placement (10–20 system) H2: EEG quality monitoring; data analysis; classification evaluation H3: Researcher instructions during motor imagery task A1: EEG signal acquisition (20 channels); signal processing (EMD, bandpass, ICA) A2: Feature extraction (30 time/frequency domain features) A5: ML/DL classification of rest vs tongue imagery H1: Initial device set-up, app installation training H2: SEHEPS assessment at time points H3: Fortnightly outpatient visits for resistance adjustment H4: Progressive resistance modification based on performance A1: Real-time sound level detection via microphone A2: Training completion monitoring, adherence tracking A3: Visual feedback (sound curve display) A6: Automatic training data recording, training diary H1: Sensor patch placement, system calibration H2: VFSS assessment, signal validation H3: Remote real-time data monitoring H5: Biofeedback data interpretation, treatment decisions A1: Dual-differential sEMG recording, swallowing event detection A2: Real-time strain/sEMG signal acquisition A3: Circular diagram biofeedback display A6: Wireless data transmission, event logging H1: Electrode placement, system set-up H2: FEES evaluation at pre/post/3-month H3: Direct supervision of all biofeedback sessions H4: Task difficulty adjustment, repetition progression A1: Real-time sEMG signal acquisition A2: Muscle activity detection, performance monitoring A3: Wave display + ‘plane game’ gamified feedback A4: Game-based motivation mechanism H1: Platform introduction, education session H2: Assessor-blind FOIS evaluation H3: Weekly video consultations with stroke coach H4: Individualised goal setting, problem-solving A3: Video exercise demonstrations via app A4: Automated exercise reminders A6: Completion recording, non-adherence tracking, peer forum hosting H1: Sensor positioning, signal quality verification H2: Monitoring for aspiration signs, BIOPAC validation H3: Caregiver receives alerts, adjusts feeding H5: Clinical response to warning alerts A1: Automatic swallowing event detection (sEMG, FSR, airflow) A2: Real-time respiratory–swallow coordination monitoring A3: Colour-coded warning display (red/yellow/green) A5: Unsafe pattern detection (piecemeal deglutition, prolonged apnoea) H1: Custom mouthpiece fitting, exercise training H2: MBSS assessment at 0, 8, 12 wk H3: Fortnightly telephone 1RM reassessment H4: Progressive resistance adjustment (60% → 80%) A1: 4-sensor tongue pressure detection A2: Target achievement tracking A3: Real-time visual biofeedback on tablet A6: Exercise completion logging H1: App installation, initial training H2: IOPI tongue pressure measurement at time points H3: Fortnightly face-to-face mediation sessions H4: Technique verification and correction A2: Automated compliance monitoring A3: Mirror function, real-time pitch graph, video demonstrations A6: Colour-coded calendar progress tracking, training data recording H1: Electrode positioning, threshold calibration H2: Water swallow test, FOIS, sEMG assessment H3: Direct supervision of all hospital sessions H4: Game difficulty adjustment based on progress A1: Real-time sEMG signal processing A2: Threshold-based swallow detection A3: ‘Rabbit Climbing Mountain’ game feedback A4: Adjustable threshold for personalised difficulty H1: SLP places sensor on suprasternal notch; initial clinical evaluation H2: Determines eligibility; validates swallow events as gold standard H3: Focus group facilitation for user feedback A1: CNN real-time swallow detection via edge computing (F1 = 0.89, sensitivity 95%) A2: Continuous swallow frequency monitoring A3: Haptic biofeedback via ERM vibration when below threshold A4: Closed-loop operation without external devices H1: Accelerometer placement on thyroid cartilage; threshold calibration H2: Submental ultrasound for hyoid displacement; FOIS assessment H3: Direct supervision of all sessions H4: Acceleration threshold adjustment based on performance A1: Accelerometer detection of laryngeal elevation A2: Real-time acceleration monitoring and performance tracking A3: Visual biofeedback via ‘Frog Catching Mosquito’ 3D animation A4: Adjustable threshold for personalised difficulty H1: SLP performs FEES; prescribes exercise parameters H2: Weekly SLP evaluation of progress H3: ICU nurses