Bilateral cerebellar intermittent theta-burst stimulation combined with laryngeal elevation training promotes neurorestoration in post-stroke dysphagia Authors: Haiquan Zhang, Jingyun Tan, Di Zhang, Wei Lu, Min Liu, Shanshan Wang Published: 2026 ABSTRACT AbstractPrevious studies have shown that pharyngeal elevation training can improve swallowing function in stroke patients, but the central nervous system interventions have limited effects. Intermittent theta burst stimulation (iTBS) applied to bilateral cerebellar hemispheres may enhance cortical excitability. However, the efficacy and underlying mechanisms of combining bilateral cerebellar iTBS with real-time biofeedback-guided pharyngeal elevation training for post-stroke dysphagia remain unclear. This study aims to investigate the combined effect of bilateral cerebellar iTBS and biofeedback-guided pharyngeal elevation training on swallowing function and cortical activity in stroke patients during the recovery phase. Methods: 45 stroke patients were randomized into the Control, Sham-iTBS + SET, or Active-iTBS + SET groups (n = 15 each). The latter two groups received additional interventions: Sham or active bilateral cerebellar iTBS combined with swallowing exercises. 42 participants (n = 14 per group) completed the study and were included in the final analysis, following 3 dropouts (n = 1 per group). Swallowing function was assessed using the Standardized Swallowing Assessment (SSA) and Penetration-Aspiration Scale (PAS) before and after 4 weeks of treatment. Functional near-infrared spectroscopy (fNIRS) recorded cortical activation during tasks and resting-state functional connectivity. After 4 weeks of intervention, all groups demonstrated significant improvements in the primary outcome (SSA) and the secondary behavioral outcome (PAS) scores (p < 0.05). Compared to the Control Group, both Sham-iTBS + SET and Active-iTBS + SET showed superior clinical improvements (p < 0.05); notably, the Active-iTBS + SET group significantly outperformed the Sham-iTBS + SET Group (p < 0.05) in both SSA and PAS recovery. Secondary fNIRS analysis revealed that the Active-iTBS + SET group significantly increased activation of the right prefrontal cortex (RPFC) and premotor areas (PM). This was accompanied by enhancing interhemispheric and intra-network functional connectivity, particularly between the PFC and other regions, and between M1 and PM (p < 0.05). Furthermore, local node connectivity was increased (p < 0.05). Correlation analysis indicated that the clinical improvements in SSA scores were positively correlated with neural activation observed in RPM and RPFC (p < 0.05). Bilateral cerebellar iTBS combined with pharyngeal elevation training significantly improves swallowing function in post-stroke dysphagia patients, may be associated with increased activation of the right prefrontal cortex and premotor areas, and enhanced cortico-cortical connectivity. The RPM region may serve as a potential neural target for cerebellar iTBS in stroke rehabilitation.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-56970-z.Keywords: Stroke, Cortical function, Repetitive transcranial magnetic stimulation, Laryngeal ascending reinforcement training, Functional near-infrared spectroscopySubject terms: Neurology, Neuroscience Introduction Post-stroke dysphagia poses a substantial challenge in neurorehabilitation, with epidemiological studies indicating a prevalence rate of up to 30 to 67%27. This impairment not only significantly increases the risk of aspiration pneumonia by more than threefold14, but it also often hinders the swift recovery of swallowing function, leading to secondary complications such as malnutrition and dehydration11,16,17,32. This, in turn, prolongs recovery time and raises the 30-day readmission rates20. While conventional rehabilitation methods may yield short-term improvements in swallowing symptoms, they often fall short of restoring the neuro-muscular coordination essential for swallowing initiation, execution, and cortical regulation, including laryngeal movement. Consequently, over 50% of patients continue to experience delayed or insufficient laryngeal lift even in later stages of recovery22. This highlights the limitations of current interventions in facilitating neuroplasticity within the corticobulbar-cerebellar pathways25. The core issue lies in the challenges that peripheral training encounters in reconstructing the dynamic coordination required for swallowing initiation, execution, and cortical control8. In recent years, non-invasive brain stimulation techniques have opened new pathways for reorganizing the neural circuits involved in swallowing. Numerous studies have demonstrated that repetitive transcranial magnetic stimulation (rTMS) can improve swallowing function by enhancing cortical excitability1,4,19. However, current interventions primarily focus on the primary motor cortex, with limited research on the cerebellum, which plays a crucial role in swallowing30,33. Notably, studies published in JAMA have highlighted the involvement of the cerebellum, particularly through the dentate nucleus-thalamus-cortex pathway, in the temporal coordination of motor activities. Its neuroplasticity may provide new therapeutic targets for functional recovery15. Furthermore, intermittent theta burst stimulation (iTBS), a novel rTMS protocol, can induce long-term potentiation-like