Tai Ji for stroke rehabilitation: a randomized controlled trial on improving upper limb function and cortico-muscular coupling in stroke patients Authors: Qiurong Xie, Yurong Li, Nan Zheng, Qi Zhang, Xiaoling Wang, Jia Luo, Yanxin Zhang, Jing Tao, Jia Huang Published: 2026 ABSTRACT AbstractObjectiveThis study investigates the impact of integrating Tai Ji into rehabilitation on sensorimotor cortex and upper limb muscle activation in stroke patients, addressing the current lack of research on its underlying mechanisms, despite its well-documented benefits for upper limb function.MethodsEighty-four stroke patients were randomized into a Tai Ji group (n = 42) and a control group (n = 42). Additionally, a healthy reference group (n=[30]) was recruited to establish normative baseline synergy patterns. Both groups received conventional rehabilitation, while the Tai Ji group practiced 8-Form Tai Ji for 60 min, five times a week, over four weeks. Functional outcomes were assessed using the Fugl-Meyer Assessment for the Upper Extremity (FMA-UE) and the Modified Barthel Index (MBI). Neural and muscular activities were evaluated by analyzing the coherence and directional connectivity of synchronous electroencephalography and surface electromyography (EEG-sEMG) using Partial Transfer Entropy (PTE), which quantifies information flow between the sensorimotor cortex and muscles.ResultsAfter 4 weeks, the Tai Ji group demonstrated significantly greater improvements in upper-limb motor function (FMA-UE total score and subscales) and activities of daily living (MBI) than the control group (P < 0.05). Neurophysiological analysis revealed that Tai Ji training enhanced descending functional connectivity from the contralateral motor cortex to key upper limb muscles and facilitated ascending connectivity from muscles to the sensory cortex (P < 0.05), changes associated with a shift toward normalization of pathological muscle synergy patterns. Furthermore, exploratory analyses indicated trend-level associations between improvements in specific motor functions (e.g., hand, wrist) and the modulation of these specific pathways.ConclusionsSupplementing standard rehabilitation with a 4-week Tai Ji program was associated with significant improvement upper-limb limb motor function and activities of daily living in stroke patients. Mechanistically, these benefits were linked to a dual-process neural adaptation: strengthening of the descending corticomotor drive concurrent with facilitation of ascending sensory feedback and normalization of pathological muscle synergies. This bidirectional optimization of cortico-muscular connectivity may constitute a neurophysiological correlate of the observed functional improvements.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12984-026-02054-y.Keywords: Tai Ji, Stroke, Upper extremity, Cortico-muscular coupling, Muscle Introduction Upper limb motor impairment is one of the most common and disabling consequences of stroke, affecting a majority of survivors and presenting a significant challenge to functional independence. Recovery of dexterous upper limb function is notably slower and less complete than that of the lower limbs. This disparity critically impairs fine motor skills essential for daily living, as evidenced by the low rate of complete functional recovery—for instance, only a minority of patients regain full dexterity even months after stroke [1]. Effective rehabilitation is therefore crucial for enhancing quality of life, reducing care dependency, and promoting societal reintegration after a stroke. Tai Ji, a traditional Chinese mind-body practice, emphasizes the harmonization of intention, movement, and breath. Its slow, fluid, and coordinated movements are hypothesized to enhance the brain’s integrative and regulatory functions, making it a promising intervention for stroke rehabilitation [2]. Meta-analyses confirm that integrating Tai Ji into upper limb rehabilitation yields superior functional outcomes compared to standard training alone [3]. Potential neurophysiological substrates for these benefits include enhanced sensorimotor integration and increased cortical thickness [4], greater muscle activation [5], and elevated neural excitability in stroke patients [6]. Furthermore, Tai Ji improves movement patterns and joint coordination [7]. Recovery of upper extremity function relies on the intact interaction within the sensory-motor network, a process quantified as functional corticomuscular coupling (fCMC). Human movement relies heavily on the dynamic interaction between the brain and muscles, governed by the cortical motor system, which involves intercortical connections, cortical-muscle pathways, and intramuscular networks [8, 9]. fCMC represents the bidirectional, closed-loop communication between the brain and muscles, encompassing both descending motor commands and ascending sensory feedback [10]. Analyzable via EEG and sEMG, fCMC is crucial for voluntary motor control [11] and its restoration parallels motor recovery post-stroke [12]. However, stroke disrupts this interplay, leading to weakened fCMC [13–16] and markedly lower fCMC levels compared with those of healthy individuals [12]. Consequently, restoring fCMC is a key rehabilitation goal, as evidenced by its association with functional gains and the efficacy of interventions like virtual reality training [17], transcranial stimulation [18], and robotic therapy [19] in enhancing fCMC. Even simpler interventions, such as unilateral training [20] and neuromusic treatment [21], can modulate fCMC, underscoring its role as a critical neurophysiological substrate for recovery. Parallel to impaired fCMC, stroke induces abnormal muscle coordination (dyssynergy), characterized by excessive co-activation (e.g., deltoid/trapezius) and reduced activation of key muscles such as triceps brachii (TRI) [22–24]. These aberrant synergies contribute directly to dyscoordination and inefficient movement [25, 26]. The severity of such coordination abnormalities correlates with clinical impairment [23, 27–30], and their normalization is essential for functional recovery. Therefore, successful rehabilitation requires interventions that can simultaneously address both the neural drive (via fCMC) and the peripheral execution (via muscle coordination). Critically, the bidirectional nature of fCMC underscores that sensory feedback is equally vital for both motor learning and recovery [31, 32]. Facilitating this cortex-muscle dialogue enhances outcomes [33, 34]. However, critical knowledge gaps persist. Specifically, direct evidence for Tai Ji’s modulatory effect on fCMC is lacking, the neurophysiological pathways through which it restores this communication are unknown, and its capacity to normalize pathological post-stroke muscle synergies remains unconfirmed. Building on this framework, we hypothesize that Tai Ji training facilitates upper