supervise daily Rephagia training (~20 min) H4: Exercise settings adjusted by SLP based on progress A1: sEMG signal acquisition from submental muscles; individual calibration A2: Automatic tracking of swallow events A3: Visual biofeedback via ‘Kangaroo’ game — jumps when threshold exceeded H1: Electrode placement; skin preparation; threshold setting H2: VFSS assessment; FOIS, PAS, DOSS, DHI evaluation H3: Direct supervision of all 50 min sessions H4: Progressive threshold adjustment based on performance A1: Real-time sEMG signal processing A2: Threshold-based game control; progress tracking A3: Visual + auditory biofeedback via Rose and Rabbit games A4: Game difficulty progression signals H1: Mobili-T sEMG set-up; guides patients through modules H2: MASA, MBSS, FEES evaluations at multiple time points H3: Delivers programme; provides delayed verbal feedback (3–4 s post-swallow) H4: Adjusts task difficulty; progresses modules at 80% accuracy A1: sEMG acquisition from submental muscles A2: Auto-calibration (110% peak from 5 baseline swallows); timing display A3: Real-time visual biofeedback of effort and timing on iPad A6: Module progression tracking (Learn → Identify → Practice → Challenge → Combo) H1: Study design; sensor kit development and shipping H2: Semi-structured interview; constant comparative analysis H6: Theme identification and interpretation A1: Swallowing and breathing event detection algorithms A3: Visual feedback via iOS app interface; data visualisation A6: Longitudinal tracking functionality (design stage) A4: Participant-suggested: reminder vibrations H1: Sensor set-up; guides three-phase RSC protocol H2: VFSS assessments at baseline, 2 wk, 1 month H3: Direct supervision of all training sessions H4: Monitors patient progress towards optimal EXP-SW-EXP pattern A1: Real-time respiratory and sEMG signal processing A2: Respiratory–swallow coordination pattern detection and analysis A3: Visual biofeedback showing RSC patterns A4: Target setting for optimal coordination pattern H1: Baseline education and exercise instruction; prescribes protocol H2: MBSS, MDADI, FOIS, PSS-HN assessments at follow-up H3: Standard clinical care (periodic contact) A3: Automated exercise guidance via video demonstrations A4: Exercise intensity adjustment based on symptoms A6: Daily challenge tracking; automated reminders; progress data storage H1: sEMG electrode placement (~15 min set-up); system calibration H2: FEES assessments pre/post and 2-month follow-up; FOIS, PAS, P-Score H3: Direct supervision of all 45 min sessions H4: Sets and adjusts sEMG target thresholds; provides verbal coaching A1: sEMG signal acquisition and amplification from submental muscles A2: Objective measurement of exercise performance and intensity A3: Real-time visual biofeedback showing sEMG waveform A6: sEMG amplitude data storage Human role codes: H1, initial set-up and training; H2, clinical assessment; H3, supervision and monitoring; H4, protocol adjustment; H5, clinical decision-making; H6, validation and verification. AI role codes: A1, signal detection and processing; A2, performance tracking and monitoring; A3, real-time feedback delivery; A4, adaptive parameter adjustment; A5, decision support and classification; A6, data management and storage. Collaboration modes: Mode A, AI-augmented direct (clinician present throughout, AI enhances clinical delivery); Mode B, supervised autonomous (AI delivers intervention autonomously, clinician reviews periodically); Mode C, hybrid periodic (patient uses AI-driven system independently, with periodic clinician consultations); Mode D, AI-driven validation (AI generates recommendations or detects events, clinician validates outputs); Mode E, fully autonomous (AI operates independently, clinician involved only for initial set-up and outcome evaluation). 