effects that enhance synaptic plasticity. Research suggests that bilateral cerebellar iTBS is more effective than unilateral stimulation in improving cortical function29. Pharyngeal elevation training offers distinct benefits as a peripheral intervention. This training employs techniques such as resistance swallowing, glottic elevation swallowing, and biofeedback techniques to enhance the timely contraction of the thyrohyoid and mentalist muscles. The mechanisms underlying this approach have evolved from anatomical studies to functional imaging evidence23. Our previous research confirms that pharyngeal elevation training, when combined with conventional rehabilitation, effectively improves swallowing function in post-stroke dysphagia patients24. However, isolated peripheral training faces challenges in modulating the corticobulbar-cerebellar pathways10,13. It remains uncertain whether a “closed-loop” approach, integrating central nervous system stimulation with peripheral training, can enhance therapeutic outcomes. Addressing these limitations, this study employs the “central-peripheral-central” (CPC) regulation theory12. We propose a novel combined intervention that targets bilateral cerebellar iTBS alongside pharyngeal elevation training, with the addition of functional near-infrared spectroscopy (fNIRS) monitoring of cortical activity. This study aims to explore whether this synergistic approach can improve swallowing function in stroke patients during recovery, thereby overcoming the limitations of individual treatments. Additionally, we aim to assess cortical functional changes under this combined protocol to provide a theoretical basis for precision rehabilitation of dysphagia. Materials and methods Participants This study is a randomized, single-blind, controlled clinical trial. This study strictly followed the Declaration of Helsinkiand ICH-GCP guidelines. Approved by the Ethics Committee of Shandong Provincial Third Hospital (KYLL-2023025), all participants/guardians provided written informed consent. Identifiable information was permanently removed. Patient recruitment initiated in January 2023 was originally performed as a pilot exploratory study without prospective trial registration. After favorable efficacy outcomes were observed from the pilot data, we completed formal trial retrospective registration at the Chinese Clinical Trial Registry (ChiCTR2300071393) on May 15, 2023, and the full trial was reported following the CONSORT 2025 statement. Inclusion criteria: (1) Diagnosed with stroke according to the criteria set by the International Stroke Conference9, confirmed by cranial CT or MRI,(2) First-onset stroke, with duration > 1 week and ≤ 6 months; (3) Age between 20 and 70 years, right-handed; (4) Mini-Mental State Examination (MMSE) score ≥ 21 points34, capable of cooperating with treatment, and without severe cognitive impairment,(5) Diagnosed with dysphagia confirmed by videofluoroscopic swallowing study (VFSS), with NIH Stroke Scale (NIHSS) scores > 5 and < 15; (6) The participant or their family signed informed consent and agreed to participate in the clinical intervention. Exclusion criteria: (1) Hemodynamic instability; (2) Severe pain or intolerance to treatment; (3) Extensive scalp skin lesions or skull defects; (4) Presence of intracorporeal metal implants such as pacemakers or metallic skull implants; (5) Structural diseases of the oral cavity or pharynx; (6) Patients with brainstem infarction or hemorrhage. Sample size Sample size estimation was conducted using G*Power (Version 3.1; https://www.psychologie.hhu.de/arbeitsgruppen/allgemeine-psychologie-und-arbeitspsychologie/gpower). Based on a prior review of swallowing therapy outcomes which reported a consistent medium effect size7, a medium effect size (f = 0.25) was assumed. With α = 0.05, a statistical power (1-β) = 0.8, and a correlation of 0.5 among repeated measures, the required minimum total sample size was calculated to be 42 participants (n = 14 per group). To ensure sufficient power despite potential attrition, we initially randomized 45 participants (n = 15 per group), allowing for a dropout rate of up to 10%. Ultimately, 42 participants completed the protocol and were included in the final analysis, which maintains the pre-specified statistical power. Randomization A total of 50 stroke patients with dysphagia during the recovery phase, who were hospitalized in the Department of Rehabilitation Medicine at Shandong Provincial Third Hospital between January 2023 and November 2024, were assessed for eligibility. Among them, 45 patients met the inclusion criteria. These participants were randomly assigned to one of three groups: a Control Group, a Sham-iTBS + SET Group, and an Active-iTBS + SET Group, according to a computer-generated randomization sequence. An independent supervisor, who was not involved in the trial, generated the random sequence and performed the group assignments. This ensured that the research staff remained unaware of and could not influence the allocation process. Blinding of participants was achieved as follows: in the control condition, the coil was rotated 90°, and the intensity was set to 10% of the resting motor threshold (RMT), producing an equivalent somatosensory experience. Outcome assessors were also blinded: a neurorehabilitation physician who was unaware of group assignments and not involved in the intervention