limb recovery by specifically enhancing bidirectional fCMC and improving muscle coordination patterns. This randomized controlled trial evaluates the clinical efficacy of Tai Ji and employs synchronous EEG-sEMG to quantitatively assess its effects on fCMC and muscle synergy, correlating these neurophysiological changes with functional gains. This research aims to elucidate the mechanisms of Tai Ji-induced cortical-muscle reorganization, advancing a targeted, mechanism-based approach to stroke rehabilitation. Materials and methods Study design and registration This study employed a parallel-group, single-blind, randomized controlled trial design, following approval from the ethics committee of the participating hospital (Approval Number: 2021KY-005-02). The study was conducted in accordance with the principles outlined in the Declaration of Helsinki. Furthermore, the trial was registered with the Chinese Clinical Trial Registry (Registration Number: ChiCTR2200061376). Participants Sample size calculation An a priori sample size calculation was performed using GPower software (version 3.1.9.7) for an independent two-sample t-test. The estimate was based on detecting a significant between-group difference in the primary outcome, the Fugl-Meyer Assessment for the Upper Extremity (FMA-UE), after the intervention. The effect size (Cohen’s d = 0.83) was derived from the post-intervention FMA-UE scores (mean ± SD: 51.09 ± 14.83 vs. 37.36 ± 17.92) reported in a prior randomized controlled trial of a comparable Tai Ji intervention in stroke patients [35]. With a two-sided alpha of 0.05 and 90% power, the calculation indicated a requirement of 32 participants per group (64 in total). To account for an estimated 20% dropout rate, the sample size was inflated: 32/(1 − 0.2) = 40 participants per group. Conservatively rounding up, we planned to recruit 42 participants per group, for a total target sample size of 84. Recruitment and groups Stroke patients Stroke patients at Fuzhou Hospital were invited to participate in the study through advertisements posted on hospital bulletin boards. Before screening, the research team explained the study’s objectives to interested individuals and obtained informed consent. Diagnosis was confirmed by professional therapists according to the Diagnostic Criteria of Cerebrovascular Diseases in China (Version 2019) [36]. Eligible patients were then formally enrolled. The inclusion criteria included: first-ever stroke, presence of unilateral cortical lesions (in either the left or right hemisphere) and/or subcortical lesions involving the motor pathways; onset time between 2 weeks and 12 months; age ranging from 18 to 70 years; Brunnstrom stages III to V for the affected upper limb; Manual Muscle Testing (MMT) grade ≥ 3 for key muscles of the affected upper limb (deltoid, TRI), with a Modified Ashworth Scale (MAS) score for muscle tone < 2; right-handedness prior to the stroke; a standing balance score of ≥ 2 as assessed by the standard clinical balance function grading scale (grades 0–4, indicating the ability to stand independently and maintain equilibrium during slight perturbations) [37]; ability to stand independently for over 5 min and walk independently for over 6 m; and no contraindications for electroencephalography (EEG) and surface electromyography (sEMG) examinations (including: open wounds or severe skin conditions at electrode sites; presence of cardiac pacemakers or other implanted electronic devices; history of photosensitive epilepsy (for EEG); and general inability to tolerate the testing procedure). Exclusion criteria included upper limb motor dysfunction caused by other brain diseases (e.g., brain tumors); patients with severe complications (e.g., shoulder-hand syndrome); individuals with severe heart disease, malignant tumors, gastrointestinal bleeding, or failure of cardiac, hepatic, or renal function; and patients not suitable for EEG and sEMG examinations. Healthy controls A cohort of age- and sex-matched healthy adults (n = 30) was recruited from the community to serve as a normative reference. This group allows us to establish a physiological benchmark for normal muscle synergy patterns, which is essential to: (1) objectively define the pathological synergy patterns present in our stroke cohort at baseline; and (2) interpret the direction and quality of changes post-intervention, distinguishing general ‘change’ from specific ‘normalization’ toward a healthy state. Inclusion criteria were: (1) no history of neurological, musculoskeletal, or cardiovascular disorders; (2) no upper limb pain or injury in the past year; (3) ability to complete the experimental task. Exclusion criteria were: (1) any regular structured exercise training that might significantly alter neuromuscular patterns (e.g., professional or competitive sports); (2) pregnancy. Randomization and blinding Eligible participants were randomly randomized in a 1:1 ratio to either the Tai Ji or control groups. An independent statistician used SAS 9.1 software to generate the randomized sequence. The assignments were sealed in sequentially numbered, opaque envelopes. To implement allocation concealment, a research assistant not involved in recruitment or assessment opened the next envelope in sequence and informed the participant of their group assignment. This was a single-blind study. Outcome assessors and the statistician performing the final data analysis were blinded to group allocation throughout the trial. Procedures Baseline assessments were conducted one week prior to the intervention, while post-intervention evaluations were scheduled within one week following the final training session, which occurred in the fourth week. During both assessment phases, research team members carried out upper limb motor function tests and EEG and sEMG assessments. Participants in the control group received standard medical treatment, health education, and upper limb rehabilitation training. The standard medical treatment included intravenous thrombolysis, endovascular therapy, antiplatelet therapy, management of blood pressure and blood glucose levels, pharmacological interventions, and management of complications. In addition, subjects received health education that addressed the severity of stroke, risk factors, triggers, secondary prevention measures, and coping and rehabilitation strategies. Experienced clinicians provide standard medical treatment and health education. Concurrently, participants received upper extremity rehabilitation provided by occupationally certified therapists to improve functional recovery outcomes. This conventional upper limb rehabilitation, focusing on task-oriented training, stretching, and strengthening exercises, was conducted for 60 min per session, five times a week, over a total duration of four weeks. Participants in the Tai Ji group received the same standard medical