1RM one-repetition maximum, ADAM Automated Detection and Assessment of Movement, AI-VG artificial intelligence video game, BCI brain–computer interface, BiSSkiT Biofeedback in Strength and Skill Training, CNN convolutional neural network, CTAR chin tuck against resistance, DDK diadochokinetic, DHI Dysphagia Handicap Index, DL deep learning, DOSS Dysphagia Outcome and Severity Scale, DRACE Dysphagia Risk Assessment for the Community-dwelling Elderly, DSS decision support system, DSRS Dysphagia Severity Rating Scale, EAT-10 Eating Assessment Tool-10, EAT-STARS Eating and Swallowing Training for Accurate, Reliable and Standardised skill-building, EEG electroencephalography, EMD empirical mode decomposition, EMST expiratory muscle strength training, EPS effortful pitch swallow, EPG effortful pitch glide, ERM eccentric rotating mass, ERT ensemble of regression trees, ETR effortful tongue retraction, EXP-SW-EXP expiration–swallow–expiration, FEES fibreoptic endoscopic evaluation of swallowing, FOIS Functional Oral Intake Scale, FR-VG facial recognition video game, FSR force sensing resistor, FSS2 Flow Short Scale-2, gbCTAR game-based chin tuck against resistance, GOHAI General Oral Health Assessment Index, GUSS Gugging Swallowing Screen, HNC head and neck cancer, HRM heart rate monitor, HRV heart rate variability, ICA independent component analysis, ICU intensive care unit, IMU inertial measurement unit, IOPI Iowa Oral Performance Instrument, IR infrared, LED light-emitting diode, MASA Mann Assessment of Swallowing Ability, MBSS modified barium swallow study, MBSImP Modified Barium Swallow Impairment Profile, mCSES modified Computer Self-Efficacy Scale, MDADI MD Anderson Dysphagia Inventory, mHealth mobile health, ML machine learning, MNA-SF Mini Nutritional Assessment-Short Form, NASA-TLX NASA Task Load Index, OHIP-14 Oral Health Impact Profile-14, OWL Web Ontology Language, PAS Penetration–Aspiration Scale, PSS-HN Performance Status Scale for Head and Neck cancer, RMS root mean square, ROM range of motion, RSC respiratory–swallow coordination, SEHEPS Self-Efficacy for Home Exercise Program Scale, sEMG surface electromyography, SLP speech-language pathologist, SMS short message service, SSA Standardised Swallowing Assessment, SUS System Usability Scale, SWAL-QOL Swallowing Quality of Life, SWRL Semantic Web Rule Language, TAM Technology Acceptance Model, tDCS transcranial direct current stimulation, TMIS tongue–machine interaction system, VDS Videofluoroscopic Dysphagia Scale, VFSS videofluoroscopic swallowing study, wk week(s). The AI:Human ratio represents the estimated proportional distribution of task responsibilities between AI-driven and human-delivered components, derived from the reported intervention protocols. Ratios were assigned based on the relative contribution of automated versus clinician-directed activities during the intervention period and do not reflect set-up or assessment phases. Among human roles, H1 (Initial setup; n = 31, 100%), H2 (Assessment; n = 30, 96.8%), and H3 (Supervision; n = 29, 93.5%) were assigned in the vast majority of studies, whereas H4 (Treatment adjustment) remained under human control in 17 studies (54.8%). Among AI roles, A3 (Feedback delivery; n = 30, 96.8%) and A1 (Detection; n = 26, 83.9%) were the most prevalent, followed by A4 (Adaptation; n = 12, 38.7%) and A5 (Decision support; n = 3, 9.7%). AI roles were concentrated at the signal detection and feedback layers, whereas treatment adjustment and clinical decision-making remained predominantly under human control. Effectiveness of human–AI collaborative interventions Among the 26 studies reporting clinical outcomes, effect directions were predominantly favourable (Fig. 3). However, given the predominance of small samples, short follow-up periods, and feasibility-oriented designs, favourable effect directions should not be equated with robust evidence of clinical effectiveness. The following synthesis distinguishes four outcome tiers that carry different evidentiary weight: (i) clinical swallowing function—direct patient-relevant benefit and the most informative tier; (ii) physiological surrogate measures—which do not invariably translate into functional improvement; (iii) patient-reported and adherence metrics—reflecting acceptability rather than efficacy; and (iv) algorithm performance—reflecting technical capability under controlled validation conditions, typically with healthy volunteers. This hierarchy is maintained throughout the reporting below. Harvest plot of effect directions across outcome domains for AI-enhanced dysphagia interventions. Each bar represents one study, categorised by outcome domain (A: clinical swallowing; B: physiology/strength; C: quality of life [QoL]/patient-reported outcome measures [PROMs]; D: adherence/feasibility) and study design (RCT vs. non-RCT). Bar height encodes sample size (tall: n ≥ 60; medium: 30 ≤ n < 60; short: n < 30), and bar colour encodes the effect direction (green: statistically significant positive effect; yellow: positive trend or mixed results; grey: no difference), and bar pattern fill encodes risk of bias (solid: low risk; hatched: some concerns or moderate risk; open: high risk). Swallowing function outcomes Five studies reported oropharyngeal swallowing function as a primary endpoint. Zhang et al. (two RCTs; n = 84 and n = 26; some concerns)41,42, using an AI-augmented video game (AI-VG) system, reported significant improvements over controls in Gugging Swallowing Screen (GUSS) scores (P < 0.05), the larger trial additionally demonstrated significant improvement in Standardised Swallowing Assessment (SSA) scores (P = 0.006). Alyanak et al. (RCT; n = 33; some concerns) reported significantly greater improvement in Dysphagia Outcome and Severity Scale (DOSS) scores (P = 0.004)54. Park et al. (RCT; n = 37; low risk of bias) observed within-group improvements in Videofluoroscopic Dysphagia Scale (VDS) scores in both arms (P < 0.05)68, although the difference between groups was not statistically significant. In a three-arm comparison (conventional therapy plus transcranial direct current stimulation [tDCS] as control, plus sEMG biofeedback, and plus gamified training), Hou et al. (RCT; n = 90; some concerns) reported a stepwise increase in clinical response rates (60, 76.7, and 90%)50. The Functional Oral Intake Scale (FOIS) was the most widely reported outcome measure (n = 12)31,41,42,45,46,50–52,54,58,67,68. Seven studies31,41,42,45,50,58,67—six RCTs (one at low risk of bias, five with some concerns) and one case-control study—demonstrated short-term superiority over controls; the remaining five reported within-group improvements only, which, in the absence of between-group differences, provide limited evidence of intervention-specific effects. Three studies incorporating longer-term follow-up (6 weeks to 3 months) consistently showed that between-group differences were no longer significant (P > 0.05)51,58,67. Li et al. additionally reported a nasogastric tube removal rate of 80%31. Three studies targeting age-related swallowing decline and Parkinson’s disease assessed oral motor function. Chan et al. (quasi-experimental; n = 70 completers; moderate risk) reported significant objective improvements in bite force (P < 0.001), masticatory efficiency (P = 0.002), and oral diadochokinetic rate (P < 0.05), although the subjective screening tool (Eating Assessment Tool-10 [EAT-10]) did not reach significance57. Jung et al. (RCT; n = 76; some concerns) reported increased salivary flow (+0.71 g min-1; P < 0.05), whilst the subjective oral health assessment (General Oral Health Assessment Index [GOHAI]) was non-significant62. Battel et al. (within-subject feasibility; n = 10; moderate risk) observed significant improvements in the severity and frequency of salivary residue in patients with Parkinson’s disease (P < 0.05)52. Grading of Recommendations, Assessment, Development and Evaluation (GRADE) certainty: Moderate (downgraded for imprecision due to small sample sizes). Physiological and biomechanical outcomes Four studies evaluated lingual muscle activation. Chan et al. reported significant improvements in tongue pressure (P < 0.001) and lingual endurance (P = 0.004)57. Jung et al. demonstrated increased anterior tongue pressure (+4.48 kPa; P < 0.05); posterior tongue pressure and buccal pressure did not reach significance62. Kim et al. (single-arm; n = 8 completers) showed an increase in swallowing tongue pressure from 17.5 to 26.5 kPa (P = 0.046); the effect was not maintained at 12-week follow-up43. Krekeler et al. (RCT; n = 19; some concerns) found no statistically significant difference in maximum isometric pressure, effect sizes were large (anterior tongue pressure: Cohen’s d = 0.95; posterior tongue pressure: d = 0.96)67. With respect to submental muscle activation, Hou et al. (RCT; n = 90) revealed a progressive increase in sEMG peak amplitude across the control (22 μV), sEMG biofeedback (30 μV), and gamified (44 μV) groups (P < 0.05), with