Temporal synchrony promotes motor recovery and ipsilesional cortical reorganization in BCI-based stroke rehabilitation: a randomized controlled trial
Clinical Question (PICO)
In selected adults with a first stroke 1 week to 6 months earlier, substantial arm weakness, preserved cognition, and adequate motor-imagery ability, does a soft robotic glove triggered by a real-time EEG engagement signal improve 3-week upper-limb recovery compared with the same motor-imagery training plus noncontingent glove movements, or sham-BCI plus passive glove movements? All groups also received the same conventional rehabilitation program. The primary endpoint was the Fugl-Meyer Assessment–Upper Extremity (FMA-UE); arm activity and daily independence were secondary outcomes. No prespecified clinical subgroup-effect estimates are reported in the main article.
Bottom Line
Synchronous, signal-contingent training was associated with better short-term motor and daily-function outcomes: adjusted FMA-UE differences were 7.47 points versus sham (95% CI 1.94–13.00; P = 0.004) and 9.72 versus asynchronous training (95% CI 4.14–15.30; P < 0.001). This small, single-center trial supports further testing of feedback contingency—not a proven cortical-reorganization mechanism or a durable treatment benefit; none of the imaging group comparisons survived false-discovery-rate correction. See the primary report's clinical results and exploratory imaging analyses.
Design
- Trial type: Three-arm, parallel randomized controlled trial; outcome assessor blinded, participants and therapists unblinded.
- N randomized: 77.
- Group A, sham-BCI plus passive glove: 26 randomized; 19 analyzed.
- Group B, real-BCI plus asynchronous glove: 25 randomized; 24 analyzed.
- Group C, real-BCI plus contingent glove: 26 randomized; 25 analyzed.
- Randomization: Computer-generated 1:1:1 allocation; sequentially numbered, sealed opaque envelopes prepared independently. No stratification, blocking, or minimization.
- Setting: Xuzhou Rehabilitation Hospital, China; one center.
- Enrollment: November 2024–April 2025. Published July 22, 2026; this is a look-back appraisal, not a new October publication.
- Follow-up: Assessment immediately after 3 weeks; no longer-term follow-up.
- Analysis: Available-case primary analysis; baseline-adjusted ANCOVA with Bonferroni-adjusted pairwise comparisons. Multiple-imputation sensitivity analysis assumed missing-at-random outcomes; the authors report a similar pattern of results.
- Primary outcome: Postintervention FMA-UE, adjusted for baseline; scale 0–66, higher indicating less motor impairment.
- Registration: ChiCTR2400091197, registered October 21, 2024. The implemented sham-device control differed from the public registry's conventional-rehabilitation-only control; the authors state that ethics approval preceded recruitment, but the registry was not updated.
Population
Inclusion Criteria
- Right-handed adults aged 18–80 years with a first stroke, stable vital signs, and onset 1 week to 6 months earlier.
- Stated imaging criterion: unilateral subcortical lesion on CT/MRI.
- Arm dysfunction at Brunnstrom stages 1–5; MMSE >20; Kinesthetic and Visual Imagery Questionnaire >55.
- No previous neurosurgical intervention or other psychiatric/neurological disease; participant and guardian consent.
Exclusion Criteria
- Other neurological conditions, recurrent stroke, or medical instability.
- Skull/scalp abnormalities interfering with EEG/fNIRS, active scalp infection, or inability to tolerate procedures.
- Severe spasticity (Modified Ashworth Scale ≥3), relevant contracture, or antispasmodic treatment during the preceding 6 months.
- Concurrent central-nervous-system-targeting interventions.
Baseline Characteristics (randomized groups A / B / C)
- Mean age: 58.58 / 50.08 / 51.77 years; men: 18/26 / 18/25 / 19/26.
- Median time since stroke: 5.50 / 8.00 / 10.00 weeks.
- Ischemic stroke: 6 / 10 / 11; hemorrhagic stroke: 20 / 15 / 15. The sample therefore includes substantial hemorrhagic-stroke representation.
- Median FMA-UE: 11.50 / 17.00 / 18.00; median ARAT: 0 in all three groups.
- Mean Modified Barthel Index (MBI): 36.15 / 41.60 / 35.38.