treatment, health education, and upper limb rehabilitation training as those in the control group, supplemented with a clinically adapted Tai Ji intervention. This intervention consisted of a simplified, eight-form program specifically designed for stroke rehabilitation in collaboration between rehabilitation therapists and Tai Ji masters. The forms focused on therapeutic movements such as weight shifting and trunk rotation, while explicitly excluding challenging postures common in community practice (e.g., deep stances, complex single-leg balances). This training was led by instructors with over five years of experience, who employed a progressive approach from isolated to complex movements. Critically, it followed a standardized protocol that incorporated real-time, individualized adjustments based on each participant’s functional capacity (e.g., Brunnstrom stage, balance score). A multi-tiered assistance strategy was implemented to ensure safety and feasibility: exercise modifications (e.g., widening stance, reducing range of motion, using chair support), hands-on guidance from instructors to facilitate correct patterns, simplification of complex forms into components, and continuous supervision in a safe environment with support bars. The selected eight forms encompassed the Starting Posture, Parting the Wild Horse’s Mane on both sides, Brush Knee and Twist Step on both sides, Forearm Rollings on both sides, Grasp the Bird’s Tail-Left Side, Grasp the Bird’s Tail-Right Side, Cloud Hands, and Closing Form. Each training session lasted 60 min, comprising a 5-minute warm-up, 50 min of Tai Ji practice, and a 5-minute cool-down. Training was conducted five times a week for a total duration of four weeks. One researcher used an exercise log to record the frequency of training sessions, completion status, and any adverse events. Outcome measures The outcome measures of this study included two clinical assessment tools: the FMA-UE and the Modified Barthel Index (MBI), as well as outcome measures obtained from data collected through EEG and sEMG, which emphasized the functional connectivity between the cortex and the muscles. EEG-sEMG data collection This study collected EEG data using a Neuroscan 64-channel EEG system (NeuroScan, El Paso, Texas, USA) in conjunction with its synchronized recording system, which included the QuikCap and SynAmp2 amplifier. sEMG data were obtained using Noraxon’s sEMG equipment. EEG signals were recorded according to the international 10–20 system across 64 scalp locations, with a reference electrode placed on the mastoid. The sampling frequency for the EEG was set to 1 kHz. Simultaneously, multi-channel sEMG signals were recorded from various muscles, including the Pectoralis Major (PM), Upper Trapezius (UT), Anterior Deltoid (AD), Middle Deltoid (MD), Posterior Deltoid (PD), Biceps Brachii (BIC), TRI, Flexor Carpi Radialis (FCR), and Extensor Carpi Radialis (ECR), using the Noraxon system, which had a sampling frequency of 1.5 kHz. The synchronization of EEG and sEMG data was achieved through a pulse synchronizer, which activated both data collection devices upon receiving a recording signal from E-Prime, ensuring precise alignment of the two datasets. Participants performed a forward-reaching movement to assess upper limb functional connectivity as the experimental task. They were seated comfortably at a table with their backs supported against the chair to minimize leaning influences. To standardize the task difficulty individually, the reaching distance was determined based on each participant’s arm length. Point A was defined as the starting position of the hand on the table, directly in front of the shoulder of the affected limb. Point B was then marked on the table at a distance equal to the participant’s full arm length (measured from the acromion to the tip of the clenched fist), directly forward from Point A. The participant’s forearms were placed on the table to facilitate the smooth execution of the reaching task. Participants were instructed to form a fist and move their hand from point A to point B, then return to point A, repeating this process ten times. A red circular cue appeared on the screen one second before each movement began, followed by images related to the forward reach action. Upon seeing the cue, participants performed the forward reach for a duration of 3 s. Once the action was completed, the screen turned black to signal a brief rest period, helping to mitigate muscle fatigue. Critically, the presentation software (E-Prime) sent a synchronized trigger pulse (marker) to both the EEG and sEMG recording systems at the exact onset of each visual cue, thereby providing an objective and precise time-lock for the initiation of every forward-reaching movement. EEG-sEMG data processing EEG data preprocessing was performed using the EEGLAB toolbox in MATLAB, involving several critical steps. First, electrode localization was conducted, and invalid electrodes were removed. The data were then subjected to whole-brain average reference re-referencing, followed by bandpass filtering in the frequency range of 0.5 to 45 Hz. Subsequently, the data were downsampled to 250 Hz. After downsampling, Independent Component Analysis (ICA) was applied for further refinement. The data were segmented, baseline corrected, and segments of poor quality were excluded. At the beginning of preprocessing, raw electrode channels were aligned with standard channels to eliminate interference signals, such as ocular artifacts. The average reference was utilized to mitigate baseline effects. Current research suggests that the normal frequency range of EEG signals is between 0.5 and 50 Hz, encompassing the δ, θ, α, β, and part of the γ frequency bands. Our focus was primarily on the β band (15–30 Hz), which is closely associated with sensorimotor function. To enable the investigation of corticomuscular coupling, a symmetric preprocessing pipeline was applied to both EEG and sEMG signals. Specifically, both were bandpass filtered with an identical range of 0.5–45 Hz. For EEG, this extracts movement-related cortical oscillations. For sEMG, this isolates the low-frequency amplitude envelope of the muscle activity while attenuating the high-frequency action potential details. This standard filtering band also suppresses high-frequency noise and 50 Hz line interference. Subsequently, all signals were downsampled to a unified rate of 250 Hz to ensure temporal alignment and facilitate synchronized analysis. Following these initial steps, data were segmented into epochs for analysis. Each epoch was time-locked to the onset of a single forward-reaching movement from Point A to Point B. The onset was objectively defined by the recorded trigger marker sent at the visual cue. Data epochs were extracted from 500 ms before this movement onset marker to 3 s after it, with the 500 ms pre-movement segment used for baseline correction. Segments with amplitudes exceeding ± 100 µV were removed to ensure optimal EEG signal quality. After completing these preprocessing steps, we aimed to investigate functional connectivity between brain regions by categorizing the cerebral cortex into four distinct functional areas based on electrode positions: Left Motor Cortex (LM; channels FC1, FC3, C1, C3), Right Motor Cortex (RM; channels FC2, FC4, C2, C4), Left Sensory Cortex (LS; channels CP1, CP3), and Right Sensory Cortex (RS; channels CP2, CP4).