Enhancing Physiotherapy Outcomes Through Multimodal Interventions in Post-Stroke Rehabilitation

Abstract

Post-stroke rehabilitation integrates technological feedback systems to enhance motor relearning and autonomic regulation. Among these, physiological biofeedback—based on electromyography (EMG), heart rate variability (HRV) and electrocardiography (ECG)—represents a multimodal approach for restoring neuromotor control and autonomic balance. EMG biofeedback enables patients to visualize and voluntarily modulate muscle activation, supporting cortical reorganization and improving movement precision through real-time feedback. Recent meta-analyses confirm that EMG biofeedback significantly improves upper- and lower-limb function in stroke survivors, particularly when combined with task-oriented physiotherapy. EMG biofeedback demonstrates improvements in swallowing function, motor control, and patient motivation. Beyond the motor domain, HRV biofeedback has shown substantial benefits lately, especially in regulating the autonomic nervous system (ANS) activity, improving vagal tone, and reducing sympathetic overdrive: a major contributor to fatigue and cardiovascular instability post-stroke. By targeting the sympathetic–parasympathetic balance, HRV biofeedback not only enhances autonomic flexibility but also supports emotional and cognitive recovery. Together, these modalities integrate neuromuscular and autonomic rehabilitation, offering a path toward individualized, feedback-driven recovery protocols. This narrative review synthesizes recent evidence on the mechanisms, the clinical outcomes, and translational potential of EMG- and HRV-based biofeedback in stroke rehabilitation, highlighting their role in advancing physiotherapy toward an adaptive, data-driven, and neuroplastic paradigm, as from now on, the emerging directions will include integrating physiological biofeedback with immersive or AI-driven platforms for enhanced personalization and motivation.

1. Introduction

Integrating electromyography (EMG) and heart rate variability (HRV) systems into post-stroke rehabilitation can offer significant advantages in monitoring and enhancing patients’ functional recovery in chronic ischemic stroke. By combining these physiological signals, comprehensive biofeedback proves the utility of rehabilitation strategies aimed at restoring post-stroke motor function, especially in spastic hemiparesis.

EMG helps the assessment of muscle activity and identifies patterns associated with motor recovery, as chronic post-stroke patients often display altered muscular activation, affecting their ability to regain functional mobility. The frequency spectrum analysis of EMG signals has proven efficacy in evaluating muscle fatigue and abnormal activity patterns, providing insights into the rehabilitation process [1]. EMG can also facilitate the development of advanced rehabilitation technologies, such as robotic systems that adapt to the real-time needs of patients, ultimately improving motor control and learning through structured feedback [2].

Parallel to EMG, HRV and electrocardiographic (ECG) measurements are essential for evaluating the cardiovascular health and autonomic nervous system function, particularly following a stroke, as autonomic dysregulation is a common finding. HRV also offers predictive insights into the rehabilitation outcomes and can correlate significantly with different recovery trajectories [3,4]. Therefore, ECG-derived HRV reflects the sympathetic and parasympathetic influences on heart rate, which are useful for understanding the physiological state of the patient, as a decrease in HRV often signifies a higher risk associated with cardiovascular complications, which is particularly relevant during post-stroke recovery phases [3].

A well-oriented integration of these systems can lead to innovative rehabilitation devices enabling real-time monitoring of both neural and muscular responses. Smart clothing equipped with ECG and EMG sensors can enhance the personalization of rehabilitation by continuously tracking patient metrics and providing immediate feedback on performance, thus reassuring the motivation and the adherence to therapeutic activities [5,6]. The fusion of ECG and EMG data is currently being used to develop advanced physiological devices for emulating traditional exercises through controlled biofeedback mechanisms [7]. The synergistic approach of integrating EMG and HRV addresses the need for smart wearable technologies that healthcare providers can further adapt their interventions to, within a holistic follow-up of the rehabilitation [1,7]. Machine learning and artificial intelligence (AI) have emerged as natural extensions to traditional signal analysis, offering tools that are expected to be powerful in extracting clinically meaningful patterns from EMG, HRV, and related biosignals. These data-driven methods promise real-time interpretation, adaptive feedback, and predictive modeling, thereby paving the way for personalized and responsive rehabilitation interventions. Nevertheless, there is still limited evidence on the integration of these multimodal systems, particularly in real-world or home-based settings. Methodological and user-acceptance barriers for synchronizing wearable and AI-assisted platforms are still underexplored [1,4,7].

While EMG and HRV biofeedback each provide valuable insights into post-stroke recovery, relying on either investigation alone offers only a partial view of a patient’s rehabilitation status. EMG captures real-time neuromuscular activation and motor control, but does not reflect the underlying autonomic state, stress response, or cardiovascular adaptability that strongly influence rehabilitation engagement and, most importantly, fatigue. On the other hand, HRV reveals key information about autonomic nervous system regulation and readiness, yet lacks the specificity to monitor voluntary movement quality or detect abnormal muscle synergies [1,4,7].

Due to an intersection of risk factors, such as cardiac arrhythmias or high blood pressure, and signs of post-stroke fatigue, orthostatic intolerance, or unexplained performance variability, simultaneous monitoring of these parameters becomes especially relevant (although at first view, these domains are not inherently or directly linked at a mechanistic level) [3,4].

Critically, ischemic stroke recovery is defined by the dynamic interplay between neuromuscular and autonomic systems, especially in chronic evolution. Deficits in one domain can impede progress in the other, just as autonomic dysregulation can exacerbate motor fatigue, while inadequate motor activity may prolong sympathetic overdrive. By integrating EMG and HRV monitoring, a much more synchronized view of both motor performance and autonomic status is expected [7]. This combined approach enables the identification of maladaptive patterns (such as motor effort undermined by autonomic instability) and also supports the development of individualized, adaptive interventions for both neuromuscular retraining and autonomic recovery [7].

Although this narrative review highlights emerging techniques, robust validation studies (especially large-scale, longitudinal clinical trials) are lacking. The long-term reliability, accuracy, and impact of these systems on functional and autonomic recovery outcomes are not yet well-established. As most of the current protocols rely on pre-set feedback parameters or static training regimens, within our research, we navigate on insufficient data on how exactly multimodal biofeedback systems can be tailored to individual patient profiles, accounting for the broad heterogeneity in stroke severity, cognitive function, and comorbidities. The aim is to critically synthesize recent advancements in the integration of EMG and HRV-based biofeedback within post-ischemic stroke rehabilitation, with a particular focus on the convergence of multimodal physiological monitoring and artificial intelligence. The purpose of the current research lies in assessing how synchronized neuromuscular and autonomic feedback, when coupled with adaptive, data-driven technologies, could address the current limitations of single-modality approaches.

2. Material and Methods

This narrative review is based on a comprehensive synthesis of the peer-reviewed literature through databases such PubMed, Scopus, and Web of Science, by searching combinations of keywords such as “post stroke HRV”, “post stroke EMG”, “poststroke surface EMG”, “poststroke conventional EMG”, “poststroke EMG biofeedback”, “stroke rehabilitation artificial intelligence”, “stroke rehabilitation artificial intelligence HRV” and “stroke rehabilitation vagal stimulation”.