a corresponding reduction in swallowing duration (1.52 s, 1.32 s, and 1.09 s; P < 0.05)50. Kim et al. (device validation; 30 healthy volunteers and 1 patient with Parkinson’s disease) quantified the effects of specific swallowing manoeuvres on submental muscle activity48. Jansen et al. (single-arm feasibility; n = 20 ICU patients) observed sEMG changes that did not reach statistical significance (52 to 57 μV); physiological changes showed no significant correlation with clinical outcomes (FOIS, PAS)49. Concerning hyoid–laryngeal biomechanics, Li et al. (case–control; n = 20), employing accelerometer-guided game-based biofeedback, showed significantly greater hyoid displacement in the intervention group (intervention group: 11.37 to 14.45 mm, P = 0.002; control group: 12.84 to 13.35 mm, non-significant)31. Seven studies assessed swallowing safety through aspiration and pharyngeal residue outcomes. Three reported significantly lower aspiration rates in the intervention group compared with the control group (P < 0.05)45,46,51. Alyanak et al. reported within-group improvement in liquid Penetration–Aspiration Scale (PAS) scores (P = 0.026) and superiority over controls for semi-solid consistencies (P = 0.031)54. Park et al. reported within-group PAS improvements in both arms (P < 0.05) without a significant between-group difference68. Battel et al. demonstrated sustained improvement in salivary and solid residue in patients with Parkinson’s disease at three-month follow-up (P < 0.05)52. Krekeler et al. reported a substantial reduction in residue (d = 1.2), PAS improvement remained non-significant (P > 0.05)67. GRADE certainty: Low (downgraded for risk of bias owing to predominant use of non-randomised designs, inconsistency due to heterogeneous measurement methods, and imprecision due to small sample sizes). Quality of life and psychological outcomes Three studies reported significant improvements in swallowing-related quality of life (Swallowing Quality of Life [SWAL-QOL] or Dysphagia Handicap Index [DHI] scores; P < 0.05)41,54,58. Six studies reported favourable trends that fell short of statistical significance (P > 0.05)43,51,52,57,62,67. Starmer et al. found that treatment adherence was associated with better MD Anderson Dysphagia Inventory (MDADI) scores in patients with HNC, although physiology-related quality of life did not differ significantly between groups59. Psychological outcomes showed favourable trends across several studies43,56,68, including enhanced participant motivation, engagement, and self-efficacy. Su et al. reported that flow state experience scores were highest at moderate game difficulty56. Benfield et al. noted lower mood in the biofeedback group when patients were repeatedly confronted with failure-indicating feedback51. GRADE certainty: Moderate, downgraded for inconsistency (instrument heterogeneity: SWAL-QOL, DHI, MDADI). Patient engagement and adherence Short-term adherence was generally high (72.7–100%, median > 80%)41,43,46,51,52,57; five studies reported 100% adherence31,49,50,56,60, though three of the five49,50,56 were conducted under direct supervision; durations ranged from a single session to approximately eight weeks. Srp et al. reported a decline from 100% to 50% between intensive and maintenance phases60; Chan et al. and Krekeler et al. reported attrition rates of 38% and 41%57,67, respectively, in studies with extended durations. GRADE certainty: Moderate, downgraded for inconsistency (variable metrics: session completion rates, repetition counts, login frequency). Safety profile and temporal sustainability Of the 31 included studies, only 15 (48.4%) provided any form of adverse event reporting; the remaining 16 did not, representing a substantial gap in safety evidence. Among reporting studies, no treatment-related serious adverse events were recorded, and minor issues were limited to mild skin irritation, transient cervical discomfort, xerostomia, and fatigue. However, the absence of serious events should be interpreted in light of incomplete reporting: most studies relied on spontaneous reporting rather than systematic active monitoring, and no study reported near-miss aspiration events or instrumentally confirmed subclinical aspiration. GRADE certainty: Moderate (incomplete