(Fig. 1). Data Collection. A.Study Schematic. The black squares represent the locations where the sEMG electrodes of the sEMG device are placed. B.EEG Electrode Configuration Functional connectivity analysis This study used the Partial Transfer Entropy (PTE) method to assess functional connectivity between the cortex and muscles. Transfer entropy, rooted in information theory, quantifies the transfer of information between variables. Its inherent asymmetry makes it particularly suitable for evaluating the directionality of information flow between two time series. PTE extends this concept by focusing on direct causal relationships within multivariate systems, thereby elucidating functional connections that interactions among channels may obscure. We evaluated the functional connectivity between the cortex and muscles by analyzing the correlations between EEG signals from cortical areas and sEMG signals from muscles. This analysis was conducted using the HERMES toolbox on the MATLAB platform. We first employed a nearest-neighbor method to determine the embedding dimension and configured the embedding delay using the moments of the autocorrelation function less than 1/e. Subsequently, we calculated PTE, which is essential for multichannel functional connectivity analysis, to compute PTE between cortical source signals and EMG signals. To facilitate data analysis and presentation of results, we designated the right hemisphere as the affected hemisphere and performed left-right channel swapping for participants with left-hemisphere lesions. To maintain consistency in lesion locations among stroke patients, we normalized the ipsilateral hemisphere to the right hemisphere by flipping the channels across the midline, standardizing the hemiplegic limb to the left side (with the right side representing the affected hemisphere). The EEG and sEMG electrode configurations remained unchanged for patients with lesions in the right hemisphere. In contrast, we adjusted the left-side EEG channels for those with left-hemisphere lesions to correspond with the right-side channels, referencing the interhemispheric fissure line, while simultaneously relocating the channels initially positioned on the right side to the left side. Extraction of muscle synergy patterns Following the filtering and envelope extraction processes, we analyzed the synergy patterns of the muscle groups by employing the Non-negative Matrix Factorization (NMF) algorithm for dimensionality reduction of the sEMG signal matrix. NMF is particularly well-suited for this purpose as it decomposes a non-negative matrix into two non-negative matrices, effectively revealing the underlying patterns within the data. This approach enabled us to identify distinct muscle activation patterns that characterize the interdependencies and synergistic functions among the various muscles involved in the movement tasks. Statistical analysis We employed independent-samples t-tests or Mann-Whitney U tests to compare baseline characteristics (e.g., age, upper-limb function) between groups. For categorical variables (e.g., gender), Chi-square tests were used. Within- and between-group changes in behavioral outcomes were assessed using paired-samples t-tests (or Wilcoxon tests) and independent-samples t-tests (or Mann-Whitney U tests), respectively. The analysis of corticomuscular connectivity was pre-specified to test our primary hypotheses concerning the descending (cortex-to-muscle) and key ascending (muscle-to-sensory cortex) pathways. To statistically analyze changes in these pathways, we calculated the difference in average β-band PTE values (post-training minus pre-training) for each participant. These change scores were then compared between the Tai Ji and control groups using independent-samples t-tests, or Mann-Whitney U tests (α = 0.05). For transparency, connectivity values for the full directional matrix (including sensory cortex-to-muscle and other pathways) were also calculated and are provided in the Supplementary Material (Table S1) for exploratory reference. We identified regions with significant differences in intergroup comparisons and extracted the relevant feature values for brain function. We then calculated the Pearson correlation coefficient to assess the relationship between these feature values, representing cortical and muscular functional connectivity in the significant regions and upper-limb motor function indicators. This analysis aimed to identify the factors most influential on cortical function. Results Screening took place from March 2022 to January 2023, involving 568 stroke patients, of whom 84 met the eligibility criteria. Participants were randomly assigned to two groups: 42 to a Tai Ji group and 42 to a control group. Fourteen participants (8 from the Tai Ji group and 6 from the control group) withdrew due to incomplete intervention or refusal to undergo a second EEG assessment. Thus, the study concluded with 70 participants: 34 in the Tai Ji group and 36 in the control group, as depicted in Fig. 2. Study flow diagram Participant characteristics The results indicated no significant differences in demographic characteristics between the two groups (see Table 1). Table 1Baseline characteristics of participantsTai Ji Group(n = 34)Control Group(n = 36)T/Z/x2P-valueAge(years)52.32 ± 15.1156.36 ± 13.18-1.1930.237Onset time(months)2.50(1.00,6.00)6.00(1.25,11.00)-1.8830.067Gender(male/female)(n)25/922/140.7240.395Hemisphere side(left/right)(n)13/2114/22<0.0011.000Type of morbidity (hemorrhage/infarction)(n)14/2016/200.0010.972SD is the standard deviation; IQR is the interquartile rangeData are presented as mean ± SD or median [IQR] Baseline characteristics of participants SD is the standard deviation; IQR is the interquartile range Data are presented as mean ± SD or median [IQR] Differences in clinical assessment outcomes Before the intervention, we conducted intergroup comparisons of FMA-UE dimensions, including upper limb, wrist, hand, and coordination, as well as MBI. The results indicated no significant differences between the two groups (P > 0.05). After the 4-week intervention, within-group comparisons revealed significant improvements for the Tai Ji group across all FMA-UE and MBI scores dimensions. Specifically, the Tai Ji group demonstrated substantial enhancements on the FMA-UE scale (P < 0.001), including