- The lesion-location table also lists cortical and cerebellar involvement, which is not readily reconciled with the stated subcortical-only entry criterion. Multiple locations per participant were permitted in that table.
Interventions
All groups trained 5 days weekly for 3 weeks. Daily conventional rehabilitation comprised 50 minutes of physical therapy, 30 minutes of occupational therapy, and 30 minutes of acupuncture/moxibustion, followed by a 20-minute BCI/device session. The experimental component therefore scheduled 15 sessions, or 300 minutes in total.
- Group A—sham/passive: Motor-imagery instructions with prerecorded feedback; glove performed a preset grip-and-open cycle every 20 seconds, unrelated to ongoing brain activity.
- Group B—asynchronous: Real-time EEG engagement drove visual success feedback, but the glove still cycled every 20 seconds independently of that signal.
- Group C—contingent: Both visual success feedback and glove movement were triggered when the real-time EEG-derived score crossed an adaptive threshold during instructed motor imagery.
The EEG controller used a normalized beta/alpha power ratio as an engagement score, not a demonstrated hand-intention-specific decoder. Features were updated every 0.1 seconds from 2-second windows. Successful-trial timing varied, and the number of trials per 20-minute session was not fixed. Comparable session duration does not establish an identical number of successful movements or sensory events. These details matter when interpreting the implemented intervention and controller.
Outcomes
Primary Outcome—baseline-adjusted postintervention FMA-UE:
| Comparison | Adjusted mean difference, points | 95% CI | Reported P value |
|---|---|---|---|
| Contingent C versus sham A | 7.47 | 1.94 to 13.00 | 0.004 |
| Contingent C versus asynchronous B | 9.72 | 4.14 to 15.30 | <0.001 |
The overall ANCOVA group effect was F = 10.067, P < 0.001, partial eta squared = 0.239. Unadjusted mean FMA-UE changes were 7.88, 5.88, and 15.50 points in A, B, and C, respectively. These are mean differences on a continuous scale—not odds ratios, relative risks, or responder rates. Pairwise comparisons were performed after the overall group test; they are not patient-subgroup analyses. No prespecified subgroup-specific HR/OR/RR estimates were available to report.
Secondary Outcomes:
| Outcome and comparison | Adjusted mean difference, points | 95% CI | Reported P value |
|---|---|---|---|
| ARAT, C versus A | 6.18 | −0.02 to 12.38 | 0.051 |
| ARAT, C versus B | 6.64 | 0.38 to 12.90 | 0.034 |
| MBI, C versus A | 21.15 | 12.49 to 29.81 | <0.001 |
| MBI, C versus B | 21.27 | 12.42 to 30.12 | <0.001 |
ARAT measures arm activity on a 0–57 scale; MBI measures daily independence on a 0–100 scale. The ARAT comparison with sham did not meet the reported significance threshold. The numerical estimates above come from the clinical outcome results in the primary article, rather than an inference from within-group improvement.
Exploratory Neuroimaging:
- Usable fNIRS data: 65 participants, A/B/C = 19/23/23.
- No resting-state or task-related between-group imaging comparison remained significant after false-discovery-rate correction.
- A separate exploratory correlation between change in sensorimotor laterality during motor imagery and FMA-UE improvement remained significant: Spearman rho = 0.404, P = 0.002, q = 0.012. Association does not establish treatment mediation or direction of causation.
Adverse Events / Safety:
Nine participants withdrew: seven in A, one in B, and one in C. Five A withdrawals and the B withdrawal were attributed to lack of perceived benefit; other A reasons were scheduling and transfer. The C withdrawal followed an unrelated ward fall with scalp abrasion. The authors report no intervention-related serious adverse events and no withdrawals attributed to device adverse events or intolerance. This small, short study does not establish rare-event safety, falls prevention, mortality benefit, or recurrent-stroke protection.
Figures
Source: Wu et al., 2026, PMC13678967 — 12984_2026_2100_Fig1_HTML.webp. Original image reproduced unchanged under CC BY-NC-ND 4.0; caption paraphrased. Click image to expand.
Source: Wu et al., 2026, PMC13678967 — 12984_2026_2100_Fig2_HTML.webp. Original image reproduced unchanged under CC BY-NC-ND 4.0; caption paraphrased. Click image to expand.