A total of 484 records were retrieved. After removing duplicates, 372 studies remained for screening. Titles and abstracts were independently screened by three main reviewers, and 124 full-text articles were assessed for eligibility. Discrepancies were resolved by consensus or consultation with a fourth reviewer.

The inclusion criteria comprised randomized controlled trials, observational studies, systematic reviews, and meta-analyses reporting on physiotherapeutic interventions for chronic motor deficit in post-stroke patients (hemiparesis), with outcomes that are relevant to EMG, HRV, ECG, or AI-driven feedback. While the primary focus was on the literature published from 2019 onward to capture the most recent technological advances, select studies prior to 2019 were included if they contributed to understanding the technological evolution that has since influenced current clinical applications (e.g., foundational developments in EMG analytics or HRV biofeedback that are now realizable through AI-driven platforms).

As exclusion criteria, animal studies, case reports, editorials, letters, commentaries, and conference abstracts without full data were avoided. Other excluded sources include studies focusing exclusively on pharmacological interventions and non-stroke motor disorders. The gray literature and non-peer-reviewed sources were also excluded to maintain the quality and reliability of the evidence.

We included studies reporting positive, neutral, and negative outcomes, focused solely on physiotherapeutic interventions aiming to provide a balanced and comprehensive overview of the existing evidence. Data from eligible articles were extracted using a structured template, focusing on study design, participant characteristics, intervention details, outcomes, and key findings. Thematic synthesis was employed to analyze the mechanisms, clinical benefits, and limitations of each technological feedback approach. The strength of evidence, clinical relevance, and feasibility of patient adherence were evaluated for translational potential. The review was structured thematically, and data were analyzed in a descriptive approach to examine the benefits, mechanisms, and limitations of each intervention, to better understand the technological feedback systems involved in the rehabilitation strategies. The ideas were synthesized based on the strength of the available evidence, clinical relevance, and feasibility of patient adherence.

Within the current review, consideration was given to the limits of the available data, especially since the process of signal acquisition can be influenced by a variety of internal and external factors. Internally, the neurological deficits through hemiparesis and spasticity, as well as the shifts in muscle tone, may transform the morphology of EMG signals. The autonomic dysregulation induced by central lesions may produce lasting changes in HRV: arrhythmias or heightened sympathetic activity. Also, post-stroke exhaustion, sedatives, beta-blockers, and anti-spasticity drugs can all leave their imprint on neuromuscular and autonomic signals, and the cognitive impairment of the patients may influence the consistency of the recorded signals. External influences are equally significant. The placement of electrodes and the condition of the skin are often overlooked but are vital details, as hemiparetic limbs may display edema, compromised skin integrity, or muscle atrophy, all of which can disrupt the contact between electrode and skin, thereby affecting the signal quality. Movement artifacts, whether originating from involuntary muscle contractions, tremors, or external electrical noise from nearby rehabilitation devices, can further contaminate the recordings. Also, environmental factors, such as room temperature, can subtly influence HRV readings, NCS parameters and patterns of muscle recruitment as measured by EMG. The following analysis considers the most suitable data that is available in the current literature, provided by authors that have expressed awareness of this intricate interplay of internal and external elements throughout their work, either in their methodology, discussion or limitations sections.

3. Integrating EMG into the Physiological Biofeedback of Post-Stroke Rehabilitation

3.1. Surface EMG Within the Concept of Biofeedback

Integrating electromyographic (EMG) biofeedback into post-stroke rehabilitation strategies has gained significant attention in recent years due to its efficacy in enhancing motor function recovery, addressing co-activation of muscles, and facilitating neuromuscular retraining. EMG biofeedback provides patients with real-time data on their muscle activity, enabling enhanced engagement and motivation during rehabilitation exercises [1,2,6].

The utility of EMG biofeedback in conjunction with other rehabilitation modalities, particularly in improving upper limb motor function and alleviating pain associated with post-stroke shoulder-hand syndrome, have shown promising results so far [8]. This method utilizes the brain’s capacity to form new neural pathways in response to visual and auditory feedback derived from EMG signals, thereby aiding in the restructuring of motor functions affected by stroke [8]. In this context, the application of myoelectric computer interface training further illustrates the potential of EMG biofeedback to target co-activation (an often-overlooked aspect in post-stroke rehabilitation). By systematically reducing co-activation of spastic muscles, patients exhibit improved arm movement and function, underscoring the complexity of muscle interactions in stroke recovery [9]. Similarly, targeting abnormal co-activation patterns through structured feedback mechanisms can significantly reduce motor impairments and enhance neuromuscular function [10]. This highlights the role of precise biofeedback in not only enhancing motor control but also optimizing the rehabilitation process.

EMG biofeedback has shown promise when it is innovatively combined with therapies like electrical stimulation and mirror therapy, as described by Kim and Jang, which not only improves postural stability but also enhances gait functionality in chronic stroke patients [11]. This integrative approach combines the strengths of visual feedback methods with the real-time muscle activity insight, provided by ultrasound correlations, thus maximizing the rehabilitation follow-up and effectiveness [12].

The feasibility and usability of EMG biofeedback systems have also been evaluated in telehealth contexts. Donnelly et al. highlighted the necessity of understanding patient acceptability to ensure successful engagement with telerehabilitation systems, demonstrating the importance of tailoring EMG biofeedback interventions to meet the comfort and operational needs of stroke survivors [13]. Such adaptations can broaden access to rehabilitation services, particularly for those suffering from severe functional impairments. The continuous feedback nature of EMG biofeedback therefore fosters patient motivation and compliance, which are critical determinants for long-term rehabilitation success. EMG-based feedback methods can contribute to improved gait biomechanics and enhance various motor and cognitive parameters when integrated thoughtfully into rehabilitation programs [14,15]. Indeed, the dynamism of this approach reveals a robust pathway towards individualized, efficient rehabilitation protocols for stroke patients.

Despite all these facts, the integration of EMG biofeedback in post-stroke rehabilitation presents several controversies, despite its potential for enhancing motor recovery. Concerns regarding its efficacy, practicality, and generalizability have emerged within the academic and clinical communities. One major concern involves the mixed results associated with the use of EMG biofeedback for motor learning and functional recovery post-stroke, as for example, improvements in arm movement and reduced co-activation through EMG biofeedback are not always achieved [9]. Furthermore, many EMG biofeedback interventions have primarily focused on muscle strengthening and spasticity reduction, rather than sufficiently addressing co-activation patterns, which may limit the overall effectiveness on functional outcomes, according to the same author’s report [9]. This inconsistency raises questions about the specificity of EMG biofeedback training and whether all patients respond positively to such interventions, indicating a need for protocol refinement.

Debate also exists regarding the effectiveness of EMG biofeedback compared to other therapeutic modalities. Integrating EMG with robotic-assisted therapy yields greater benefits than using EMG biofeedback alone [16]. Nevertheless, Gámez et al. still note concerns about the limited success of EMG biofeedback in older populations and the variability of results across different demographics, suggesting that age and pre-existing conditions may influence rehabilitation outcomes, thus highlighting the need for tailored interventions [17].