the FMA upper limb (P < 0.001), FMA wrist (P = 0.002), FMA hand (P = 0.001), and FMA coordination (P < 0.001), as well as in the MBI (P < 0.001). The control group also exhibited significant improvements in the FMA-UE scale (P < 0.001), FMA upper limb (P = 0.001), FMA hand (P = 0.003), FMA coordination (P < 0.001), and MBI (P < 0.001); however, no significant difference was observed in the FMA wrist (P > 0.05). Additionally, comparisons of the changes in scores from pre- to post-intervention showed that the Tai Ji group achieved significantly more significant improvements in the FMA-UE change scores (P < 0.001), FMA upper limb change scores (P < 0.001), FMA wrist change scores (P = 0.002), FMA coordination change scores (P = 0.020), and MBI change scores (P = 0.004) compared to the control group. Notably, the change scores for the FMA hand did not show a statistically significant difference (P > 0.05). Detailed results are presented in Table 2. Table 2Clinical assessment results of upper limb motor function and activities of daily livingOutcome measuresGroupBaselineAfter interventionWithin-groupP-valueBaseline between groupP-valueChanges between groupP-valueUpper Limb Motor Function FMA-UETai Ji Group43.50(17.00,52.25)55.00(27.25,64.25)<0.0010.359<0.001Control Group42.50(21.00,54.75)45.00(24.25,58.00)<0.001 FMA upper limbTai Ji Group27.00(16.00,34.00)33.00(23.00,36.00)<0.0010.719<0.001Control Group27.00(18.25,33.00)28.50(21.00,33.00)0.001 FMA wristTai Ji Group5.50(0.00,6.00)6.50(0.00,10.00)0.0020.2490.002Control Group6.00(1.00,9.00)6.00(1.00,9.00)0.180 FMA handTai Ji Group7.00(1.75,10.00)9.00(2.00,14.00)0.0010.3040.174Control Group9.00(3.25,11.00)9.00(4.25,13.00)0.003 FMA coordinationTai Ji Group0.00(0.00,1.00)3.00(0.00,6.00)<0.0010.2260.020Control Group0.50(0.00,3.00)2.00(0.00,4.00)<0.001ADL MBITai Ji Group80.50(62.00,95.00)93.00(76.00,100.00)<0.0010.5180.004Control Group72.00(56.00,90.50)76.50(63.75,94.75)<0.001SD is the standard deviation; IQR is the interquartile rangeData are presented as mean ± SD or median [IQR]FMA-UE: Fugl-Meyer Assessment for the Upper Extremity; ADL: Activities of Daily Living; MBI: Modified Barthel Index Clinical assessment results of upper limb motor function and activities of daily living SD is the standard deviation; IQR is the interquartile range Data are presented as mean ± SD or median [IQR] FMA-UE: Fugl-Meyer Assessment for the Upper Extremity; ADL: Activities of Daily Living; MBI: Modified Barthel Index Differences in cortico-muscular reorganization outcomes When comparing pre and post-intervention assessments, significant differences were observed in cortical descending functional connectivity between the Tai Ji training and control groups. Notably, this difference was reflected in the values of cortical descending PTE, with the Tai Ji group demonstrating marked alterations in the visual connectivity maps. Specifically, the PTE values for the left motor area and the MD (T = 2.053, P = 0.044), as well as for the RM and the TRI (T = 2.079, P = 0.041), were significantly higher in the Tai Ji group compared to the control group. Moreover, significant differences were identified in the ascending pathways of PTE values related to Tai Ji training versus the control group, which were also visually represented in the connectivity maps. Functional connectivity analysis revealed that the PTE value between the PD in the Tai Ji group and the LS was significantly higher than that of the control group (T = 2.188, P = 0.032). To fully present the analytical results, we have compiled the strength of corticomuscular connectivity across all computed directions (including motor cortex → muscle, muscle → sensory cortex, sensory cortex → muscle, and muscle → motor cortex) in Supplementary Table S2. Exploratory analysis of the full connectivity matrix For completeness, the full directional corticomuscular connectivity matrix, comprising all pairwise connections (motor cortex→muscle, muscle→sensory cortex, sensory cortex→muscle, and muscle→motor cortex), is provided in Supplementary Table S1. Within this comprehensive dataset, statistically significant connections were observed in select pathways, notably including those from the sensory cortex to muscles. These exploratory findings extend beyond the primary hypotheses of the current study. Exploratory correlation between corticomuscular connectivity and motor improvement An exploratory correlation analysis was conducted to examine potential associations between changes in corticomuscular connectivity (for pathways showing significant between-group differences) and improvements in motor function (FMA total and subscores). The complete matrix of Pearson correlation coefficients, P-values, and descriptive statistics for all tested pairings is provided in Supplementary Table S2. Within this exploratory dataset, several correlations reached a nominal P-value threshold of 0.05. Notably, two findings were conceptually aligned with the study’s theoretical framework: a negative correlation was observed between the enhancement of the PD-LS pathway and improvement in hand function (FMA-hand: r = -0.351, P = 0.042), and between the enhancement of the RM-TRI pathway and improvement in wrist function (FMA-wrist: r = -0.350, P = 0.042). These correlational findings are preliminary and are presented here as hypothesis-generating observations. Their primary value lies in identifying specific pathways (e.g., PD-LS, RM-TRI) for future, dedicated studies to confirm their role in motor recovery (Fig. 3). Cortical-Muscle Coupling. A. Changes in Downstream Pathway Coupling Strength. The blue and red lines represent the functional connection of the cortex to the muscle. B. Changes in Upstream Pathway Coupling Strength. The red lines represent the functional connections of the muscles to the cortex. C. Correlation Analysis. LM - Left Motor Cortex; RM - Right Motor Cortex; LS - Left Sensory Cortex; RS - Right Sensory Cortex; PM - Pectoralis Major; UT - Upper Trapezius; AD - Anterior Deltoid; MD - Middle Deltoid; PD - Posterior Deltoid; BIC - Biceps Brachii; TRI - Triceps Brachii; FCR - Flexor Carpi Radialis; ECR - Extensor Carpi Radialis Muscle synergy structures and activation profiles The quantitative outcomes of the muscle synergy analysis are summarized in Table 3. Critically, the between-group comparison of improvement (post-intervention minus baseline) revealed no statistically significant difference for any synergy metric (H1, H2, W1, W2; all P > 0.05). Both the Tai Ji and control groups demonstrated significant within-group improvements from baseline to post-intervention across all metrics (all P < 0.001). Baseline comparisons confirmed no initial between-group differences (all P > 0.05), ensuring comparable starting points. Table 3Changes in muscle synergy metrics before and after interventionOutcome measuresGroupBaselineAfter