Source: Wu et al., 2026, PMC13678967 — 12984_2026_2100_Fig3_HTML.webp. Original image reproduced unchanged under CC BY-NC-ND 4.0; caption paraphrased. Click image to expand.
Source: Wu et al., 2026, PMC13678967 — 12984_2026_2100_Fig4_HTML.webp. Original image reproduced unchanged under CC BY-NC-ND 4.0; caption paraphrased. Click image to expand.
Criticisms
- Differential attrition: 7/26 in sham versus 1/25 and 1/26 in the active arms. Perceived lack of benefit drove several withdrawals. Imputation under a missing-at-random assumption cannot, by itself, dismiss outcome-dependent dropout in an unblinded device trial.
- Short horizon and restricted sample: Only 3-week outcomes; good imagery ability and MMSE >20 were required. Results do not establish effectiveness in severe cognitive impairment, poor imagery ability, later chronic stroke, or routine unselected stroke populations.
- Contingency is not isolated perfectly: Duration was standardized, but movement counts, success rates, expectancy, and reward experience could differ. Without matched event counts and recorded signal-to-movement latency, the experiment does not identify a precise therapeutic timing window.
- Mechanism exceeds the strongest evidence: The title invokes cortical reorganization, while corrected imaging group comparisons were null. The surviving exploratory correlation is not evidence that reorganization caused the treatment effect. fNIRS also lacked short-separation channels and had a limited cortical montage.
- Registration and reporting: The public control description was not updated. The stated subcortical eligibility criterion and the reported lesion categories need clarification. Table 3 repeats baseline summaries alongside smaller analyzed-group denominators; participant-level data would help resolve the analysis populations.
- No established responder threshold here: The FMA-UE comparison with sham has a lower confidence limit of only 1.94 points, and ARAT versus sham was uncertain. A statistically positive impairment endpoint is not proof of meaningful daily hand use for every patient.
Funding
Supported by the Xuzhou Key Medical Talents Project (XWRCHT20220045), the Science and Technology Program of Xuzhou (KC25081), and the Pengcheng Talent Program's high-level healthcare talent project (2025TD14). The funding statement does not explicitly describe the funders' role in design, analysis, or publication. The authors declare no competing interests. Registration: ChiCTR2400091197. Figures are reproduced without alteration from the source under its CC BY-NC-ND 4.0 license; accompanying descriptions are editorial paraphrases.
The paper
- Authors. Yu Wu et al.; Qiumeng Dong, Jie Chen, Cong Huang, and Ming Zhang.
- Title. Temporal synchrony promotes motor recovery and ipsilesional cortical reorganization in BCI-based stroke rehabilitation: a randomized controlled trial.
- Journal. Journal of NeuroEngineering and Rehabilitation.
- Year. 2026; published July 22.
- DOI. 10.1186/s12984-026-02100-9
- PMCID. PMC13678967
- PMID. 42894183
Deep Dive — click to expand
What this is
This July 2026 rehabilitation trial asks whether the relationship between a brain signal and a robot's movement matters more than merely having both technologies in the room. Among 77 randomized patients, 68 supplied the primary post-treatment outcomes; contingent feedback outperformed both asynchronous and sham training on baseline-adjusted FMA-UE. The interesting clinical signal is short-term improvement, not confirmation of the cortical mechanism advertised in the title. The commentary below separates the paper's observations from editorial implementation proposals.
1. Shadow Audit
The strongest mechanistic sentence is the negative one: no imaging group comparison survived multiplicity correction. A correlation of rho = 0.404 with q = 0.012 is worth following, but cannot rescue a causal cortical-reorganization claim. Nor does a 21-point adjusted MBI advantage establish the same degree of hand-specific functional recovery: ARAT versus sham was 6.18 points with a confidence interval crossing zero (−0.02 to 12.38).
The other shadow is the control experience. Seven sham participants withdrew, including five who perceived little benefit. That does not prove the benefit is expectation bias—dropout can distort results in more than one direction—but it weakens a tidy interpretation of the available-case contrast. The paper's missing-data methods and withdrawal account must travel with its headline effect.