Still, EMG biofeedback, particularly when combined with task-oriented training, improves functional outcomes in both the upper and lower limbs (according to a meta-analysis on post-stroke shoulder-hand syndrome, which included 45 RCTs, comprising 3379 participants) [8]. Effect sizes are generally moderate for upper limb function and pain relief, although results are more variable for lower limb recovery. A critical appraisal of the evidence base reveals significant heterogeneity in sample sizes and study quality [8,9]. Although small cohorts limit generalizability, comparisons of outcomes indicate that while EMG biofeedback shows promise, especially for post-stroke shoulder-hand syndrome, its efficacy is not universally observed. Inconsistent functional gains can be found across studies, suggesting modest improvements or no significant effect on strength. Variability is increased in elderly populations and those with severe baseline impairment, suggesting that patient selection and individualized protocol adaptation are key [8,9,17].

Also, some degree of technological discomfort and usability challenges (such as interface complexity, sensor placement, equipment maintenance and limited technological experience) can limit the effectiveness of such interventions in practice. This indicates that despite evidence supporting EMG biofeedback, its implementation in real-world settings may be met with skepticism due to operational challenges faced by stroke survivors during rehabilitation [18]. Still, EMG biofeedback combined with rehabilitation training is ranked as the most favorable intervention option for the combined aims of improving upper-limb motor function and relieving pain [8]. However, this comparative analysis of short-term/endpoint clinical improvement does not relate to the persistence of relearned motor skills. This limitation suggests that reliance solely on technology-assisted biofeedback may be inadequate unless complemented by traditional therapies.

There are also compliance considerations regarding patient autonomy and subjective experiences in biofeedback training that deserve attention. Usually, in neurological conditions, EMG biofeedback heavily depends on patient engagement and self-monitoring; practitioners must recognize the psychological implications of relying on technology that may not align with all patients’ capabilities and preferences [19].

3.2. Surface EMG Versus Conventional EMG

For all the limitations regarding biofeedback EMG, which is based on surface EMG techniques, there is a possible alternative when returning to basics. The use of conventional EMG, which includes nerve conduction studies (NCS) and needle EMG, in post-stroke rehabilitation may be significant when contrasted with the surface EMG methodologies. This discussion relies on the evidence from various studies, emphasizing how the differing EMG approaches can inform rehabilitation strategies for stroke survivors. While surface EMG remains widely adopted due to its non-invasive nature, its limitations in detecting deep muscle or isolated motor unit abnormalities have therefore prompted renewed interest in the conventional EMG techniques.

Conventional EMG provides detailed insights into motor unit behaviors influenced by stroke-related neurological changes. Abnormalities in electrophysiological signals from motor units in paretic muscles are often more detectable via intramuscular techniques compared to surface EMG. This is relevant in post-stroke patients who exhibit fibrillation potentials and positive sharp waves that are indicative of alterations in muscle innervation and motor neuron integrity [20]. Furthermore, Li et al. have shown that different EMG techniques can reveal varying degrees of motor unit physiology, admitting that needle EMG may uncover dysregulations that surface EMG fails to capture, such as isolated motor unit firing characteristics post-stroke [21].

Needle EMG’s capability to provide specific information about muscle fibers makes it essential for assessing the severity of motor unit degeneration or reinnervation post-stroke [22]. Significant changes in motor unit firing behavior in chronic stroke patients allow for a deeper understanding of complications such as spinal motoneuron degeneration, which surface EMG does not fully reveal [21,22]. Such insights are invaluable for tailoring rehabilitation protocols to address specific motor weaknesses.

On the other hand, surface EMG has been widely adopted due to its non-invasive nature and ease of use. However, challenges persist, particularly regarding the ability to reliably detect voluntary muscle activity against the background of involuntary contractions that are frequently observed in stroke patients [23]. The spurious spikes that are prevalent in surface EMG recordings can obscure the true onset of voluntary muscle activity—a critical aspect when evaluating patient rehabilitation outcomes and designing task-specific interventions [23]. Thus, while surface EMG provides a broader overview of muscle activation patterns, it may not effectively capture the complexities of motor control following a stroke. Brief comparisons between the two techniques emerge, as shown in Figure 1.

Integrating both conventional EMG approaches with surface EMG could enhance rehabilitation outcomes. Electromyography-driven neuromuscular electrical stimulation (NMES) systems, which adapt based on real-time muscle signals, show promise in improving motor recovery. So far, EMG-triggered NMES has effectively enhanced functional outcomes in upper limb rehabilitation [24,25]. These systems benefit from the precision of conventional EMG in optimizing stimulation parameters, addressing the specific neuromuscular deficits that are identified through detailed analyses like needle EMG [26].

The choice between using conventional EMG or surface EMG in post-stroke rehabilitation may ultimately depend on the specific clinical context. For patients demonstrating significant potential for motor recovery, surface EMG combined with functional task training may foster better engagement and quicker rehabilitation outcomes [27]. Conversely, in cases of severe impairment where motor unit pathology is pronounced, the precision offered by conventional EMG could be more beneficial in crafting targeted rehabilitation strategies [28]. Conventional EMG offers detailed, diagnostic insights into motor unit integrity, neuromuscular junction function, and the presence of underlying neurological or peripheral nerve pathologies that may not be detected by surface EMG alone. Also, recognizing specific patterns of impaired muscle activation or nerve involvement allows for more targeted electrode placement, personalized feedback thresholds, and the adaptation of training intensity [28]. Therefore, the inclusion of conventional EMG in a biofeedback context may also be useful to consider, when possible, for a patient-centered approach.

NCS, as part of the EMG investigation protocol, can also serve as a useful diagnostic tool in elucidating the involvement of peripheral nerves and guiding the rehabilitation strategies. An important aspect of NCS is their ability to detect conditions such as peripheral neuropathy, which can exacerbate functional impairments in stroke patients. In this context, a significant number of chronic stroke patients exhibit slowed motor nerve conduction velocities and reduced sensory nerve action potentials, indicating potential peripheral nerve involvement that may also require targeted therapeutic approaches [29].

The motor nerve conduction studies on the axillary nerve in patients with hemiplegia suggest significant differences between affected and unaffected sides, providing objectivity in assessing motor deficits that are related to stroke events [30]. Such findings emphasize the importance of NCS in assessing not only the extent of stroke-related impairment but also in identifying underlying peripheral nerve pathologies that could be treated to improve rehabilitation outcomes, and also the utility to be considered within the biofeedback protocol.

Moreover, NCS can be particularly valuable in distinguishing between stroke-related symptoms and other neuropathies when central causes have been ruled out. The ability of NCS to detect abnormalities in nerve conduction, even in the absence of overt symptoms, can lead to earlier intervention and more appropriate therapeutic strategies [31]. Furthermore, the investigations carried out by Folyovich et al. highlight that normal nerve conduction study results can redirect the diagnostic focus toward neuroimaging when cerebral lesions are suspected, thus avoiding the risk of misdiagnosis and inappropriate treatments [32].