interventionWithin-groupP-valueBaseline between groupP-valueChanges between groupP-valueH1Tai Ji Group0.17 ± 0.170.43 ± 0.20<0.0010.0600.078Control Group0.23 ± 0.160.39 ± 0.19<0.001H2Tai Ji Group0.17 ± 0.140.40 ± 0.18<0.0010.3240.205Control Group0.21 ± 0.140.37 ± 0.18<0.001W1Tai Ji Group0.25 ± 0.260.63 ± 0.29<0.0010.9200.481Control Group0.18 ± 0.120.52 ± 0.33<0.001W2Tai Ji Group0.31 ± 0.230.63 ± 0.20<0.0010.7510.078Control Group0.29 ± 0.190.50 ± 0.23<0.001The extracted muscle synergy metrics include: H1 and H2, the activation timing profiles (coefficients) for Synergy 1 and Synergy 2, respectively; W1 and W2, the muscle weighting vectors (patterns) for Synergy 1 and Synergy 2, respectively Changes in muscle synergy metrics before and after intervention The extracted muscle synergy metrics include: H1 and H2, the activation timing profiles (coefficients) for Synergy 1 and Synergy 2, respectively; W1 and W2, the muscle weighting vectors (patterns) for Synergy 1 and Synergy 2, respectively The corresponding muscle synergy patterns are visually illustrated in Fig. 4, which compares the Tai Ji and control groups before and after intervention against a reference healthy group. We analyzed sEMG signals from nine muscles (ECR, FCR, TRI, BIC, PD, MD, AD, UT, and PM) to develop synergy models. To retain essential information while ensuring clarity, we present the two-dimensional synergy models, which accounted for more than 80% of the variance (VAF > 80%). In the figure, blue and red represent Synergy 1 and Synergy 2, respectively. For each synergy, the left panel displays the muscle weight vector (W), and the right panel shows the corresponding activation profile (H). Muscle Synergy Structure. PM - Pectoralis Major, UT - Upper Trapezius, AD - Anterior Deltoid, MD - Middle Deltoid, PD - Posterior Deltoid, BIC - Biceps Brachii, TRI - Triceps Brachii, FCR - Flexor Carpi Radialis, ECR - Extensor Carpi Radialis In the healthy group, Synergy 1 was characterized by a dominant weight in the TRI, reflecting agonist activation during movement acceleration, whereas Synergy 2 showed a dominant weight in the BIC, representing antagonist activation during deceleration. Qualitatively, the post-intervention synergy patterns of the Tai Ji group appeared to show a shift in the activation profiles (H) relative to baseline. However, this visual trend must be interpreted within the context of the quantitative results (Table 3), which indicate no statistically superior improvement in the Tai Ji group. Discussion This four-week randomized controlled trial demonstrates that integrating adjunctive Tai Ji training into standard rehabilitation significantly improves upper limb motor function (FMA-UE) in stroke patients. Beyond confirming the benefits of this combined approach, our integrated neurophysiological analyses reveal a dual-mechanism adaptation: the Tai Ji-supplemented program simultaneously strengthens bidirectional fCMC and normalizes pathological muscle synergy patterns. This combination of enhanced neural communication and refined motor execution underpins the observed functional gains. Furthermore, the absence of adverse events underscores the safety and practical feasibility of this mind-body practice as an adjunct therapy. Tai Ji modulates bidirectional corticomuscular connectivity Enhancement of descending cortical drive to muscles Our central finding is that participants who received the Tai Ji-supplemented training showed a specific strengthening of the directed functional connectivity from the contralateral motor cortex to key proximal upper limb muscles, namely the MD and TRI. This suggests that the adjunctive Tai Ji practice may effectively reinforce the descending neural commands essential for voluntary movement, which are characteristically impaired following stroke due to disrupted corticomuscular pathways [15, 16]. The observed increase in this efferent drive likely constitutes a primary neurophysiological correlate for the concurrent improvement in FMA-UE scores, linking the supplemented intervention package to functional recovery at a systems level. Specifically, the enhanced drive to these prime movers may facilitate more normalized recruitment during reaching tasks, thereby reducing compensatory synergistic patterns. This enhancement may be linked to Tai Ji’s fundamental principle of “intention leads form,” where focused mental attention precedes and guides deliberate physical movement—a core component of mind-body practices supported by contemporary frameworks [38, 39]. This principle of mindful movement guidance is thought to facilitate motor learning and strengthen corticospinal projections through mechanisms such as Hebbian plasticity. Our finding—that the Tai Ji program was associated with enhancement of the critical pathways rehabilitation aims to restore—is consistent with evidence that such training can promote adaptive neuroplasticity and functional optimization of the motor network, which may involve a more efficient integration of bilateral resources [20]. Furthermore, the use of PTE in our analysis isolates direct causal influences, strengthening the interpretation that the adjunctive Tai Ji training directly contributed to enhancing cortical control over specific muscles. This finding resonates with studies on other targeted therapies, such as neuromusic therapy, which also report modulating cortical-muscle connectivity to improve motor outcomes [21]. Collectively, the specific enhancement of the corticomotor drive reported here, when integrated with concurrent normalization of pathological muscle synergies and the significant improvement in motor function, indicates that Tai Ji-supplemented rehabilitation facilitates a positive, functionally oriented reorganization of the motor system. This process involves the adaptive optimization of the entire sensorimotor network—including a more efficient integration of bilateral resources—rather than a maladaptive compensatory substitution. Thus, this mind-body practice, as part of a comprehensive program, promotes a recovery strategy that maximizes functional outcomes by optimally engaging both restorative and adaptive neuroplastic mechanisms. Facilitation of ascending sensory feedback to the cortex Concurrently, we observed an increase in directed connectivity from the PD muscle back to the ipsilateral sensory cortex following the Tai Ji-supplemented training. This finding highlights the potential role of the adjunctive Tai Ji practice in enhancing proprioceptive feedback, which is essential for refining movement accuracy and inter-joint coordination during complex tasks [31, 32]. The specific enhancement of the PD—a key muscle for shoulder stabilization—suggests that Tai Ji training may fine-tune the sensory representation of shoulder kinematics, which is often impaired post-stroke and vital for limb control. This enhanced sensory feedback is likely driven by Tai Ji’s