procedures performed all assessments, with no access to allocation information. All procedures, including the generation of the random sequence, the allocation process, and the supervisor’s identity, were thoroughly documented. A flowchart illustrating patient screening, randomization, assessment, and intervention is provided in Fig. 1. The CONSORT flow diagram of patient screening, randomization, assessment, and intervention. Interventions The Control Group received conventional swallowing training only. The Sham-iTBS + SET Group received sham cerebellar bilateral iTBS stimulation and strength-endurance swallowing exercise training (SET), in addition to conventional rehabilitation. The Active-iTBS + SET Group received active cerebellar bilateral iTBS combined with SET, along with the same conventional rehabilitation program. All participants in the three groups received SET as a baseline intervention. This design was chosen to adhere to clinical ethical standards, ensuring that no patient was deprived of standard-of-care physical training during the recovery phase, while allowing for the evaluation of the synergistic effects of iTBS as an adjunctive therapy. The intervention protocols are illustrated in Fig. 2a. Experimental procedure and schematic of fNIRS channel locations on the head (All individuals depicted are volunteers)0.2.4.1 Conventional Swallowing Training. The training included lingual muscle exercises, sensory stimulation, and swallowing reflex facilitation. Each session lasted 30 min, administered once per day, five days per week, for a total of four weeks. Strength-endurance swallowing exercise training The training was conducted using a biofeedback-based strength-endurance swallowing exercise device (Model: LES100, South Korea). Participants were seated in a standard position and held a sensor placed under the tongue. Baseline muscle strength and endurance thresholds were automatically recorded based on three standard swallowing actions, with real-time swallowing dynamics curves generated by the system. The formal training consisted of two modules: (1) Isotonic Module: The intensity was set at 80% of the peak baseline value. The muscle strength threshold could be dynamically adjusted by ± 2% based on the participant’s performance. (2) Isometric Module: The intensity was set at 50% of the peak baseline value. Participants were required to maintain hyolaryngeal elevation for 60 s, with the endurance threshold fine-tuned dynamically by ± 2% according to performance. Both modules provided real-time feedback on task completion via an interactive interface. Each session lasted 30 min, conducted once daily, five days a week, for four weeks. The procedure is illustrated in Fig. 2b. Cerebellar bilateral iTBS protocol All participants remained seated and relaxed during the transcranial magnetic stimulation session, which was delivered using a MagNeuro T140 device (3 T, Weisiman Medical, Nanjing, China). A figure-of-eight coil was placed tangentially over the skull surface of the dominant hemisphere’s M1 area. The coil was finely adjusted by hand to identify the cortical site eliciting the maximal motor response and shortest latency in the first dorsal interosseous muscle, and the resting motor threshold (RMT) was determined accordingly. The iTBS protocol was applied at 100% RMT29, with an intra-burst frequency of 50 Hz and an inter-burst frequency of 5 Hz. Each burst consisted of 3 consecutive pulses, delivered over 2 s, followed by an 8-s interval. This cycle was repeated 40 times, resulting in a total of 1200 pulses per session. A conical coil was positioned over the cerebellar pharyngeal representation area (2.5 cm below the inion and 4.3 cm laterally)31. Each cerebellar hemisphere received 600 pulses, beginning with the left side, followed by the right. The sham group received placebo iTBS using the same coil positioning but with an intensity of 1% RMT, 100% auditory feedback, and the coil rotated 90°. iTBS was administered once daily, five days per week, for four weeks. See Fig. 2b for details. Assessment methods Swallowing function was assessed using the Standardized Swallowing Assessment (SSA) as the primary outcome measure and the Penetration-Aspiration Scale (PAS) as the secondary outcome measure. In addition, hemodynamic responses were recorded using functional near-infrared spectroscopy (fNIRS). All outcome measures were evaluated at baseline (pre-intervention) and after 4 weeks of treatment (post-intervention). Assessments for each participant were completed within a single session lasting approximately 90 min. All evaluations were conducted by an experienced swallowing therapist who was blinded to group allocation and assessment time points. The SSA is a validated clinical tool for evaluating swallowing function and consists of three component (1) Clinical examination (covering 8 indicators including pharyngeal reflex and cough, with a total score ranging from 8 to 23 points); (2) Water drinking test (continuously swallowing 5 ml of water three times, observing manifestations such as water leakage at the corners of the mouth and laryngeal movement, with a total score ranging from 5 to 11 points); (3) Eating assessment (those who meet the criteria in the first two components proceed to a 60 ml water swallowing assessment, observing drinking status and aspiration, with a total score ranging from 5 to 12 points). The