Skip to content
Log in
#small language model Review

Brain-computer interface training for motor recovery after stroke.

Yu Qin Mei-xuan Li Yanfei Li Min-Chul Ma Jianguo Xu Yaqin Lu Xiue Shi Gecheng Cui Haitong Zhao Kehu Yang
Aug 2026 · Cochrane Database of Systematic Reviews · Vol 8, pp. CD015065 · 0 citations · 94 references
Medicine

Abstract

Rationale

Stroke is one of the leading causes of death and disability worldwide. Motor dysfunction is a highly prevalent and disabling consequence of stroke. Brain-computer interface (BCI) training has emerged as a promising neurorehabilitative strategy that utilises closed-loop feedback to promote targeted neural plasticity and motor recovery. However, current clinical evidence remains fragmented.

Objectives

To assess the effects of brain-computer interface training for motor recovery in people after stroke. SEARCH

Methods

We searched the Cochrane Stroke Group's Specialised Register, CENTRAL, MEDLINE, Embase, 11 other databases, trial registries, reference lists, and Google Scholar up to 27 October 2025, without language or time restrictions. ELIGIBILITY CRITERIA We included randomised controlled trials (RCTs) involving adults with stroke and motor dysfunction. We compared BCI training versus conventional rehabilitation, sham-BCI, or other active non-BCI interventions. OUTCOMES Critical outcomes were motor function (upper and lower extremity), activities of daily living (ADL), and adverse events. Important outcomes included measures of balance, muscle strength, spasticity, and neurological function. RISK OF BIAS Four review authors independently assessed risk of bias using the Cochrane risk of bias (RoB 1) tool. SYNTHESIS

Methods

We pooled data using random-effects models, calculating risk ratios (RRs) and mean differences (MDs) or standardised mean differences (SMDs). We calculated 95% confidence intervals (CIs) using the Hartung-Knapp-Sidik-Jonkman (HKSJ) method. We calculated 95% prediction intervals when at least five studies were available. Heterogeneity was assessed using the I2 statistic and evidence certainty was evaluated using the GRADE approach. INCLUDED STUDIES We included 43 RCTs involving a total of 1628 participants. The trials were conducted in 11 countries across hospital, rehabilitation unit, or outpatient-clinic settings. The trials primarily recruited participants with both ischaemic and haemorrhagic stroke, mostly in the subacute and chronic phases. The interventions predominantly utilised EEG-based motor imagery - BCI combined with robotic systems, functional electrical stimulation, or neurofeedback. SYNTHESIS OF

Results

Most studies were at low risk of bias for incomplete outcome data and blinding of assessors, but at high or unclear risk for participant blinding and allocation concealment. Overall, the certainty of evidence was low to very low, downgraded primarily for these risk of bias concerns, severe imprecision (due to small sample sizes), and potential publication bias. BCI training versus conventional therapy BCI training may slightly improve upper extremity motor function (MD 4.67, 95% CI 2.25 to 7.09; 10 studies, 611 participants; low-certainty evidence). It is uncertain whether BCI training improves ADL due to very low-certainty evidence. It may result in little to no difference in lower extremity motor function (MD 2.62, 95% CI 2.17 to 3.07; 1 study, 64 participants; low-certainty evidence) and the risk of adverse events (RR 1.11, 95% CI 0.72 to 1.69; 7 studies, 624 participants; low-certainty evidence). BCI training may result in little to no difference in upper extremity spasticity compared to conventional therapy (MD -0.04, 95% CI -0.24 to 0.16; 1 study, 296 participants; low-certainty evidence). No studies reported information on balance, muscle strength, and neurological function for this comparison. BCI training versus active controls It is uncertain whether BCI training improves upper extremity motor function (SMD 0.58, 95% CI 0.23 to 0.92; 14 studies, 331 participants; very low-certainty evidence) and ADL (MD 8.52, 95% CI 1.76 to 15.29; 5 studies, 163 participants; very low-certainty evidence). It may result in little to no difference in lower extremity motor function (MD 2.46, 95% CI 0.71 to 4.21; 4 studies, 130 participants; low-certainty evidence). Furthermore, it is uncertain whether it affects adverse events (RR 0.72, 95% CI 0.23 to 2.32; 9 studies, 234 participants; very low-certainty evidence). BCI training may improve balance (MD 3.25, 95% CI 1.07 to 5.43; 6 studies, 165 participants; low-certainty evidence). It is uncertain whether it improves neurological function or muscle strength due to very low-certainty evidence. It may result in little to no difference in spasticity (low-certainty evidence). BCI training versus sham-BCI training The evidence is uncertain about the effect of BCI training on upper extremity motor function (SMD 0.22, 95% CI -0.08 to 0.52; 9 studies, 279 participants; low-certainty evidence), lower extremity motor function (MD 0.55, 95% CI -5.88 to 6.97; 3 studies, 106 participants; low-certainty evidence), and ADL (MD 6.64, 95% CI -9.95 to 23.23; 1 study, 28 participants; low-certainty evidence). It is uncertain whether BCI training increases the risk of adverse events compared to sham-BCI training (RR 1.35, 95% CI 0.32 to 5.72; 7 studies, 205 participants; very low-certainty evidence). BCI training may result in little to no difference in balance (MD 1.28, 95% CI -0.16 to 2.72; 1 study, 28 participants; low-certainty evidence). It is uncertain whether it improves muscle strength on the wrist extensor (MD 0.70, 95% CI 0.33 to 1.08; 2 studies, 51 participants), spasticity (SMD 0.30, 95% CI -1.00 to 1.61; 3 studies, 82 participants), and neurological function (MD 2.60, 95% CI -3.35 to 8.55; 1 study, 27 participants), all based on very low-certainty evidence. AUTHORS'