emphasis on precise posture, slow movement, and heightened bodily awareness, which amplifies proprioceptive engagement during practice. Such focused, continuous afferent input during practice is hypothesized to strengthen the feedback to the cortex, aiding in the recalibration of sensorimotor circuits. This mechanism establishes a virtuous cycle: improved descending commands generate more coordinated movements, which in turn produce richer sensory feedback, further refining motor output and promoting integrated sensorimotor recovery [33, 40]. This closed-loop facilitation may contribute to the observed gains in movement smoothness, intermuscular coordination, and overall motor control stability. Considerations on exploratory corticomuscular connectivity findings Exploratory analyses suggest that the remodeling of corticomuscular information flow following the Tai Ji-supplemented training may be pathway-specific. This is reflected in a trend toward enhanced connectivity from sensory cortices to specific muscles (e.g., LS-AD, RS-MD), while changes in other pathways (e.g., muscle-to-motor cortex) were not significant. This finding may reflect adaptive compensation within the sensorimotor integration pathways post-stroke. According to sensorimotor integration theory, the brain relies on a “forward model” to predict the sensory consequences of movement [41]. Following damage to descending motor pathways, this feedforward predictive mechanism, potentially anchored in sensory cortices, may be upregulated to optimize motor control. Concurrently, sensory cortices themselves are recognized as critical hubs for functional reorganization post-stroke, and their network connectivity status is closely linked to motor recovery [42, 43]. Therefore, the observed enhancement in “sensory cortex-to-muscle” connectivity following the intervention may embody dual implications: the upregulation of a predictive mechanism and compensatory engagement of sensory cortices. It must be emphasized that the interpretations above are preliminary and speculative. These exploratory findings offer a novel, hypothesis-generating direction for understanding post-stroke brain network remodeling, suggesting that the Tai Ji-supplemented program may preferentially facilitate regulatory pathways from sensory cortices to motor output. Considerations on exploratory correlational findings Exploratory analyses revealed negative correlations between motor improvement and the enhancement of specific pathways: hand function with the PD-LS pathway, and wrist function with the RM-TRI pathway. While seemingly counterintuitive, these findings may tentatively reflect processes of neural efficiency or a shift in compensatory strategy. For instance, as sensorimotor control refines with training, dependence on overt feedback loops may diminish, potentially reducing measured connectivity despite better performance. These interpretations remain speculative, derived from hypothesis-generating analyses. The correlations require validation in future pre-registered studies but highlight PD-LS and RM-TRI as candidate pathways for understanding recovery mechanisms. Tai Ji promotes normalization of pathological muscle coordination sEMG analysis revealed that, at baseline, our stroke cohort exhibited aberrant muscle synergies commonly reported post-stroke [23, 24]. This pathological pattern was characterized by abnormal co-activation of the trapezius and deltoid muscles (associated with compensatory shoulder girdle elevation) and markedly reduced activation of the TRI—a key agonist for elbow extension [22, 26]. Following the Tai Ji-supplemented training, we observed a shift in muscle coordination patterns. Qualitatively, the post-intervention synergy profiles of the Tai Ji group appeared to move toward the more selective activation patterns seen in the healthy reference group. Quantitatively, both the Tai Ji and control groups showed significant within-group improvements in all synergy metrics. The lack of a statistically significant between-group difference in the extent of change suggests that the control intervention also induced considerable neuromuscular adaptation. However, the observed visual trend toward normalization in the Tai Ji group warrants mechanistic consideration. We propose that the Tai Ji program may facilitate this qualitative shift through its unique movement characteristics: First, Its slow, controlled motions are ideal for training eccentric contractions and refining coordination between agonist and antagonist muscles. Second, Specific postural directives, such as “sinking the shoulders and dropping the elbows,” directly target the abnormal shoulder elevation synergy. Third, and distinctively, the integration of movement with breath and mindful relaxation—a core mind-body component of Tai Ji [38]—may create an internal state conducive to the motor learning required for de-linking pathological muscle synergies. Tai Ji’s holistic emphasis on postural alignment, slow and controlled motion, and mindful integration of movement fosters refined neuromuscular control. This approach is characterized by balanced coordination between prime movers and stabilizers and is associated with improved movement quality and joint safety [44–46]. This mechanistic framework is supported by the established feasibility of adapted Tai Ji in stroke rehabilitation [47] and its alignment with core neurorehabilitative principles [2]. Integration of mechanisms and clinical implications The observed enhancements in fCMC and muscle coordination displayed notable specificity, primarily involving proximal muscles such as the deltoid and TRI, despite the broader functional improvements captured by the FMA-UE. This apparent specificity can be explained by a convergent, multi-level mechanism that likely forms the foundation for the general recovery. First, the adapted Tai Ji forms specifically emphasized shoulder and elbow movements, directly targeting the prime movers most critical for reaching and most impaired by post-stroke pathological synergies [39]. Second, neuroplastic recovery may follow a hierarchical principle. This concept is supported by evidence of hierarchical abnormalities in neuromuscular control post-stroke [48] and is discussed in the context of hierarchical maturational processes during recovery [49]. Within this framework, the fundamental corticospinal pathways to key proximal agonists may be among the most responsive to task-oriented interventions, thereby acting as primary drivers for overall functional gain. Finally, detectable fCMC itself relies on intact, rapid sensorimotor feedback loops [50]; the analysis is thus most sensitive to changes in muscles with well-defined cortical representations and efficient circuits, such as the superficial proximal muscles studied. Therefore, the selective increase in fCMC following the Tai Ji-supplemented training reflects