Brain-computer interface for post-stroke upper-limb recovery: a systematic review integrating clinical efficacy with neuroplasticity evidence

Abstract

Objective Over 20 systematic reviews have evaluated brain-computer interface (BCI) training for post-stroke upper-limb rehabilitation, but published syntheses have focused predominantly on clinical efficacy, and the neuroplasticity mechanisms remain comparatively underexplored. This review integrates pooled clinical efficacy with a structured synthesis of neuroplasticity evidence from the same trials. Methods PubMed, Embase, CINAHL via EBSCO, Web of Science, Scopus, and Cochrane CENTRAL were searched from inception to May 2026. Randomized controlled trials (RCTs) of BCI-based training for post-stroke upper-limb recovery were eligible. Two parallel syntheses were conducted. A random-effects meta-analysis pooled mean differences for three upper-limb motor outcomes, RoB 2 risk-of-bias assessment, GRADE certainty-of-evidence grading, and publication-bias diagnostics. Neuroplasticity evidence was synthesized qualitatively following the Synthesis Without Meta-analysis (SWiM) guideline. Results Thirty-five RCTs (n = 1,188) were included. BCI training improved Fugl-Meyer Assessment-Upper Extremity [FMA-UE; MD = 4.55, 95% CI (2.84, 6.25), P < 0.001], Action Research Arm Test [ARAT; MD = 3.90, 95% CI (1.31, 6.50), P = 0.003], and Wolf Motor Function Test [WMFT; MD = 8.53, 95% CI (3.01, 14.04), P = 0.002]. Comparator type significantly moderated the effect [Q-between(3) = 13.43, P < 0.01], ranging from MD = 6.70 against usual care to MD = 1.97 against sham-contingent controls; stroke stage, signal strategy, and feedback modality did not. Qualitative synthesis of 22 trials identified multi-level neuroplastic changes spanning corticospinal excitability, sensorimotor rhythm modulation, and network reorganization. A subset of studies reported correlations between neuroplastic changes and motor improvement. Conclusion This systematic review found that BCI-based multimodal rehabilitation improves post-stroke upper-limb motor function, with the estimated effect varying by comparator type. Neuroplasticity outcomes were examined across electrophysiological, hemodynamic, and structural modalities, but the current evidence is too heterogeneous and sparse to support a mechanistic account of recovery. Further research is needed to determine how training-associated neural changes relate to functional improvement. Systematic review registration https://www.crd.york.ac.uk/prospero/view/CRD420251053630, identifier: CRD420251053630.