The implications for rehabilitation are further reinforced by the work of Chen et al., which highlighted changes in peripheral axonal properties post-stroke. When nerve conduction studies indicate alterations in axonal excitability, this provides essential data for addressing the rehabilitation efforts, based on individual patient needs [33]. Recognizing these electrophysiological changes allows for advanced rehabilitation protocols to facilitate better recovery.

Nowadays, surface EMG and NCS offer higher specificity for identifying subclinical motor unit dysfunction and peripheral nerve involvement after stroke. Studies with moderate-to-large samples demonstrate subtle alterations in motor unit firing and axonal excitability [21,33]. The degree of alterations in axonal characteristics appears to be correlated with the severity of weakening. The methodological rigor of these studies is generally high, with consistent use of blinded assessors and standardized protocols, in spite of the single-center design limitation [20,21,33].

High-density surface EMG decomposition can be considered a feasible, noninvasive means to assess single motor unit firing behavior, without reliance on needle EMG, which would be much more invasive and more technically demanding. This type of testing is an option that is worthy to be considered in rehabilitation follow-up, due to its accessibility [21].

NCS further complements EMG by objectively characterizing nerve conduction velocity and action potential amplitudes. A larger study reveals that up to 40% of chronic stroke patients exhibit clinically relevant peripheral nerve changes, which can influence rehabilitation strategies [30]. However, the impact of these findings on functional recovery is less well-quantified, and longitudinal studies are needed [20,21,30,33].

Some of the comparison characteristics between surface and conventional EMG, other than the aspects already discussed, can be synthesized, as seen in

Table 1. Comparison of practicality between the two EMG methods.

Figure/Parameter	Surface EMG	Conventional EMG
Data Source	Muscle groups, superficial muscles	Individual motor units, deep/small muscles
Temporal Resolution	High (real-time feedback possible)	High (but focused on bursts/diagnostics)
Biofeedback Use	Widely used (real-time patient feedback)	Still rarely used (mainly diagnostic, not feedback)
Signal Artifacts	Susceptible (crosstalk, movement, noise)	Less susceptible (but more technically demanding)
Limitations	Poor specificity for deep muscles, signal noise [21,23]	Invasive, limited for repeated feedback when used as a single technique [26,27,33]

. These characteristics highlight that surface EMG is preferred, especially in routine physiotherapy and telerehabilitation, while conventional EMG, which is less commonly used for feedback, is essential for diagnosis and complex cases where detailed pathophysiology must be understood; thereby, it has potential to complete and provide personalizing feedback-driven rehabilitation protocols.

Looking forward, advances in AI and machine learning models offer exciting potential to integrate analytics algorithms into rehabilitation. Unlike conventional approaches that rely on manual or semi-automated interpretation, modern machine learning and explainable AI protocols are capable of extracting subtle, multidimensional patterns from EMG and HRV signals, allowing for more precise classification of movement quality, fatigue, and compensatory strategies. One example is the ability to distinguish between pathological and volitional muscle activation, predict rehabilitation outcomes, and adapt feedback in real time, based on the patient’s physiological state. These capabilities pave the way for the development of data-driven rehabilitation platforms, where therapy parameters can be dynamically adjusted [1,2,7,9].

For all that, there is still a paucity of research investigating truly adaptive, closed-loop systems that can dynamically adjust therapy intensity, feedback, and modalities in real time, based on physiological data and patient progress. Given the complementary strengths and limitations of surface EMG, conventional EMG, and NCS, there is an increasing trend toward integrating these modalities within broader, multidisciplinary rehabilitation strategies.

4. Interdisciplinary Perspectives of EMG Biofeedback

The integration of advanced technologies such as brain–computer interfaces (BCIs), electrical stimulation therapies, and transcutaneous auricular vagus nerve stimulation (tVNS) also represents significant advancements in this field (Figure 2).

BCIs enable direct control of peripheral muscles through decoded neural information, effectively establishing a “closed-loop pathway” between the central and peripheral nervous systems. This promotes neuroplasticity and cortical remodeling [34]. Furthermore, methods of direct electrical stimulation of peripheral nerves, such as median nerve stimulation, target the neuromuscular junctions to enhance motor recovery and activate the paretic muscles, providing a focused approach to rehabilitation [34]. Transcutaneous electrical nerve stimulation (TENS) complements these strategies by inducing proprioceptive improvements in upper extremities, significantly enhancing motor function [35].

Vagus nerve stimulation (VNS), paired with rehabilitation, reflects a novel approach focused on enhancing neural recovery. VNS synchronized with rehabilitation exercises improves motor function through the activation of cholinergic and noradrenergic neurons [36,37]. This modality has shown promise in improving upper limb motor function, underscoring the importance of combining peripheral stimulation with active rehabilitation exercises to leverage the neuroplastic mechanisms [38].

Moreover, the assessment of the PNS can inform intervention strategies that are tailored to individual patient needs. Advanced imaging techniques, such as diffusion-weighted imaging, help to identify stroke lesions and their correlation with functional impairments, allowing for the design of a more personalized rehabilitation program [39]. The Fugl-Meyer Assessment, which is useful for evaluating motor function post-stroke, can be effectively integrated with neuromuscular electrical stimulation (NMES) therapies to objectively measure recovery improvements [40].

Also, combining VNS with conventional rehabilitation methods highlights substantial enhancements in overall recovery and functional efficiencies [41]. The assessment of peripheral muscle activity through methods like ultrasound, in addition to EMG, can track muscle atrophy and guide rehabilitation interventions effectively, reinforcing the importance of multi-faceted evaluations in stroke recovery [42].

EMG also plays a role in the management of shoulder subluxation in post-stroke rehabilitation. Such an approach is beneficial in stroke patients, where shoulder subluxation is a common complication that can hinder recovery and functional independence. EMG-triggered FES effectively reduces shoulder subluxation by targeting specific muscles (namely, the supraspinatus and posterior deltoid) to promote muscle activation and support shoulder stability, thereby reducing dislocation occurrences [43]. This helps in the functional recovery of weak musculature, promoting proper shoulder alignment during rehabilitation [44]. Training programs utilizing real-time feedback from EMG allow patients to recognize and improve their muscle activation patterns in recovering shoulder function [45]. This enhancement contributes to better motor outcomes and fosters neuromuscular control that is essential for preventing complications associated with shoulder subluxation, such as adhesive capsulitis and rotator cuff injuries [46,47].

When combined with functional electrical stimulation and positioning aids, EMG biofeedback enables a good adjustment of the rehabilitation protocols, dynamically based on muscle response, thereby optimizing the rehabilitation process [48]. This adaptability ensures more personalized treatment plans that cater to the specific weaknesses and progress of individual patients, ultimately leading to improved shoulder function and pain management, but also identifying associated risk factors within the shoulder subluxation. Assessing muscle thickness and activation patterns could help to classify patients who are at risk and adjust their rehabilitation strategies accordingly [49]. This proactive approach also allows for a good implementation of preventative measures during the early stages of rehabilitation, thereby reducing the risk of developing shoulder-related complications [12,50]. Lastly, the necessity of early intervention through EMG-guided training suggests that addressing shoulder subluxation during the acute phase post-stroke can significantly enhance motor recovery outcomes [44].

