When EMG Biofeedback Works Best in Neurorehab

This review by Mirecki and Field-Fote revisits a classic rehabilitation question with a very modern twist: when does electromyographic biofeedback (EMG BF) actually help upper-extremity recovery after stroke or spinal cord injury (SCI), and when does it merely add shiny tech to a therapy session? Because the field has seen a resurgence of EMG-driven games, exoskeletons, and home programs, this is new emerging research with novel insights—especially in how it organizes the evidence.

EMG BF uses surface electrodes to measure muscle activation and then feeds that information back to the person training, usually through visual and/or auditory cues. More broadly, biofeedback and neurofeedback are learning-based approaches that give people real-time information about their physiology or brain activity, with the goal of improving self-regulation through practice and reinforcement. In neurorehabilitation, that matters because recovery is not just “getting stronger”—it is relearning coordination, timing, and confidence in movement.

What makes this review clinically useful is its motor-learning lens. Instead of grouping EMG BF by device type (game vs exoskeleton vs stimulation), the authors categorize interventions by the kind of augmented feedback delivered: knowledge of performance (KP)—feedback about how a movement is being performed—or knowledge of results (KR)—feedback about whether the person achieved the goal. That distinction sounds subtle, but it’s the difference between a coach saying “keep your wrist neutral and recruit the extensors smoothly” versus “you hit the target—nice.” The review asks a simple clinical question with big implications: which features of EMG BF are linked to better functional outcomes in stroke and SCI?

Methods

The authors conducted a structured (non-systematic) narrative review of EMG BF interventions targeting upper-extremity deficits in stroke and SCI. Searches were performed in PubMed, CINAHL, and EBSCO for English-language publications through June 2024, primarily within the last 15 years, with additional older SCI studies included via hand-searching due to limited SCI literature.

Inclusion criteria emphasized interventions where EMG originated from the affected limb (i.e., the signal required volitional activation of target muscles). Studies relying on the unaffected arm to drive therapy for the affected side were excluded. Case studies were also excluded. Included interventions could be EMG BF alone or combined with usual care.

A key methodological step was the classification of interventions into:

• Knowledge of performance (KP): real-time feedback during task performance about movement quality or ongoing activation relative to a target (e.g., live EMG trace with auditory threshold tone; some videogame designs where avatar movement reflects continuous performance).

• Knowledge of results (KR): feedback after an attempt about success/failure in achieving a target threshold, often using EMG to trigger an external event such as functional electrical stimulation (FES), exoskeleton movement, or discrete game actions.

The authors also organized studies by:

• Study design: experimental (randomized/comparator) versus quasi-experimental (pre–post without a control group).

• Acuity: acute/subacute (< 6 months post-injury) versus chronic (> 6 months post-injury).

Twelve studies were included. Seven used experimental designs, five quasi-experimental. In experimental studies, KP EMG BF was more common than KR. In quasi-experimental work, KR approaches were more common. Outcomes across studies included functional measures (e.g., Fugl-Meyer Assessment upper extremity subscore, Action Research Arm Test, Wolf Motor Function Test, Motor Activity Log), spasticity (e.g., Modified Ashworth Scale), and biomechanical/EMG metrics.

Results

Across stroke and SCI, KP EMG BF (real-time visual/auditory EMG) often showed within-group improvement but did not consistently outperform usual care on functional endpoints. In acute/subacute stroke, some KP-based protocols—particularly when paired with interventions that amplified activation (such as EMG-linked stimulation approaches described in the review)—were associated with improvements in impairment measures (e.g., spasticity and range of motion) and, in some cases, functional scales. However, several experimental comparisons still found that functional gains were similar between EMG BF + usual care and usual care alone.

In chronic stroke, KP EMG BF delivered via videogame-style training showed mixed results: improvements in specific biomechanical measures (e.g., co-contraction ratios) did not reliably translate into broad functional gains.

KR EMG BF approaches in chronic stroke—where EMG activation triggers discrete events (game action, exoskeleton assistance, or stimulation)—were more likely to show meaningful pre–post improvements on functional measures such as Motor Activity Log, Action Research Arm Test, Fugl-Meyer Assessment, and spasticity ratings. The catch is methodological: many KR studies were quasi-experimental, limiting confidence about whether EMG BF itself caused the improvements versus factors like practice dose, novelty, or concurrent therapy. One experimental trial comparing EMG-linked approaches to other active comparators reported no significant between-group differences.