the preferential recalibration of the most critical and compromised sensorimotor loops, which, in turn, provides a stable neural platform for the observed broader functional gains. The mindful practice of Tai Ji is ideally suited to optimize these fundamental circuits. The exploratory correlation findings, while preliminary, add a nuanced layer to this model. They suggest that within this overarching framework of enhanced bidirectional connectivity, the degree of modulation in specific, training-engaged pathways (like PD-LS and RM-TRI) may relate to improvements in specific motor domains (hand, wrist), albeit in a complex manner that future work must clarify. Collectively, our findings support a dual-process model of motor recovery associated with the Tai Ji-supplemented program. The intervention simultaneously targets both the neural command level—by strengthening the bidirectional, closed-loop dialogue between the cortex and muscles—and the peripheral execution level—by normalizing pathological muscle synergies. These processes are likely synergistic: enhanced neural drive facilitates more appropriate muscle activation, while refined coordination provides cleaner sensory feedback, further optimizing cortical commands. This integrated mechanism, captured by our combined EEG-sEMG and clinical metrics, offers a neurophysiological explanation for the significant functional gains (FMA-UE) observed. It underscores Tai Ji’s value as a holistic, mind-body intervention that addresses the multi-level complexity of post-stroke motor impairment, informing the design of comprehensive rehabilitation strategies. Limitations and future directions This study has several limitations that should be considered when interpreting the results. First, and most importantly, the total intervention time was unequal between groups. The Tai Ji group received additional supervised training. Consequently, the observed improvements likely reflect a combination of Tai Ji’s specific effects and the non-specific benefits of increased training dose. This design, while pragmatic for assessing Tai Ji as an adjunct therapy, means that the large between-group effect size for FMA-UE improvement (Cohen’s d ≈ 1.84) incorporates both components and may overestimate the unique effect of Tai Ji per se. Therefore, future hypothesis-driven trials using a dose-matched active control should adopt a more conservative effect size (e.g., d = 0.4–0.5) for sample size calculation, which would require approximately 64 to 100 participants per group (α = 0.05, power = 80%). Second, the four-week intervention duration is relatively short. Longer-term trials are needed to evaluate the sustainability of benefits and their translation into daily functioning. Third, the absence of structural neuroimaging data limits our ability to link the observed functional connectivity changes to underlying brain structure (e.g., corticospinal tract integrity). Finally, as an initial mechanistic investigation, our findings require replication in broader and more diverse populations to establish generalizability. Future research should build upon these findings by: (1) conducting large-scale, dose-matched randomized trials to isolate Tai Ji’s unique benefits and explore predictors of response; (2) integrating multimodal neuroimaging (e.g., diffusion tensor imaging) with electrophysiology to delineate structural-functional plasticity; (3) implementing longer-term studies to assess sustained functional gains; (4) validating the exploratory neurophysiological findings (e.g., involving the PD-to-sensory cortex pathway) in independent cohorts; and (5) investigating the cross-cultural applicability of these mechanisms. Conclusions This study demonstrates that integrating a 4-week Tai Ji program into standard rehabilitation is an effective and well-tolerated approach associated with improved upper-limb motor function after stroke. Mechanistically, these benefits were linked to a dual-process reorganization: primarily characterized by strengthening the contralateral descending pathway, while concomitantly optimizing the function of the entire sensorimotor network. This adaptive optimization is evidenced by enhanced ascending sensory feedback, normalized muscle synergies, and the reinforcement of specific corticomotor drive. Exploratory analyses further suggested that the modulation of specific corticomuscular pathways might relate to functional gains, offering testable hypotheses for future research. Collectively, these shifts—encompassing both restorative and adaptive plasticity—provide preliminary support for integrating Tai Ji as a promising adjunct strategy within comprehensive post-stroke rehabilitation programs to promote comprehensive, functionally-adaptive motor recovery. Electronic Supplementary Material Below is the link to the electronic supplementary material. Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Acknowledgements The authors thank the subjects from the Affiliated Rehabilitation Hospital of Fujian University of Traditional Chinese Medicine for participating in the experiment. Author contributions Xie Qiurong, Ph.D., writing - original draft, methodology, formal analysis, project administration, funding acquisition. Li Yurong, Ph.D., writing - review & editing, funding acquisition, methodology, formal analysis.Zheng Nan, M.S., writing - review & editing, funding acquisition, methodology, formal analysis.Zhang Qi, Ph.D., writing - review & editing, funding acquisition.Wang Xiaoling, Ph.D., writing - review & editing, funding acquisition.Luo Jia, Ph.D., writing - review & editing.Zhang Yanxin, Ph.D., writing - review & editing, supervision.Tao Jing, Ph.D., writing - review & editing, funding acquisition, conceptualization.Huang Jia, Ph.D., writing - review & editing, funding acquisition. Funding This study was supported by grants form the National Natural Science Foundation of China (Grant Numbers 82305357 and 62373108 ) and the Natural Science Foundation of Fujian Province (Grant Number 2025J01948). Data availability Data related to the paper are not publicly available but can be obtained from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the Helsinki declaration and its later amendments or comparable ethical standards. Ethical approval was granted by the Ethics Committee of the Rehabilitation Hospital Affiliated with Fujian University of Traditional Chinese Medicine (Ethics Approval No.: 2021KY-005-02). Informed consent was obtained from all individual participants included in the study. Consent for publication All participants provided consent for the publication of their anonymized data. 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. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Data Availability Statement Data related to the paper are not publicly available but can be obtained from the corresponding author upon reasonable request.