reporting). Beyond safety, the temporal sustainability of treatment effects warrants explicit consideration. Nine studies incorporated post-intervention follow-up. Among these, effect attenuation was a consistent finding: between-group FOIS differences were no longer significant at 6–12 weeks51,58,67; tongue pressure gains at 8 weeks were not maintained at 12 weeks43; and adherence declined substantially during maintenance phases57,60,67. Exploratory analysis of potential effect modifiers Stratified by human–AI collaboration mode, Modes A and B (continuous or frequent clinician involvement; n = 16; including 8 RCTs, 50.0%): nine studies reported significant improvements in swallowing function or functional oral intake31,41,42,45,46,50–52,68, two reported physiological measure improvements31,50, and adherence rates were consistently high (83.3–100%). Mode C (hybrid periodic consultation; n = 7; including 4 RCTs, 57.1%): six studies reported functional or physiological improvements57–60,62,67, with more variable adherence rates (37–100%) and effect attenuation at longer-term follow-up in three studies58,60,67. Modes D and E (high AI autonomy; n = 8; including 1 RCT, 12.5%) were predominantly focused on technology validation, with three studies reporting clinical functional outcomes30,56. Across dysphagia aetiologies, post-stroke dysphagia (n = 13; 10 RCTs): 10 studies reported FOIS improvements31,41,42,45,46,50,51,58,67,68, five reported aspiration-related improvements45,46,51,54,68, and three reported physiological improvements31,50,67. Among Parkinson’s disease studies (n = 5; 0 RCT): three reported improvements in swallowing function or frequency30,52,60 and two were device validation studies48,63. Age-related swallowing decline (n = 4; 1 RCT), three reported tongue pressure improvements43,57,62, although sample sizes were uniformly very small. Across technology types, sEMG biofeedback systems (n = 11): five studies reported functional improvements45,46,50,52,54 and six reported enhanced physiological activity46,48–50,52,54. Facial recognition systems (n = 5), three reported functional improvements41,42,57 and user acceptance was consistently favourable. mHealth application studies (n = 10), five reported short-term adherence rates exceeding 70%43,44,56,57,60, whilst three demonstrated declining adherence during the maintenance phase43,57,60. With respect to study design, 12 of the 13 RCTs reported significant within-group improvements41,42,45,46,50,51,54,56,58,62,67,68 and eight reported significant between-group differences41,42,45,46,50,54,58,67. The 18 non-randomised, validation, and qualitative studies—rated moderate-to-high risk of bias for interventional and validation designs (n = 16) and low risk for qualitative designs (n = 2)—predominantly reported feasibility, usability, and physiological outcomes but provided limited clinical effectiveness evidence. AI system performance, usability and clinician experience Eight studies evaluated AI algorithm performance. The ADAM wearable sensor (validated in 58 healthy adults and 20 patients with Parkinson’s disease during single laboratory sessions) achieved a swallowing detection sensitivity of 95% and specificity of 99% (F1 = 0.89), and correlated strongly with speech-language therapist visual observation (r = 0.92)30. Lee et al. achieved 100% accuracy for liquid swallowing detection47, and Kim et al. (30 healthy adults and 1 patient with Parkinson’s disease) reported signal validation results exceeding 0.9548. Aslan et al. employed ensemble machine learning for EEG-based motor imagery classification, achieving 99.8% accuracy—substantially outperforming single algorithms (k-nearest neighbours: 79.4%; CNN combined with continuous wavelet transform: 83%); this analysis was conducted offline rather than in real time64. The Swallowscope achieved a thickened liquid swallowing detection accuracy of 75.9%61. The tongue–machine interface system demonstrated a click accuracy of 70% and an information transfer rate of 130 bits min−171. The ontology-based clinical decision support system received a clinician approval rate exceeding 66.7%65. GRADE certainty: Very low (risk of bias—predominantly healthy-volunteer validation; inconsistency—performance ranging from 70% to 