total score of the scale ranges from 18 to 46 points, with lower scores indicating milder swallowing dysfunctio27. Videofluoroscopic Swallowing Study (VFSS) Protocol: VFSS was performed with participants seated upright in a wheelchair. The fluoroscopic field was adjusted to ensure coverage from the oral cavity superiorly to the level of the sixth cervical vertebra (C6) inferiorly. Continuous fluoroscopic imaging was used to dynamically assess the oral, pharyngeal, and laryngeal phases of swallowing, as well as airway protection mechanisms, including the ability to voluntarily cough in response to aspiration. During the assessment, patients were administered boluses of 3 mL, 5 mL, and 10 mL consisting of contrast-enhanced materials with different consistencies, including thin liquid, mildly thick liquid, and thick liquid. Each condition was tested in three consecutive trials. All examinations were conducted under real-time monitoring. If aspiration was observed, the procedure was immediately discontinued, and appropriate airway clearance measures, including suctioning, were performed when necessary.Following the lateral view assessment, participants were repositioned for an anteroposterior (AP) view to further evaluate pharyngeal symmetry, structural movement, and post-swallow residue35. PAS Assessment:PAS scores were obtained using a videofluoroscopic swallowing study (VFSS). For analysis, the worst (highest) PAS score observed across all trials and conditions was recorded. In cases where multiple swallows occurred during a single bolus, the highest (most severe) score was used. Scoring was performed after completion of the swallow and reopening of the airway, in accordance with standard PAS evaluation procedures. PAS scores were interpreted as follows: Score 1: normal (no airway invasion); Scores 2–5: laryngeal penetration (material enters the airway but remains above the vocal folds); Scores 6–8: tracheal aspiration (material passes below the vocal folds). All VFSS recordings were independently evaluated by two experienced doctors blinded to group allocation, and discrepancies were resolved by consensus2. NIRS Assessment: This study utilized near-infrared brain functional imaging (Model: Huichuang, Danyang, China, NirSmart-3000B), employing two near-infrared wavelengths of 730 nm and 850 nm to measure changes in oxyhemoglobin (HbO2) concentration, with a sampling frequency of 11 Hz. A total of 14 light source emission points and 14 signal reception points were used, with a 30 mm separation between the probe and the light sources, yielding 35 channels. These channels overlie the subject’s cerebral cortex, covering bilateral prefrontal lobes, bilateral motor areas, bilateral primary sensory cortices, and bilateral supplementary motor areas. For further data analysis, the channels were merged into regions of interest (ROI) according to relevant recommendations18: LPFC: left prefrontal cortex,RPFC: right prefrontal cortex; LPM: left supplementary motor cortex; RPM: right supplementary motor cortex; LS1: left primary somatosensory cortex; RS1: right primary somatosensory cortex. The assessment consists of resting state and task state. The patient first completes a 5-min eyes-closed resting state test, avoiding head movements during this period. The task state adopts a continuous swallowing paradigm: the subject sits in front of a computer and performs operations according to on-screen instructions, when “Please start swallowing” appears, the subject continuously swallows 3 ml of warm water with the therapist’s assistance for 30 s; when “Please rest” appears, swallowing stops, and the subject closes their eyes to rest for 30 s. The experiment consists of 5 blocks of “swallowing-rest” cycles, with 10 s of resting data collected as a baseline before the formal task begins. The total duration of the task state is 310 s. The first 30 s of each cycle constitute the warm water swallowing task period. The summary of ROI and task procedure is shown in Fig. 2d. NIRS data processing the fNIRS data were preprocessed using the HOMER2 toolbox, a graphical user interface program based on MATLAB 2013b (MathWorks, Inc.; Natick, USA). Initially, motion artifacts were eliminated using a method based on the standard deviation of motion and spline interpolation. Motion artifacts were identified by detecting sliding window standard deviations exceeding a specific threshold. Any signal variations in the data channels surpassing the threshold (standard deviation STD and amplitude AMP) were marked as motion artifacts. Spline interpolation was applied to remove these artifacts. Subsequently, a second-order Butterworth bandpass filter with cutoff frequencies of 0.01–0.1 Hz was employed to filter out physiological noise caused by heartbeat, respiration, and high-frequency noise. Finally, the filtered optical data were converted into HbO2 and deoxyhemoglobin (HbR) concentrations using the modified Beer-Lambert law. This study focused on the HbO2 signal as it better reflects cortical activity. Brain network construction the characteristics of resting-state functional networks were investigated using graph theory. 