Physiological biofeedback using EMG can also enhance the rehabilitation of vascular parkinsonism in post-stroke patients. This approach leverages real-time feedback on muscle activity to support motor training and address specific deficits associated with bradykinesia, rigidity, and postural instability [19].

One primary benefit of EMG biofeedback in this scenario is its capability to facilitate voluntary control over muscle activation. Diotaiuti et al. highlight that EMG biofeedback can improve motor function in individuals with Parkinson’s disease by enabling them to self-regulate muscle activation [19]. This self-regulation is important for patients experiencing vascular parkinsonism, as it can reduce reliance on compensatory movement patterns and help to restore more natural motor behaviors. By training patients to increase muscle activation in specific muscle groups, rehabilitation can therefore target functional impairments more effectively [51,52].

Another important aspect of EMG biofeedback is its potential to promote neural plasticity. Although Hidayati et al. focus on the role of biofeedback in a specific case, the overall notion that biofeedback may enhance neural plasticity using various modalities is plausible and relevant to neurorehabilitation [51]. This is particularly beneficial for patients who have undergone cerebrovascular accidents, as it may facilitate the brain’s reorganization following injury, helping recovery within affected neural pathways. The integration of real-time EMG biofeedback in rehabilitation exercises allows for the precise monitoring of muscle activity and coordination in vascular parkinsonism; where motor coordination can be severely compromised, EMG biofeedback enables an adjustment of the training regimens more dynamically. Tacca et al. demonstrated how high-density EMG technology could identify alterations in movement coordination, emphasizing its role in tailoring rehabilitation strategies to individual patient needs [52]. Such adjustments can improve immediate rehabilitation outcomes and may lead to long-term functional improvements.

Moreover, EMG biofeedback serves as a motivational tool by providing patients with tangible feedback about their progress. Continuous feedback can sustain engagement in rehabilitation activities, improving adherence and overall outcomes. EMG biofeedback maintained patient motivation by visually representing rehabilitation progress, even when gross movements were minimal [53]. This is important, especially for patients with vascular parkinsonism, who may experience fluctuations in motivation due to the chronic nature of their condition.

Additionally, EMG can help to compensate for anxiety associated with motor tasks, empowering patients in their rehabilitation. As patients regain control over their muscle activation patterns, they may experience increased confidence, leading to greater participation in therapeutic exercises, as evidenced in various contexts of EMG biofeedback usage [54,55].

It is also important to utilize a multipronged approach in neurorehabilitation by utilizing EMG biofeedback alongside other interventions to enhance muscle activation and coordination, thus supporting the comprehensive rehabilitation of upper limbs in stroke patients, including those with vascular parkinsonism [56].

Overall, EMG biofeedback represents a crucial bridge between neurophysiological monitoring and active rehabilitation. Its capacity to quantify voluntary muscle activation, detect abnormal co-contractions, and provide instant corrective feedback transforms the rehabilitation process into a data-driven, patient-engaged experience. By visualizing muscle recruitment patterns, stroke survivors gain not only improved motor control but also heightened motivation and self-efficacy, which are indispensable for long-term adherence to therapy.

Even so, the extensive amount of EMG data generated during repetitive motor training—combined with inter-individual variability in signal morphology, fatigue responses, and electrode placement—poses major analytical challenges. Traditional manual interpretation cannot fully capture the multidimensional information embedded in EMG waveforms, especially in chronic stroke patients who exhibit fragmented or low-amplitude activation patterns. Recent work employing explainable artificial intelligence (XAI) demonstrates that machine learning models can decode pathological EMG signatures, classify impairment severity, and distinguish compensatory from volitional activation with higher precision than human raters [1,52].

The same trend is evident in EMG-driven robotic rehabilitation. Adaptive algorithms have been shown to integrate EMG features with cortical or kinematic data to modulate robotic assistance in real time, enhancing motor relearning while preventing over- or under-assistance [26]. Furthermore, multi-signal fusion approaches—combining EMG with physiological markers such as accelerometry, torque or even autonomic indices—enable reliable detection of fatigue and fluctuating effort, allowing exoskeleton systems to dynamically adjust the training intensity and resistance [6].

As a result, the integration of EMG biofeedback with intelligent adaptive systems represents the next frontier of stroke rehabilitation. Machine learning models that are capable of decoding residual muscle activity, even in patients with severe motor deficits [28], offer the foundation for closed-loop physiotherapy platforms that respond automatically to the patient’s physiological state. This technological convergence between EMG analytics, autonomic profiling, and algorithm-driven adaptation establishes the basis for an adaptive, responsive, and personalized rehabilitation paradigm—a concept explored in detail in the following section.

A main advantage of these integrated approaches is their promising adaptability to individual patient needs, though barriers such as technology acceptance, cost, and the need for clinician training persist. Nonetheless, the variability in acquisition methods, signal processing, and outcome measures still makes comparability and translation difficult, especially within clinical practice. Standardized protocols for integrating, calibrating, and interpreting multimodal feedback remain to be developed, with a perspective beneficial addition from HRV parameters.

5. Heart Rate Variability in Post-Stroke Rehabilitation

HRV quantifies the cyclic variations between the ECG recording of consecutive R-R intervals, and represents a useful index of autonomic adaptability [57,58]. In the context of stroke, HRV reflects the integrity of the central autonomic network involving the insular cortex, amygdala, hypothalamus, and medullary centers that coordinate the cardiac vagal output. Post-stroke autonomic dysfunction is frequent and manifests as a shift toward sympathetic dominance and parasympathetic withdrawal [57,58]. This imbalance predisposes patients to arrhythmias, hemodynamic instability, and impaired neuroplasticity. Systematic analyses have so far demonstrated that reduced HRV, particularly decreased SDNN (standard deviation of NN intervals), RMSSD (Root mean square of successive differences), and HF (high-frequency) power, is strongly correlated with stroke severity, cognitive decline, and unfavorable long-term outcomes [59,60]. Conversely, higher HRV levels predict better recovery trajectories, lower complication rates, and reduced mortality [58,59,60]. Therefore, HRV serves both as a prognostic biomarker and as a physiological target for modulation during neurorehabilitation, as the interplay between cardiovascular and neuromuscular signals allows for the development of integrative protocols that optimize both autonomic and motor recovery.

Vagal efferent activity also manifests upon neural and vascular recovery. The baroreflex loop stabilizes the arterial pressure and optimizes the cerebral perfusion. Simultaneously, cholinergic anti-inflammatory pathways mediated via the vagus nerve attenuate systemic cytokine release, thereby protecting neuronal tissue from secondary inflammatory injury. In experimental models, the parasympathetic activation promotes brain-derived neurotrophic factor (BDNF) expression and enhances synaptic reorganization within the perilesional areas. Clinically, patients with preserved or restored HRV present faster improvements in attention, mood regulation, and executive function: outcomes that mirror the neural benefits of heightened vagal tone. These observations position HRV as a physiological bridge between autonomic restoration and functional neuroplasticity [61].