99.8%; indirectness—controlled laboratory conditions; imprecision—including single-patient pilot testing). Fourteen studies30,41–43,47–49,51,55–57,60,63,65 analysed system usability, acceptance, and satisfaction. User-reported benefits included objective data supporting clinical diagnosis and decision-making; enhanced engagement and autonomy through gamified human–AI interaction and wearable device feedback; user-friendly interface design; and soft, stretchable wearable materials. Healthcare professionals’ interaction with the systems was notably shaped by perceived system confidence—when professionals disagreed with AI judgements, they tended to selectively override swallowing risk predictions. The impact of these technologies on clinical workflows was examined in four studies. Positive effects included improved team dynamics, enhanced patient communication, and strengthened objective evidence for clinical decision-making. Identified concerns included incorrect execution of intervention recommendations, over-reliance on AI, and potential interference with clinical judgement. No study examined effects on patient empowerment. Implementation barriers: NASSS framework analysis Systematic mapping of implementation barriers to the NASSS framework domains revealed domain-specific complexity patterns (Table 3). Table 3Implementation complexity assessment based on the NASSS frameworkStudyD1ConditionD2TechnologyD3ValueD4AdoptersD5OrganisationD6WiderSystemD7EmbeddingOverallComplexityPrimary Implementation BarrierZhang (2025)●●○●○○●●○●●○●○○○○○●●○ModerateTechnology acceptance; home implementation feasibilitySpoladore (2024)●●○●●●●●○●●●●●○●●○●●●HighOntology modelling complexity; black-box avoidanceZhang (2025)●●○●○○●●○●●○●○○○○○●○○ModerateSmall pilot sample; equipment requirementsChan (2024)●○○●●○●●○●●●●○○○○○●●○ModerateDigital literacy; sustained engagement (38% dropout)Su (2025)●○○●○○●●○●○○●●○○○○●○○ModerateSmartphone ownership; learning curveJung (2024)●○○●●○●●○●●○●●○○○○●○○ModerateSmartphone access; digital literacy in elderly populationBenfield (2023)●●●●●○●●○●●○●●○○○○●●○ModerateEquipment set-up; time-intensive sessions (45 min)Park (2019)●●○●●○●●○●●●●●○○○○●●○ModerateEquipment requirements; 19.6% dropoutKhan (2024)●○○●●●●○○●●●●○○○○○●●●HighComplex calibration; silicone device acceptanceLee (2017)●○○●●●●●○●●●●○○○○○●●●HighWireless integration needed; cable fragility; n = 4Zhang (2024)●●○●●○●○○●●●●●○○○○●●○ModerateElderly digital literacy; supervision neededKuramoto (2016)●●○●●○●●○●●○●○○○○○●●○Moderate75.9% accuracy insufficient for clinical use; validation onlyKim (2020)●○○●○○●●○●●●●○○○○○●●○ModerateDigital literacy critical; preventive population onlyAslan (2024)●○○●●●●○○●●●●○○○○○●●●HighSpecialised EEG equipment; no patient validationSrp (2024)●●○●○○●●○●●○●○○○○○●●○ModerateComparative trial needed; therapist training requirementsKim (2019)●●○●●●●●○●●●●○○○○○●●●HighCost concerns; single patient validationBattel (2023)●●●●●○●●○●●●●●○○○○●●○ModerateIPD progression complexity; trained clinician neededWang (2023)●●○●○○●●○●●○●●○○○○●○○ModerateInternet literacy; coach time requirementShieh (2023)●○○●●●●●○●●●●○○○○○●●●HighSensor placement complexity; thresholds require validationKrekeler (2023)●●●●●○●●○●●○●●○○○○●●○Moderate41% attrition; fortnightly telephone contact neededKim (2020)●○○●○○●●○●●●●○○○○○●○○ModerateTechnology unfamiliarity; 3/11 non-compliantHou (2024)●●●●●○●●●●●●●●○○○○●○○ModerateSpecialised VitalStim equipment; trained therapist requiredXu (2025)●●○●●●●●●●●●●○○●●○●●○HighPatch application difficulty; skin-tone preferenceLi (2016)●●●●●○●●●●●●●●○○○○●●○HighThreshold adjustment complexity; therapist supervision requiredJansen (2025)●●●●●○●○○●●○●●○○○○●●○ModerateICU constraints; sEMG signal quality issuesAlyanak (2025)●●○●●○●●○●●●●●○○○○●●○ModerateSpecialised equipment; therapist training requiredBahia (2025)●●○●●●●●○●●●●●○○○○●●●HighCognitive demands; programme refinement needed; n = 2Kantarcigil (2025)●●○●●●●●○●●●●○○○○○●●●HighSensory distractions; affordability concernsWang (2026)●●○●●○●●●●●○●●○○○○●●○ModerateBimodal equipment requirements; coordination training complexityStarmer (2023)●○○●○○●●○●●○●●●○○○●●○ModerateMulti-centre coordination; app-only support limitationsNordio (2022)●●○●●○●●○●●●●●○○○○●●○ModerateHigh equipment cost; 15 min set-up time per