2. Inversion Engine
For a literal reversal of the adjusted FMA-UE advantage, a replication would need a contingent-minus-control difference at or below zero, rather than the observed +7.47 versus sham or +9.72 versus asynchronous training. Those are point-estimate benchmarks, not calculated bias corrections. For the narrower claim that a meaningful advantage is already secure, the hurdle is lower: the sham comparison's lower confidence limit is only +1.94, and ARAT versus sham remains inconclusive.
A proper inversion test would preserve blinded assessment, match delivered glove movements and rewards, record end-to-end latency, retain participants for follow-up, and analyze all randomized participants with sensitivity analyses that permit informative dropout. If the advantage disappeared under those conditions, the explanation could shift from temporal coupling toward expectancy, movement dose, or attrition. The current data cannot quantify how much each contributes.
3. Second-Order Catalyst
The first reasonable adopter is a rehabilitation research service already operating EEG-guided robotic training, not an acute stroke unit purchasing a new system on this paper alone. Its near-term change would be an engineering and measurement audit: log threshold crossings, actual glove activation, missed triggers, movement counts, and post-session fatigue; then prospectively compare contingent and noncontingent delivery with blinded clinical outcomes.
The initial research population should resemble those studied—subacute stroke, substantial paresis, sufficient cognition and imagery ability—while documenting ischemic versus hemorrhagic stroke and time since onset. A feasibility audit can begin promptly; claiming a routine standard of care should wait for independent replication and longer follow-up. That is an editorial implementation proposal, not a guideline statement.
4. Asymmetric Leverage
The asymmetric opportunity is not more hardware exposure: all groups already had a glove, motor-imagery instructions, and 20-minute sessions. It is potentially making an existing session more useful by changing its control logic. The experimental schedule amounted to 300 minutes over 3 weeks; if replication preserves benefit while holding movement count and staffing constant, a software/control improvement could offer value without adding scheduled treatment time.
That remains conditional. The study did not measure cost-effectiveness, staff setup burden, actual matched movement dose, or durability. The relevant denominator is a selected BCI-capable rehabilitation population—not all stroke admissions—and no population-level savings estimate can be derived here.
5. Paradigm Destroyer
The habit challenged is equating a robotic movement with a closed therapeutic loop simply because an EEG cap is present. This experiment makes contingency a variable worth specifying, rather than a marketing adjective.
Two-sentence research-protocol update: "For EEG-assisted robotic rehabilitation, document whether peripheral assistance is contingent on the patient's online signal and record the delivered timing and movement dose. Continue established rehabilitation, and evaluate any contingency-based upgrade with blinded outcomes and longer follow-up rather than treating this small trial as sufficient for routine adoption."
MVP — Minimum Viable Proof
The minimum useful replication is a sustained, baseline-adjusted clinical advantage of contingent over otherwise matched noncontingent training in an all-randomized analysis, accompanied by arm-activity benefit and acceptable harms. Here the starting numerical observation is +9.72 FMA-UE points versus asynchronous training (95% CI 4.14–15.30) at 3 weeks. Independent confirmation beyond the treatment period, with matched movement dose and an interpretable responder endpoint, would materially strengthen the case; another nominal brain-map difference would not.
Best Combination
Combine this result with earlier randomized evidence as a hypothesis about why BCI protocols differ, not proof that every BCI intervention works. In Brunner et al.'s 2024 randomized pilot, a different motor-imagery/FES protocol for severe subacute paresis showed no significant between-group ARAT difference at 3 months (P = 0.382; 35 participants analyzed). Its patient selection, effector, and outcome horizon differ, so it neither directly refutes nor independently confirms this glove-contingency result.
The useful combination is a trial design: preserve rigorous conventional rehabilitation, select and characterize patients carefully, and explicitly test the timing/control ingredient with an active comparator and durable activity outcomes. The present study adds a design clue to a heterogeneous literature; it does not settle treatment choice or supply a current guideline recommendation.
Overvalue Warning
- Do not turn the title into a causal mechanism. Corrected imaging group comparisons were negative; the remaining exploratory correlation cannot prove that ipsilesional reorganization mediated recovery.
- Do not turn a 3-week impairment advantage into universal functional recovery. ARAT versus sham remained uncertain, attrition was unequal, and patients with poor cognition or imagery ability were not represented. Neither long-term independence nor rare-event safety was established.
Educational use notice. The Time Is Brain Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.