Neuroimaging further supports this link, as functional MRI (fMRI) and resting-state connectivity analyses demonstrate that higher HRV correlates with strengthened coupling between the insular cortex, anterior cingulate, and prefrontal regions: key nodes of the central autonomic network that are responsible for emotional regulation and motor planning [62]. Restoration of these connections appears to facilitate better motor initiation and interhemispheric balance following stroke. This autonomic–cortical interplay suggests that the modulation of HRV not only reflects recovery progress but actively contributes to cortical reorganization.

From a mechanistic standpoint, HRV biofeedback and vagal stimulation converge on overlapping neural circuits that modulate both the hypothalamic–pituitary–adrenal axis and prefrontal inhibitory networks. The increased vagal tone reduces sympathetic hyper-excitability, normalizes limbic over-activation, and re-establishes top-down control over sensorimotor pathways. In turn, this neurovisceral coherence fosters a physiological environment that is conducive to neuroplasticity, enhancing synaptic efficacy, motor-relearning and emotional stability. In this context, targeting HRV represents not merely a monitoring strategy but a therapeutic intervention that is capable of promoting systemic recovery through autonomic–somatic integration [63].

HRV biofeedback (HRV-BF) is a non-invasive technique designed to strengthen baroreflex sensitivity and enhance parasympathetic tone through controlled breathing at the individual’s resonance frequency—typically around 0.1 Hz (approximately six breaths per minute). At this rhythm, heart rate and blood pressure oscillations enter a synchronized pattern that maximizes the vagal activity and cardiovascular efficiency. Paced breathing and visual feedback allow for individuals to perceive their own physiological rhythms and learn to modulate them volitionally, reinforcing autonomic flexibility and emotional stability [63,64]. All these applications are synthesized in Figure 3.

In clinical contexts, HRV-BF is administered in short, structured sessions—generally over 8 to 12 weeks, lasting 20 min each—combining supervised and home-based training. During practice, the patient receives real-time feedback from photoplethysmography or ECG sensors, visualized as a coherence curve. As training progresses, improvements are observed in time-domain metrics (SDNN, RMSSD) and frequency-domain components (HF power, LF/HF ratio), reflecting enhanced vagal modulation. Importantly, HRV-BF induces not only physiological but also psychological benefits—reducing anxiety, depression, and post-stroke fatigue, and improving concentration and self-efficacy [65].

Combining HRV-BF with neuromotor training amplifies its therapeutic effect. When conducted before task-oriented physiotherapy or robotic sessions, HRV regulation optimizes autonomic readiness, facilitating smoother cortical–motor engagement and reduced stress reactivity. Likewise, HRV-guided breathing integrated with EMG-triggered functional electrical stimulation can enhance coordination between autonomic and somatic outputs, resulting in improved muscle control and fatigue resistance. This integrative approach reflects a paradigm shift toward synchronizing cardiovascular and neuromuscular rehabilitation, aligning physiological stability with functional recovery [66]. Reliable HRV-BF implementation requires high-quality signal acquisition and strict adherence to methodological standards. Short-term recordings should last at least 5 min under stable conditions, avoiding arrhythmias or artifacts that may distort frequency analysis [67]. Individual calibration of the resonance frequency—rather than fixed breathing rates—has been shown to maximize baroreflex gain and ensure therapeutic precision [64].

Although future protocols will explore dose–response relationships, optimal session frequency, and combined applications with vagus nerve stimulation (tVNS) or robotic feedback systems, the data that is available so far suggests that patients receiving HRV-BF achieve greater gains in cognitive function (the MoCA evaluation), motor performance (FMA-UE scaling), and psychological well-being (HADS-D), together with improvements in autonomic indices [65]. Usually, at the six month endpoint, HF power and RMSSD increase, whereas the LF/HF ratio declines, indicating a durable shift toward parasympathetic predominance. These outcomes are consistent with earlier findings that vagal normalization enhances cerebral autoregulation and functional recovery [68,69]. HRV-BF therefore acts as both a therapeutic intervention—by strengthening baroreflex sensitivity—and a diagnostic tool for monitoring rehabilitation progress. Beyond the cardiovascular benefits, HRV-BF contributes to emotional stabilization and cognitive endurance, which are key determinants of therapy engagement and independence [68].

Integrating HRV-BF into physiotherapy offers a structured method for autonomic retraining alongside neuromotor exercises. A practical protocol involves 10–20 min of guided breathing at six breaths per minute, three to five times weekly, using ECG or PPG-based monitors. Real-time visualization of HR coherence motivates patients and provides immediate reinforcement. HRV measures such as RMSSD, HF power, and LF/HF ratio can be tracked to evaluate parasympathetic reactivation [70]. Ideal candidates are patients displaying low baseline HRV, anxiety, or fatigue but sufficient cognitive function to follow instructions. Incorporating HRV-BF at the start or end of a session may improve physiological readiness for robotic or task-oriented therapy [65]. Such integration promotes self-efficacy, reduces stress reactivity, and aligns physiotherapy with the broader goals of homeostatic rehabilitation.

Although a fixed six-breaths-per-minute rate standardizes training, individual calibration of the resonance frequency may optimize baroreflex engagement [64]. Reliable HRV analysis requires artifact-free signals, stable recording windows (≥5 min for short-term assessments), and adherence to methodological standards [67,71]. Future investigations should clarify the dose–response relationship between HRV-BF intensity and clinical improvement, explore synergy with non-invasive vagus nerve stimulation (tVNS), and determine cost-effectiveness in routine neurorehabilitation. Emerging neuroimaging data also suggests that restored HRV correlates with strengthened connectivity within the insula–brainstem–prefrontal circuit, supporting the concept of autonomic–neuroplastic co-rehabilitation [62].

6. AI-Driven Adaptive Feedback and Predictive Analytics in Post-Stroke Rehabilitation

The integration of artificial intelligence (AI) and machine learning (ML) algorithms into rehabilitation offers a transformative shift toward personalized and adaptive therapy. Physiological signals such as EMG and HRV contain complex, multidimensional information that reflects the patient’s motor intent, autonomic state, fatigue level, and neuromuscular coordination. Traditional interpretation methods are insufficient for capturing these nonlinear patterns; however, modern ML models—including convolutional networks, gradient boosting algorithms, and explainable AI frameworks—can reliably decode the temporal and spectral features of physiological signals in ways that exceed human perception [1].

AI-driven systems enable real-time adaptation of rehabilitation protocols. By integrating EMG patterns, HRV, kinematic parameters, and robotic sensor data, these models can dynamically adjust the task difficulty, feedback intensity, or robotic assistance levels, based on the patient’s current performance [52]. High-density EMG and ML classification suggest that AI can detect pathological co-activation, distinguish fatigued from non-fatigued states, and track incremental neuromuscular improvements throughout the recovery process [52]. In EMG-driven robotic rehabilitation, adaptive controllers have shown the ability to modulate assistance in response to effort fluctuations, improving motor engagement and minimizing compensatory movements [2,26]. The main perspectives of AI applications are summarized in Figure 4.