35 fNIRS channels were defined as network nodes, while the edges were represented by the Pearson correlation coefficients between channel pairs. To enhance normality, the resulting correlation matrices underwent Fisher R-to-Z transformation. Subsequently, the z-score matrices were converted to absolute values and ranked from maximum to minimum. A sparsity threshold ranging from 0.1 to 0.8 (in increments of 0.01) was applied to the network. Connections with values exceeding the threshold were assigned a value of 1, and those below were assigned 0, effectively eliminating weak or irrelevant correlations and yielding binary networks at each threshold. We calculated small-world properties and six network parameters: clustering coefficient (Cp), characteristic path length (Lp), global efficiency (GE), local efficiency (LE), node degree (ND), and node efficiency (NE). Statistical analysis Statistical analyses were performed using IBM SPSS Statistics for Windows, Version 27.0 (IBM Corp, Armonk, NY, USA; https://www.ibm.com/products/spss-statistics). Continuous variables that satisfied both normality and homogeneity of variance assumptions are presented as mean ± standard deviation (SD). Categorical variables are summarized as counts and percentages. A two-way repeated-measures analysis of variance (ANOVA) was performed for SSA scores, PAS ratings, and cortical activation intensities, with two within-subject factors: time (pre-treatment vs. post-treatment) and group (Control, Sham-iTBS + SET, and Active-iTBS + SET). For any significant main effects or interactions, post-hoc tests with Bonferroni correction were applied to control for type I error. Effect sizes for all ANOVAs are reported as partial eta-squared (η2p), with values interpreted as follows: 0.01 for small, 0.06 for medium, and 0.14 for large effects5. Paired t-tests and one-way ANOVA were used to compare within- and between-group differences in functional connectivity across brain regions, with family-wise error (FWE) correction applied for multiple comparisons. One-way ANOVA was also used to examine between-group differences in brain network metrics, while paired t-tests assessed within-group changes over time. Pearson correlation analysis was conducted to evaluate the relationship between channels showing significant activation and swallowing function outcomes across groups. Demographic and baseline clinical data were compared using chi-square or F tests, as appropriate. Results General results During the treatment period, three patients dropped out: one from the Control group due to changes in their condition, one from the Sham-iTBS + SET group who transferred hospitals and interrupted treatment, and one from the Active-iTBS + SET group who voluntarily withdrew. Ultimately, a total of 42 patients were enrolled in the study. These included 14 patients in the Control group, 14 in the Sham-iTBS + SET group, and 14 in the Active-iTBS + SET group. No significant differences were observed among the three groups in baseline characteristics (p > 0.05). Please see Table 1 for details. Table 1Characteristics of demographic and swallowing function assessment at baseline.Control group(n = 14)Sham-iTBS + SET group (n = 14)Active-iTBS + SET group(n = 14)F/χ2pGender1.0260.368Male9 (64.3%)12 (85.7%)9 (64.3%)Female5 (35.7%)2 (14.3%)5 (35.7%)Age(years)58.21 ± 13.0251.43 ± 13.1762.14 ± 7.813.0550.059Type of stroke0.9070.412Hemorrhagic10 (71.4%)10 (71.4%)7 (50%)Ischemic4 (28.6%)4 (28.6%)7 (50%)Duration since stroke(months)1.93 ± 0.921.50 ± 0.942.00 ± 0.781.3120.281Lesion Side0.1150.892Left5 (35.7%)4 (28.6%)4 (28.6%)Right2 (14.3%)4 (28.6%)2 (14.3%)Bilateral7 (50%)6 (42.9%)8 (57.1%)MMSE24.43 ± 3.1123.79 ± 2.2925.07 ± 2.870.7500.479NIHSS12.00 ± 1.1111.64 ± 0.7511.71 ± 0.730.6480.528SSA38.71 ± 1.9838.86 ± 1.7539.57 ± 1.650.9140.409PAS5.93 ± 0.485.64 ± 0.635.43 ± 0.512.9690.063 Characteristics of demographic and swallowing function assessment at baseline. Clinic outcomes Before treatment, no significant differences were observed in primary outcome (SSA) or the secondary behavioral outcome (PAS) scores among the three groups. After 4 weeks of treatment, all three groups showed a significant reduction in SSA and PAS scores compared to baseline (p < 0.05). Post-treatment analysis revealed that both the Sham-iTBS + SET Group and the Active-iTBS + SET Group indicated significantly lower SSA scores than the Control Group (p < 0.05). Notably, the Active-iTBS + SET Group exhibited a more pronounced reduction in SSA scores compared to the Sham-iTBS + SET Group (p < 0.05). Similar to the primary findings, both iTBS intervention groups showed significantly better PAS scores than the Control Group after treatment (p < 0.05). Furthermore, the Active-iTBS + SET Group demonstrated greater improvement in PAS scores compared to the Sham-iTBS + SET Group (p < 0.05). Detailed results are presented in Table 2. Table 2SSA and PAS scores comparison.GroupPre-interventionPost-interventionFpη2pSSAControl group38.71 ± 1.9832.93 ± 1.27a10.450 < 0.0010.349Sham-iTBS + SET Group38.86 ± 1.7531.29 ± 1.49abActive-iTBS + SET Group39.57 ± 1.6529.57 ± 1.56abcPASControl group5.93 ± 0.484.71 ± 0.73a5.5400.0080.221Sham-iTBS + SET Group5.64 ± 0.634.00 ± 0.78abActive-iTBS + SET Group5.43 ± 0.513.29 ± 0.61abcF, P, η2p are the values for the interaction effect. Compared with pre-treatment, a represents p < 0.05; at the same time point, compared with the control group, b represents p < 0.05; at the same time point, compared with the sham stimulation group, c represents p < 0.05. 