Conclusions

Compared with conventional therapy, BCI training may slightly improve upper extremity motor function, with little to no difference in lower extremity function. ADL effects are very uncertain. Compared with active controls, it may improve balance, with little to no difference in lower extremity function. Effects on upper extremity function and ADL are very uncertain. Compared with sham-BCI, BCI training may result in little to no difference in motor function or ADL. Regarding adverse events, it may result in little to no difference versus conventional therapy, but remains uncertain across other comparisons. The certainty of the evidence ranged from low to very low across all comparisons. These findings were primarily limited by high risk of bias, small sample sizes, heterogeneity, and possible publication bias. Future larger, adequately powered, and methodologically rigorous RCTs, particularly those utilising sham-BCI controls and standardised outcome measures, are needed to determine the specific therapeutic benefits of BCI training and to adequately assess potential harms.

Funding

This Cochrane Review had no dedicated funding. REGISTRATION Protocol (2022) DOI: 10.1002/14651858.CD015065.

DOI PubMed Semantic Scholar
View source

Similar papers

#small language model Open access Aug 2026

LifeSciBench: Evaluating Language Models on Realistic, Expert-Level Tasks in the Life Sciences

LifeSciBench is introduced, a benchmark of 750 expert-authored tasks designed to evaluate whether language models can handle realistic life science research work, with each constituent task paired with a human expert-written rubric.

Amelia Liu, Andrew Ho, Anne Marie Droste et al. · 2 citations
#small language model Preprint Aug 2026

CritICL: Inference-Time Weak-to-Strong Generalization from Small Language Model Failure Modes

Experimental results show that CritICL consistently outperforms standard in-context learning and achieves performance competitive with or superior to test-time scaling methods, while requiring significantly fewer generations and lower token cost.

Yu-Fan Wu, Yinghui He, Zhengyi Hu et al. · 1 citation
#artificial intelligence Preprint Aug 2026

TestifAI: Tomography-Based Testing for Deep Learning Systems

TestifAI, a deep learning testing framework for efficient and accurate estimation of robustness against combinations of perturbations, is proposed and partial model tomography is introduced, a novel approach to reconstructing model behaviour in a multi-perturbation space from tests that apply only a small number of perturbations.

Arooj Arif, T. Hartung, E. Botoeva et al. · 1 citation
#small language model Sep 2026

Low Carbon Scheduling of Integrated Energy System Based on Large Language Model-Embedded Multi-Agent Reinforcement Learning

The complex multi-energy coupling characteristics inherent to integrated energy system (IES) present unprecedented challenges for the implementation of low-carbon scheduling. Existing optimization methods often exhibit limitations in system scalability, algorithm adaptivity, and carbon reduction efficacy for complex IES. This paper proposes a Large Language Model (LLM)-Embedded Multi-Agent Reinforcement Learning (LEMARL) to address the aforementioned issues. The proposed method integrates the global perception capability of LLMs with the dynamic optimization capability of MARL. Specifically, the LLM-Embedded module generates high-quality reward functions and policy frameworks from a global perspective, while the MARL module leverages these LLM-generated strategies for distributed interactive iterations—greatly enhancing computation efficiency and scalability. Simulation results demonstrate that LEMARL reduces carbon emissions by 7.76% and simultaneously decreases operating costs by 4.49% in a small-scale IES. Furthermore, LEMARL also exhibits superior applicability and scalability in large-scale IES of the IEEE 141-bus power grid integrated with 51-node thermal system.

Chen Xia, Tong Gou, Yinliang Xu et al. · 1 citation
#small language model Preprint Aug 2026

Apodex 1.1: Scaling Agentic Intelligence for Complex Work

General-purpose language models can reason and synthesize knowledge, but complex work also requires sustained interaction with files, information sources, and executable code, together with state maintenance, failure recovery, and verifiable delivery. We call this \emph{working capability}: sustained, verifiable progress toward a real-world objective. Apodex 1.1 develops this capability along two complementary dimensions. \emph{Environment Scaling} expands the diversity and verifiability of executable file, search, and code environments, while \emph{Agentic Coordination Scaling} trains agents to decompose long-horizon tasks, delegate parallel work, integrate asynchronous results, and replan. A shared execution harness and AgentOS maintain task state and provenance across tools and agents, and training turns environment trajectories and coordination traces into reliable behavior. Across complex professional work, finance, scientific research, mathematics, coding, and search, Apodex 1.1 reaches the leading performance band despite using a substantially smaller model than many frontier systems. The 35B-parameter Apodex 1.1 Mini further retains strong working capability in a locally deployable form. These results ground agentic intelligence in useful, verifiable work completed over time and advance our goal of building a \emph{Heavy-Duty Solver} for ambitious, long-running tasks.

Apodex Team B. An, B. Li, B. Wang et al. · 1 citation

Related blog posts