Beyond immediate motor performance, predictive analytics anticipate long-term recovery outcomes. Machine learning models trained on EMG, HRV, and clinical features have been able to forecast functional gains, identify patients who are at risk of plateauing, and estimate the optimal intensity and timing of therapy. Multi-information fusion approaches, combining EMG with additional physiological markers, yield good potential for predicting session-to-session variability and personalizing rehabilitation trajectories [6]. The ability to predict recovery patterns not only facilitates more efficient planning but also supports adaptive, closed-loop systems that evolve alongside the patient’s physiological state.

Taken together, it can be considered that AI-based signal interpretation and prediction models represent a major advancement toward precision rehabilitation. By transforming raw physiological data into clinically actionable information, these technologies enable intelligent biofeedback paradigms that are responsive, individualized, and capable of accelerating neuroplastic recovery mechanisms [2,52]. The convergence of EMG analytics, HRV regulation, and machine-learning-assisted adaptation points toward a future in which post-stroke therapy becomes increasingly automated, personalized, and grounded in physiological evidence [52,54].

Physiological biofeedback gathers several complementary modalities that together target distinct components of neuromotor and autonomic regulation. HRV biofeedback primarily modulates vagal activity, baroreflex sensitivity, emotional regulation, and attentional control [63,72]. In contrast, EMG biofeedback enables precise voluntary control of skeletal muscle activation, reduces pathological co-contraction, and enhances motor relearning, particularly in the upper limb after stroke [8,9]. ECG monitoring complements these modalities by providing continuous cardiovascular and autonomic surveillance, which is important in stroke patients who frequently display cardiac dysautonomia [58,69].

More recent developments incorporate AI for multimodal data interpretation and therapy personalization (

Table 2. Comparative physiological modalities, computational method (AI), and their targets.

Modality	Primary Physiological Target	Clinical Benefits	Limitations	Optimal Use in Rehabilitation
HRV Biofeedback	Autonomic nervous system (vagal tone, baroreflex)	Improves emotional regulation, stress tolerance, autonomic balance [63,64];	Requires cognitive participation	Pre-session autonomic regulation
EMG Biofeedback	Muscle activation, coactivation	Enhances motor relearning, reduces abnormal synergies [8,9]	Susceptible to noise/artifacts	Task-oriented motor training
ECG Monitoring	Cardiac/autonomic function	Detects dysautonomia, monitors safety [58,69]	Low specificity for motor function	Safety and autonomic tracking
AI-Driven Data Analysis (processing and fusion method)	Multimodal physiological fusion	Personalized adaptation, predictive modeling [1,2,52]	Requires datasets/computation	Closed-loop adaptive rehab

Abbreviations: EMG = electromyography; HRV = heart rate variability; AI = artificial intelligence; and ECG = electrocardiography.

). Machine learning algorithms have been shown to reliably decode complex EMG activation patterns, detect impaired coordination using high-density EMG systems, and adapt robotic assistance in real time using corticomuscular integration models [1,2,52]. Such systems enable the dynamic adjustment of exercise difficulty and feedback precision based on the patient’s physiological state, improving motor engagement and optimizing the rehabilitation outcomes. Although EMG biofeedback remains the predominant type of biofeedback, novel technologies are being explored for their potential as biofeedback instruments, including the emerging applications of VR-based biofeedback and exergaming biofeedback in rehabilitation [54].

Together, these modalities form a complementary framework that is capable of addressing motor, autonomic, cognitive, and emotional impairments that are commonly observed after stroke. The literature up to this point consistently supports a multimodal approach, where HRV biofeedback enhances autonomic readiness, EMG biofeedback improves neuromuscular precision, ECG ensures cardiovascular stability, and AI-driven analytics orchestrate a personalized, real-time adaptation of therapy [1,9,58,72].

We can synthesize the main benefits by implementing AI-driven protocols in the rehabilitation process, as shown in

Table 3. EMG and HRV feedback in stroke patients, enhanced by AI.

EMG + AI	HRV + AI	AI (EMG + HRV)
Myoelectric patterns in stroke patients aiding in diagnosis and prediction of gait rehabilitation enhanced by AI through machine learning techniques [8]	Effects of stress by HRV monitoring and analysis [63]	Improvement of diagnostic accuracy and real-time monitoring capabilities through advanced predictive analytics and telemedicine integration [1]
Trigger robot-assisted training in stroke patients due to residual EMG signals [8,9]	Analyzing ECG data to identify stroke risk patterns and enabling early stroke detection [63,69]	Improved motor control and autonomic regulation with potential for personalization by wearable technology [2]
Enhances conventional therapy for orofacial functions, improving quality of life and neuromuscular patterns, particularly for swallowing [9]	Utilizing machine learning to analyze discomfort associated with rehabilitation [2,6]	Personalized feedback and predictive analytics [1,52]
Enhances proprioception and control of the affected leg in stroke patients [1,8]	Reflects real-time autonomic nervous system state and cardiac health [2,63]	Tailored rehabilitation through adaptive robotic technologies [2,52]
Enabling real-time monitoring and accurate gesture recognition for upper-limb exercises [1,8,9]	Can improve cardiac management and treatment strategies [2,63]	Can improve monitoring and support for stroke patients by integrating various physiological modalities for better state assessment [1]

Abbreviations: EMG = electromyography; HRV = heart rate variability; AI = artificial intelligence; and ECG = electrocardiography.

.

Despite growing evidence supporting the use of HRV biofeedback, EMG-driven rehabilitation, neuromodulation, and AI-assisted monitoring, several implementation challenges limit their widespread clinical adoption [52,54]. First, methodological variability remains a major barrier. HRV requires strict standardization of acquisition and analysis procedures, as highlighted by Catai et al., and improper measurement protocols may lead to inconsistent or non-interpretable autonomic indices. EMG-based interventions face similar issues: electrode placement, signal noise, cross-talk, and patient-specific neuromuscular alterations significantly affect the data quality [21,22,70]. This variability complicates integration into routine physiotherapy workflows and requires adequate clinician training.

A second challenge relates to patient heterogeneity. Stroke survivors present diverse patterns of muscle activation, autonomic imbalance, sensory deficits, and cognitive impairments, making the personalization of biofeedback protocols essential, yet technically demanding [57,59]. For example, individuals with severe autonomic dysfunction demonstrate attenuated HRV responses, while those with extensive neuromuscular impairment may show minimal voluntary EMG activity [28,68]. Such variability underscores the need for adaptive systems that are capable of adjusting feedback intensity, electrode configuration, and task complexity in real time.

A third obstacle concerns technological accessibility and usability. Several studies highlight the difficulty of translating sophisticated biofeedback systems into home-based or telerehabilitation settings. Donnelly et al. document barriers such as interface complexity, sensor discomfort, unreliable connectivity, and reduced confidence in technology use among stroke survivors [13,18]. Wearable sensors and consumer-level devices—while increasingly available—still require improvements in accuracy, comfort, and robustness before being clinically dependable [5].