3.3 Brain activity. SSA and PAS scores comparison. F, P, η2p are the values for the interaction effect. Compared with pre-treatment, a represents p < 0.05; at the same time point, compared with the control group, b represents p < 0.05; at the same time point, compared with the sham stimulation group, c represents p < 0.05. 3.3 Brain activity. Before treatment, no statistically significant differences in functional connectivity values were observed between groups across channel pairs (p > 0.05). However, after treatment, a notable increase in activation was observed in the RPM in the Active-iTBS + SET Group. A significant interaction effect between group and time was found (ch1: F(2, 39) = 5.131, p = 0.011, η2p = 0.208). A significant main effect of time was also observed, F(1, 39) = 8.687, p = 0.005, η2p = 0.182, with post-hoc tests revealing significantly increased activation in the Active-iTBS + SET Group after treatment compared to baseline (t = 11.794, p < 0.001, 95% CI[-2.076E-06, -5.35E-07]). Furthermore, a significant main effect of group was identified, F(2, 39) = 5.275, p = 0.009, η2p = 0.213. Post-hoc analysis indicated that the Active-iTBS + SET Group exhibited significantly stronger activation than the Control Group (p = 0.009, 95% CI [3.463E-07, 2.877E-06]) and the Sham-iTBS + SET Group (p = 0.038, 95% CI [5.591E-08, 2.587E-06]) after treatment. Similarly, enhanced activation was detected in theRPFC, with a significant group × time interaction (ch20: F(2, 39) = 3.308, p = 0.047, η2p = 0.145). No significant main effect of time was observed. However, a significant main effect of group was found, F(2, 39) = 3.323, p = 0.046, η2p = 0.146. Multiple comparisons showed that the Active-iTBS + SET Group had significantly higher activation than the Sham-iTBS + SET Group after treatment (p = 0.026, 95% CI [1.496E-07, 3.05E-06]) (Table 3 and Fig. 3). Table 3Activated channels after treatment.GroupPre-interventionPost-interventionInteraction effectMeanSDMeanSDFpη2pCH1Control group3.87E-075.30E-071.13E-084.47E-075.1310.0110.208Sham-iTBS + SET Group4.37E-074.47E-073.02E-074.44E-07Active-iTBS + SET Group3.21E-073.22E-071.62E-062.23E-06CH20Control group6.09E-071.03E-061.58E-078.73E-073.3080.0470.145Sham-iTBS + SET Group6.56E-071.08E-065.49E-093.90E-07Active-iTBS + SET Group4.68E-077.70E-071.61E-062.48E-06 Activated channels after treatment. Comparison of cortical activation among the three groups Note: The brain activation maps from left to right represent the F-values of the time main effect, interaction effect, and group main effect, respectively. Redder colors indicate higher F-values, while bluer colors indicate lower values. The colorbar represents the range of F-values. Function connection Before treatment, there were no statistically significant differences among the groups (p > 0.05). After treatment, all three groups showed improvements compared to before treatment (p < 0.05). Figure 4 (1) shows that the improvement was more significant in the Active-iTBS + SET group. Post-hoc analysis revealed that, compared to the Control group, the Active-iTBS + SET group significantly enhanced functional connectivity between the following brain regions (p < 0.05, FWE-corrected): RPM and LS1 (ch1-ch13, ch1-ch14, ch13-ch15), RPM and LPM (ch1-ch29, ch1-ch33, ch1-ch35, ch15-ch33, ch15-ch35, ch18-ch35, ch30-ch34, ch30-ch35), RPM and LM1 (ch1-ch32, ch15-ch32, ch30-ch32), RPFC and LS1 (ch5-ch14), RPFC and LM1 (ch5-ch32, ch19-ch32), LPFC and LS1 (ch8-ch13, ch10-ch13, ch13-ch26, ch13-ch33, ch13-ch35), LPFC and RS1 (ch9-ch16, ch10-ch16, ch10-ch18), LPFC and RPM (ch9-ch18, ch15-ch26, ch15-ch27, ch18-ch26), LPFC and LM1 (ch10-ch32), LPFC and LPM (ch10-ch33), LS1 and RM1 (ch13-ch17, ch14-ch17), LS1 and LM1 (ch13-ch32), LS1 and RS1 (ch14-ch16), LPM and RS1 (ch16-ch29, ch16-ch33, ch16-ch34, ch16-ch35), LM1 and RS1 (ch16-ch32), RM1 and LPM (ch17-ch29), LM1 and RM1 (ch17-ch32), LPM and RM1 (ch17-ch33, ch17-ch34, ch17-ch35), and LPM and LM1 (ch32-ch35). Comparison of functional connectivity before and after treatment among the three groups Note: Fig. 4 (1) shows the mean functional connectivity values and differences across all channels for the three patient groups before and after treatment. From left to right are the pre-treatment, post-treatment, and difference values. From top to bottom are the Control group, Sham-iTBS + SET Group, and Active-iTBS + SET Group. The colorbar indicates the range of Δr. Red represents post-treatment values greater than pre-treatment, while blue represents post-treatment values less than pre-treatment. Figure 4 (2) displays the results of post-treatment intergroup multiple comparisons. From left to right are the topological and channel maps of significant channels where the Active-iTBS + SET Group differed from the Control group, followed by the topological and channel maps of significant channels where the Active-iTBS + SET Group differed from the Sham-iTBS + SET Group. FWE correction, p < 0.05. Brain network topological properties Within the threshold range of 0.1 to 0.8, all three groups’ brain networks exhibited small-world characteristics (σ > 1). Figure 5 presents a comparison of the areas under the curve (AUC) for six topological network properties both before and after treatment within the specified threshold range. From the topological maps of network properties, it is evident that, the brain network indices in the Control group showed varying degrees of improvement after treatment, with the exception of ND. The Sham-iTBS + SET group