sessionNASSS Non-adoption, Abandonment, Scale-up, Spread, and Sustainability framework, D1 domain 1 (condition), D2 domain 2 (technology), D3 domain 3 (value proposition), D4 domain 4 (adopters), D5 domain 5 (organisation), D6 domain 6 (wider system), D7 domain 7 (embedding and adaptation over time). ADAM Automated Detection and Assessment of Movement, AI-VG artificial intelligence video game, BCI brain–computer interface, BiSSkiT Biofeedback in Strength and Skill Training, DSS decision support system, EAT-STARS Eating and Swallowing Training for Accurate, Reliable and Standardised skill-building, EEG electroencephalography, FR-VG facial recognition video game, gbCTAR game-based chin tuck against resistance, HNC head and neck cancer, ICU intensive care unit, IPD idiopathic Parkinson's disease, RSC respiratory–swallow coordination, sEMG surface electromyography, TMIS tongue–machine interaction system.Complexity ratings were assigned according to the NASSS framework: ●●●, high (multiple interconnected challenges requiring substantial resources or systemic change); ●●○, moderate (manageable with appropriate resources and planning); ●○○, low (minimal barriers to implementation). Overall complexity was determined by the predominant pattern across all seven domains. Ratings were derived from explicit barrier reports, implicit challenges described in study methods and discussion sections and the degree of infrastructure, expertise and end-user adaptation required for deployment. Colour coding for overall complexity: red shading, high; yellow shading, moderate. Implementation complexity assessment based on the NASSS framework NASSS Non-adoption, Abandonment, Scale-up, Spread, and Sustainability framework, D1 domain 1 (condition), D2 domain 2 (technology), D3 domain 3 (value proposition), D4 domain 4 (adopters), D5 domain 5 (organisation), D6 domain 6 (wider system), D7 domain 7 (embedding and adaptation over time). ADAM Automated Detection and Assessment of Movement, AI-VG artificial intelligence video game, BCI brain–computer interface, BiSSkiT Biofeedback in Strength and Skill Training, DSS decision support system, EAT-STARS Eating and Swallowing Training for Accurate, Reliable and Standardised skill-building, EEG electroencephalography, FR-VG facial recognition video game, gbCTAR game-based chin tuck against resistance, HNC head and neck cancer, ICU intensive care unit, IPD idiopathic Parkinson's disease, RSC respiratory–swallow coordination, sEMG surface electromyography, TMIS tongue–machine interaction system. Complexity ratings were assigned according to the NASSS framework: ●●●, high (multiple interconnected challenges requiring substantial resources or systemic change); ●●○, moderate (manageable with appropriate resources and planning); ●○○, low (minimal barriers to implementation). Overall complexity was determined by the predominant pattern across all seven domains. Ratings were derived from explicit barrier reports, implicit challenges described in study methods and discussion sections and the degree of infrastructure, expertise and end-user adaptation required for deployment. Colour coding for overall complexity: red shading, high; yellow shading, moderate. Condition complexity (Domain 1) was rated moderate-to-high in 21 studies (67.7%). Principal challenges included comorbidity interactions; trajectory unpredictability; and severity heterogeneity. ICU-acquired dysphagia exhibited the highest condition complexity. Technology complexity (Domain 2) showed a three-tier distribution: nine studies (29.0%) documented high, 14 (45.2%) moderate, and eight (25.8%) low technology complexity. Technological barriers encompassed calibration requirements (individualised threshold setting for sEMG systems; personalised difficulty progression in gamified interventions); hardware dependencies (availability of dedicated equipment; sEMG device material flexibility and aesthetics; sensor placement precision; cable fragility in early prototypes); algorithmic limitations (non-interpretability of deep learning systems); and interoperability gaps (absence of electronic health record integration; platform-specific restrictions). The value proposition (Domain 3) remained largely unquantified. Only two studies reported device costs: the TMIS (