When discussing the reliability and signal acquisition concerning the investigations described, it is worth mentioning that although the surface EMG is widely recognized for its utility in monitoring voluntary muscle activity, its reliability can be affected by the underlying muscle atrophy and skin–electrode impedance issues. High-density surface EMG and standardized electrode placement protocols improve repeatability; however, in chronic stroke, deep muscle analysis may require needle EMG for more accurate motor unit characterization [21]. The reliability of HRV as a surrogate for autonomic function has been established in both general and stroke populations, provided recordings are artifact-free, of sufficient length (≥5 min), and not confounded by arrhythmias. In this context, the validity is highest when methodological standards (e.g., resting supine, controlled breathing, exclusion of atrial fibrillation) are strictly followed [68].

EMG biofeedback is a valid method for improving Fugl-Meyer Assessment (FMA) scores, particularly in the spastic and recovery phases, and HRV has been shown to correlate with post-stroke fatigue, cognitive outcomes, and functional status; nonetheless, its prognostic value depends on standardized acquisition [8,68]. The monitoring of EMG, ECG, and HRV in stroke rehabilitation has also evolved significantly over recent decades, driven by advances in sensor technology, wireless communication, and signal processing. Recent years have seen the emergence of wireless surface EMG systems, which allow for unobtrusive, real-time monitoring during both in-clinic and facilitated home-based rehabilitation [52]. Wireless EMG sensors now offer high-density, multi-channel capabilities and improved signal fidelity, even during dynamic movements, enabling seamless integration with wearable garments and smart rehabilitation devices [5]. The advent of portable, single-lead ECG devices and wearable photoplethysmography (PPG)-based monitors has revolutionized cardiac and autonomic assessment. These devices support continuous, ambulatory monitoring, and are capable of transmitting data wirelessly for remote analysis and telehealth applications. Key improvements include enhanced artifact rejection, cloud-based data storage, and integration with AI algorithms for automated HRV analysis [5,71].

For all that technological progress, ethical considerations still influence the implementation of AI-assisted biofeedback. AI-driven analysis of HRV, EMG, and ECG generates sensitive physiological data that require strict compliance with privacy and data-security standards. Future clinic protocols will be necessary for AI-based algorithms’ predictions to ensure safe clinical decision-making [1]. Moreover, the potential for algorithmic bias, arising from inadequate training datasets or underrepresented subpopulations, poses a risk of unequal treatment outcomes [73,74].

Equitable access also represents an important ethical concern. Digital rehabilitation tools and biofeedback devices must be designed to accommodate a wide range of cognitive, sensory, and physical abilities, with user interfaces that are intuitive for stroke survivors who may have visual, motor, or language impairments. Also, advanced neuromodulatory interventions, robotic systems, or AI-enabled platforms may only be available in specialized centers, potentially widening healthcare disparities [73,74]. Successful implementation will require cost-effective solutions, interdisciplinary training, standardized clinical pathways, and integration with existing rehabilitation infrastructure [73].

Patient consent is another critical issue, particularly as digital systems increasingly collect, analyze, and share real-time data. Transparent communication about data usage, potential risks, storage duration, and third-party access is essential for informed consent and patient trust. Dynamic consent models may allow patients to update their preferences as technology or their circumstances evolve. Ensuring robust data privacy and security is also relevant, as health records and behavioral data are highly sensitive. Strong encryption, anonymization, and secure data storage protocols will be necessary [74].

While highlighting the reported benefits of EMG- and HRV-based biofeedback in post-stroke rehabilitation, it is important to acknowledge that much of the current evidence is still associative up to this point, rather than strictly causative. There are indeed correlations between physiological signal modulation and functional improvement; however, the underlying mechanisms and direct causal pathways remain incompletely established. Furthermore, the existing body of literature is characterized by small sample sizes, heterogeneous patient populations, and considerable variation in intervention protocols, outcome measures, and follow-up durations. This methodological diversity complicates the interpretation and generalization of the findings and may contribute to inconsistent or inconclusive results across studies. Large-scale, well-controlled randomized trials and standardized research methodologies are needed to substantiate causality and, ultimately, refine clinical protocols.

Although the integration of AI-driven analytics and multimodal feedback systems in post-stroke rehabilitation is highly promising, it is also important to recognize that the vast majority of these technologies remain at the pilot stage. Most published studies to date are either limited to preliminary trials, or small sample sizes, single-center cohorts, or short-term usability. Until further data are available, the clinical adoption of AI-enhanced, multimodal rehabilitation platforms should be approached as being experimental and supplemental to the established standard-of-care protocols.

7. Future Perspectives

The integration of autonomic, neuromuscular, and neurocognitive markers is expected to redefine post-stroke rehabilitation in the coming decade. Biofeedback-based interventions are progressively evolving from single-modality protocols toward multimodal, synchronized therapeutic approaches, in which HRV modulation, EMG-driven motor retraining, and neuromodulation techniques such as auricular or cervical vagus nerve stimulation operate within unified frameworks. Vagus nerve stimulation paired with motor practice can potentiate neuroplasticity and accelerate upper-limb functional gains, and future systems may integrate HRV-guided pacing to optimize autonomic readiness prior to stimulation sessions [36,37,38].

Advances in AI and physiological signal fusion represent another major direction. Machine learning models that are capable of decoding subtle EMG alterations [1,52] and predicting fatigue, effort variance, or motor intent open the possibility for continuous adaptive rehabilitation [2,6]. Such systems will enable closed-loop controllers that adjust the task difficulty, robotic assistance, or biofeedback intensity based on real-time HRV, EMG, and kinematic signatures. The growing availability of wearable sensors and smart textiles [5] further supports the long-term vision of home-based, high-frequency, AI-supervised rehabilitation. Overcoming integration and validation challenges, developing and testing truly adaptive and personalized feedback systems, and establishing standardized protocols for the real-world deployment of multimodal, wearable rehabilitation technologies for stroke survivors are still desired aspects to be fulfilled in the years to come.

Another emerging area involves telerehabilitation and remote physiological monitoring. Early studies report good acceptability of remote biofeedback systems among stroke survivors [13,18], though further work is required to optimize usability, data security, and adherence. In future practice, cloud-based platforms may integrate HRV trends, EMG activity maps, and performance metrics to guide clinicians in real time, enabling individualized, longitudinal care that is independent of geographic constraints.

From a neuroscientific standpoint, future research should explore the relationships between HRV, brain connectivity, and neuroplastic changes using advanced imaging techniques. Recent findings linking autonomic indices to functional connectivity in insular–prefrontal networks [62] highlight the potential of autonomic biomarkers as predictors of cortical reorganization. Similarly, studies demonstrating structural and peripheral axonal plasticity after stroke suggest that combining EMG- and HRV-based biofeedback with neuromodulation may unlock synergistic mechanisms of recovery that are not yet fully understood [33].

Nonetheless, there is a pressing need for standardized clinical pathways, including consensus on HRV assessment, EMG biofeedback dosages, integration with robotic protocols, and long-term follow-up strategies [70]. Large-scale randomized trials, cost-effectiveness analyses, and implementation science frameworks will be necessary to transition advanced biofeedback systems from specialized research centers into routine clinical care.