indicated significant enhancements in Cp, Lp, and ND, alongside a marked reduction in GE and NE. Meanwhile, the Active-iTBS + SET group maintained superior overall effects. One-way ANOVA results indicated significant differences in ND among the three groups before treatment (F(2, 39) = 46.486, p < 0.001, ω2 = 0.684). After treatment, ND continued to show significant differences among groups (F(2, 39) = 62.054, p < 0.001, ω2 = 0.744). No significant differences were observed in other network properties. Topological properties of brain networks in the three groups Note: Cp: Clustering coefficient, Lp: characteristic path length, GE: global efficiency, LE: local efficiency, NE: node efficiency, ND: node degree. Topographic maps of each network property depict the AUC within the relevant threshold range (0.1–0.8). The AUC is presented as the mean value across all subjects. Normalized mapping was applied, with the colorbar indicating the normalized data values of network properties. Redder colors represent higher values, while bluer colors indicate lower values. Significant baseline differences in ND were observed among the groups. In small-sample studies, such imbalances may arise from individual variability and stochastic distribution. To address this, we prioritized longitudinal within-group comparisons. At baseline, both the Sham-iTBS + SET group (p < 0.001, 95% CI [− 0.578, − 0.342]) and the Active-iTBS + SET group (p < 0.001, 95% CI [− 0.628, − 0.392]) exhibited significantly higher ND than the Control group. After treatment, both the Sham-iTBS + SET group (p < 0.001, 95% CI [− 0.621, − 0.390]) and the Active-iTBS + SET group (p < 0.001, 95% CI [− 0.698, − 0.468]) maintained significantly higher ND compared to the Control group. Within-group longitudinal analyses revealed no significant improvement in the Control Group (t(13) = -0.609, p = 0.553, d = − 0.163, 95% CI = [− 0.687, − 0.368]). In contrast, both the Sham-iTBS + SET Group (t(13) = − 2.658, p = 0.020, d = − 0.710, 95% CI = [− 1.289, − 0.110]) and Active-iTBS + SET Group (t(13) = -2.572, p = 0.023, d = − 0.687, 95% CI = [− 1.262, -0.091]) showed significant enhancements in ND following treatment. These analyses indicated that significant post-treatment enhancements occurred exclusively in the Active-iTBS + SET and Sham-iTBS + SET groups, while the Control group remained stable (Fig. 6). Multiple comparisons of Node degree before and after treatment among the three groups ND: Node degree. * Indicates p < 0.001 compared with the Control group before treatment, # indicates p < 0.001 compared with the Control group after treatment. Correlation between cortical activation and behavioral To further investigate the association between treatment-induced changes in cortical activation intensity and swallowing function, Spearman correlation analysis was performed to assess the relationship between mean HbO₂ concentration and SSA scores. The results revealed a significant positive correlation between activation in the RPM (ch1: r = 0.570, p < 0.001) and the RPFC (ch20: r = 0.468, p = 0.001) and SSA scores. As shown in Fig. 7. Correlation analysis between behavioral data and activation intensity. Adverse events No serious adverse events occurred during the clinical study. The reported adverse events involved transient palpitations in three participants from the experimental group during iTBS treatment. These episodes were short-lived, expected, and did not interfere with the treatment protocol. All participants had been fully informed of these potential side effects before the commencement of the study. Discussion This study demonstrates that the combination of bilateral cerebellar iTBS with SET synergistically enhances swallowing function and cortical neuroplasticity in post-stroke dysphagia patients. Our findings revealed that this dual intervention significantly improves clinical scores (SSA and PAS) and positively impacts neural substrates, particularly through increased activation in the RPFC and premotor areas, alongside enhanced whole-brain network integration. This supports a multimodal rehabilitation framework in which central neuromodulation and peripheral strength training work together to facilitate functional recovery. Synergistic effects on neurobehavioral recovery Both intervention groups showed improved SSA and PAS scores compared to the Control Group. This suggests that SET, facilitated by real-time audiovisual biofeedback, effectively activates swallowing centers by strengthening relevant muscle groups and optimizing laryngeal movement26,31. Notably, the Active-iTBS + SET group experienced a 25% improvement, which exceeded both our previous findings and standalone cerebellar stimulation results SSA scores. This increased efficacy likely stems from a “central-peripheral positive feedback loop.” While cerebellar iTBS regulates the dentato-thalamo-cortical pathway to enhance timing control over brainstem swallowing centers28, SET addresses muscular coordination deficits through isometric and isotonic training. Together, these approaches align with the Central-Peripheral-Central (CPC) closed-loop theory12, where central timing regulation and peripheral sensory feedback reinforce each other. Cortical activation and changes in brain connectivity A significant finding from our research is the significant activation of RPM and RPFC in the Active-iTBS + SET Group, which distinguishing