In SCI, the evidence base was older and concentrated in KP paradigms. Across controlled studies in acute/subacute and chronic SCI, EMG BF did not demonstrate superior functional outcomes compared with usual care or stimulation-based comparators. Notably, KR-style EMG BF approaches were not represented in the SCI literature captured by this review, leaving a major gap.

Discussion

This review’s most useful contribution is not a single “EMG biofeedback works/doesn’t work” verdict, but a more precise clinical framing: the learning signal embedded in EMG BF—KP versus KR—may shape outcomes.

From a motor-learning standpoint, KP is often most helpful during skill acquisition, when a learner needs guidance about movement quality, timing, and strategy. After stroke or SCI—especially early after injury—many people are, functionally speaking, novice learners again. Their nervous system is exploring a new landscape: altered strength, disrupted sensation, spasticity, pain, fatigue, and compensations that feel efficient short-term but can limit long-term recovery. In that context, KP-like EMG feedback can act like a mirror for muscles that are hard to “feel.” It can externalize internal signals, making subtle activation visible and coachable.

KR, in contrast, may be more powerful for retention and self-directed learning. When feedback is about success/failure in hitting a target, the nervous system is nudged to solve the problem—find a strategy that works—without being over-instructed. That can foster durable learning, even if the short-term performance looks messy. Clinically, this resembles how we often train function: repeat meaningful goals, reward success, and gradually reduce guidance.

The pattern in the reviewed literature fits this theory, but with important caveats. KP studies more often included controlled designs and still frequently looked similar to usual care on functional measures, suggesting that simply adding real-time EMG traces may not be enough—especially if the underlying task practice is already strong. KR studies were more likely to show functional gains, but many lacked control groups, which makes the true effect size uncertain.

A practical clinical take-home is to treat EMG BF as a delivery system for feedback, not a treatment in itself. The clinician’s job becomes: (1) choose the right feedback type for the person’s learning stage, (2) embed it in functionally meaningful tasks, and (3) titrate feedback—more guidance early, more autonomy later.

This review also highlights a translational gap in SCI. The SCI evidence largely reflects older KP approaches and did not show superiority over comparators. Meanwhile, KR-style innovation (games, exoskeleton triggers, home programs) has mostly been explored in stroke. Given the centrality of upper-extremity function to independence in SCI, the absence of contemporary KR trials is striking—and represents a clear research priority.

Interpretive thread: across neurorehabilitation, the most reliable improvements often come from combining high-quality practice with feedback that is informative but not overwhelming. If feedback becomes noise—too continuous, too attention-grabbing, too disconnected from real-world goals—it can compete with the very automaticity we are trying to rebuild. The “right” EMG BF might be less about fidelity of the signal and more about timing, dose, and how elegantly it reinforces problem-solving.

Brendan’s perspective

Coupling EEG neurofeedback with EMG biofeedback is one of those ideas that feels almost obvious once you say it out loud: if recovery is a conversation between brain and body, why would we train only one side of the dialogue?

EMG biofeedback is a bottom-up tool. It tells the nervous system, in real time, what the muscles are doing—how much activation is present, whether a threshold is reached, whether co-contraction is changing. For many clients, especially after stroke, that signal is a substitute for sensation: a visual or auditory “touch” that makes an otherwise faint motor command feel tangible.

EEG neurofeedback, by contrast, is a top-down tool. It trains the state that movement is built on: arousal, attention, readiness, and the brain’s ability to shift efficiently between engagement and inhibition. In clinic, the difference shows up as a simple question: does the person have the capacity to access the movement they technically can perform? Fatigue, pain, frustration, hyperarousal, and reduced attentional control can keep the motor system locked in “high effort, low precision.”

When you combine the two, you can build a training stack that respects both realities: movement is learned through peripheral practice, and it is expressed through central state regulation.

Why the pairing can be synergistic

A common pattern in upper-extremity rehab is that the right task is chosen, the right muscles are targeted, and the person still gets stuck. Often, it is not because the muscle can’t activate; it’s because the nervous system can’t consistently find the state that supports clean recruitment.

Here’s where EEG neurofeedback can act like the “traffic controller,” and EMG biofeedback becomes the “road feedback.” If EEG training improves the brain’s capacity for calm focus and flexible engagement, EMG training can immediately translate that improved state into better motor output—less overflow, smoother onset/offset, and more accurate threshold hits.

A clinically practical way to structure sessions

I like to think in phases within the same visit:

1. State-setting (EEG neurofeedback): brief training to establish a stable attentional set and reduce unnecessary motor “noise.”

2. Skill expression (EMG biofeedback): task-specific practice where the EMG feedback reinforces either movement quality (knowledge of performance) or goal success (knowledge of results), depending on the learning stage.

3. Transfer: remove or fade the feedback and immediately practice the same movement in a functional context.

Concrete EEG ideas that pair well with EMG work

The point isn’t to make the EEG protocol fancy; it’s to make it useful.

• For clients who are over-aroused, anxious, or bracing through effort, enhanced alpha training can be a powerful primer for smoother recruitment. A common starting point is posterior alpha support (often around POz) to promote a calmer baseline before EMG-guided movement practice.

• For clients who are under-aroused, inattentive, or “drifting,” SMR-range training (often around 12–15 Hz) at sensorimotor sites can support behavioural inhibition and steadier motor output. A common starting montage is C3 or C4 (or more rarely Cz depending on presentation), chosen to support stable readiness without pushing into excessive activation.

• For clients with marked performance variability, protocols that reward stability—rather than intensity—can reduce the all-or-nothing effort spikes that show up as EMG bursts, co-contraction, or tremulous recruitment.

Where EMG feedback choice matters

This review’s KP versus KR distinction becomes even more meaningful when EEG training is in the mix.

• In earlier stages, KP-style EMG feedback can be paired with EEG training that reduces noise and improves sustained attention. The combo is: refine the movement while stabilizing the brain state that supports refinement.

• In later stages, KR-style EMG feedback (threshold-triggered events) can be paired with EEG training that supports autonomy and resilience under challenge. The combo is: reward success while the brain learns to stay organized under effort.

Integration, not stacking gadgets

The temptation is to do “EEG first because it’s cool,” then “EMG because it’s measurable.” The better approach is to let the clinical problem choose the order.

• If the person arrives dysregulated—frustrated, fatigued, overloaded—start with EEG to create access.

• If the person arrives ready and motivated, start with EMG to capture that readiness, then use EEG later to consolidate a steadier baseline for home carryover.

A note on research versus real clinic life

Studies often isolate a single mechanism. Clinics rarely have that luxury. People bring sleep debt, pain, fear of failure, and months of reinforcement history. That is why the top-down/bottom-up pairing is so attractive: it addresses both the “system settings” (EEG state) and the “output channel” (EMG recruitment) in the same learning loop.

In the future, I’d love to see trials that explicitly test this combined model: EEG-guided state training followed by EMG-guided task practice, with planned feedback fading and real-world functional transfer. If the field is serious about moving beyond mixed results, it will need designs that treat learning as a staged process, not a single intervention.

Used thoughtfully, coupling EEG neurofeedback with EMG biofeedback can turn rehabilitation into something closer to what it already is at its best: a guided conversation between intention and action, where the brain learns to set the conditions and the body learns to deliver the movement.

Conclusion

EMG biofeedback remains one of rehabilitation’s most intuitive tools: show the nervous system what it is doing, reinforce what we want, and repeat it in meaningful tasks. This review clarifies why the literature can look “mixed” when EMG BF is treated as a single category. When EMG BF delivers knowledge of performance, it may help some people—especially earlier after injury—but often performs similarly to strong usual care on functional measures. When EMG BF delivers knowledge of results through threshold-triggered games, exoskeleton assistance, or stimulation, functional gains appear more consistent in stroke studies, though many designs limit causal certainty.

For clinicians, the actionable move is to match the feedback to the learning goal: use performance-focused cues to shape a movement pattern, then transition toward results-focused reinforcement to build autonomy and retention. For researchers, the message is equally clear: the next wave of trials—particularly in SCI—should test KR approaches with rigorous comparators and standardized definitions of usual care. Done well, EMG biofeedback can be more than an add-on; it can be a targeted learning signal that helps the nervous system practice smarter, not just harder.

References

Mirecki, M. R., & Field-Fote, E. C. (2026). Clinically relevant factors impacting the efficacy of electromyographic biofeedback in stroke and spinal cord injury rehabilitation. Discover Neuroscience, 21, 6. https://doi.org/10.1186/s13064-025-00226-5