this intervention from traditional therapies targeting the M1. While pure cerebellar stimulation typically activates the primary motor area33, our results suggest that the incorporation of SET shifts neural activity toward higher-order planning regions. This indicates that the iTBS + SET protocol may reduce compensatory inhibition in the prefrontal cortex, thereby enhancing its regulatory role in swallowing execution15,21. Furthermore, functional connectivity analysis demonstrated improved bilateral hemispheric coordination (LPFC-LPM and LPFC-LM1). This highlights the cerebellum’s role in optimizing the temporal and spatial integration of sensory inputs and motor outputs through cortico-brainstem circuits. The positive correlation between RPM/RPFC activation and SSA scores further identifies these regions as potential therapeutic targets for post-stroke recovery, resonating with Koch’s15 findings on the auxiliary motor area as a core swallowing hub for swallowing. Brain network reorganization The Active-iTBS + SET Group made significant advances in optimizing small-world properties, which are essential for efficient neural information processing3. While stroke often disrupts both local connectivity and global efficiency, our intervention successfully enhanced network integration by strengthening key nodes, including the RPM and LPFC. This reorganization extends beyond localized changes to encompass systemic improvements, providing a solid neural foundation for the functional gains observed. These topological modifications may also serve as valuable biomarkers for tracking personalized rehabilitation progress. Limitations Although this study presents significant findings, several limitations must be acknowledged. Firstly, the modest sample size constrained the depth of neural analysis and hindered stratified assessments based on the clinical heterogeneity of stroke, such as lesion location or lateralization. This limitation precluded a more granular interpretation of functional connectivity patterns and their potential modulation by lesion topography. Secondly, notable baseline differences in ND were observed among the groups, attributed to individual variability. While we prioritized longitudinal within-group comparisons to confirm that enhancements occurred exclusively in the iTBS-intervention groups, with the Control group remaining stable, these initial disparities still pose a methodological challenge. Thirdly, the technical limitations of fNIRS restricted our mechanistic analysis to superficial cortical hemodynamics, meaning that the observed modulation of deep structures critical for swallowing, such as the cerebellum or brainstem, is inferred indirectly through statistical covariance rather than directly visualized6. Lastly, outcomes were assessed post-intervention immediately, and the lack of long-term follow-up data prevents us from determining the durability and sustainability of the observed neurorestorative gains. Future research should employ multi-center, large-scale designs combined with high-spatial-resolution neuroimaging (e.g., fMRI) to conduct longitudinal follow-up and stratified analyses, further validating the long-term mechanisms and clinical generalizability of this dual-mode intervention. Conclusion The combination of bilateral cerebellar iTBS and pharyngeal elevation training significantly enhances swallowing recovery in individuals with post-stroke dysphagia. This improvement is achieved through the coordinated activation of the right premotor and prefrontal cortices, leading to reorganization within the whole-brain network. These findings are particularly relevant to the post-stroke recovery phase and the positive effects of this intervention on swallowing function and neural remodeling necessitate further validation in larger clinical cohorts. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements The authors would like to thank all of the clinicians and researchers who were involved with this study, and all of the participants who took part in this study. Author contributions Haiquan Zhang was responsible for methodological development, data analysis, and manuscript preparation. Jingyun Tan contributed to data analysis, methodological implementation, software programming, and initial draft writing. Wei Lu conducted data curation, experimental validation, and investigative oversight. Di Zhang performed data collection and experimental execution. Min Liu secured funding, supervised resource allocation, and participated in manuscript revision. Shanshan Wang served as the Principal Investigator, leading study conceptualization, funding acquisition, methodological design, resource coordination, and manuscript critical editing. All authors reviewed and approved the final submitted manuscript. Funding This study was supported by Jinan Clinical Science and Technology Innovation Program (No: 202328060) and Provincial Key Specialties Program of Shandong Provincial Third Hospital (No: SJZK002). Data availability The datasets generated during this study are available from the corresponding author upon reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Haiquan Zhang and Jingyun Tan contributed equally to this work and should be regarded as co-first authors. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets generated during this study are available from the corresponding author upon reasonable request.