Advancing fNIRS-Driven Stroke Recovery: From Quantitative Assessment to Closed-Loop Interventions

Abstract

Stroke is a cerebrovascular disease with high disability rates, necessitating effective strategies for functional assessment and rehabilitation. However, current research and clinical workflows still lack a closed-loop framework linking objective assessment to adaptive intervention. Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique that monitors cerebral hemodynamics and provides practical biomarkers for evaluating post-stroke impairment and guiding rehabilitation. This review systematically summarizes recent fNIRS-based stroke research across three dimensions: assessment paradigms, analytical methodologies, and intervention strategies. This review first addresses the rationale for selecting regions of interest and designing experimental paradigms to support greater standardization of stroke assessment. It then summarizes widely used analytical methods and quantitative indices, with emphasis on how different approaches capture local activation, inter-regional connectivity, and network-level reorganization. Finally, it examines the use of fNIRS in evaluating peripheral and central interventions, highlighting its potential to objectively monitor therapeutic efficacy and clarify the neural mechanisms underlying functional recovery. fNIRS provides a promising bridge between quantitative assessment and adaptive rehabilitation in stroke. Continued progress will depend on improving methodological standardization, deepening multimodal and network-level analysis, and integrating into closed-loop systems.

Keywords

Functional near-infrared spectroscopy / Stroke / Assessment / Intervention / Rehabilitation

1 Introduction

Stroke remains a leading cause of death and long-term disability worldwide [1]. Survivors often show motor [2], cognitive [3], and language impairments [4]. Although rehabilitation assessment still relies mainly on clinical scales that lack objective physiological markers for precise, individualized evaluation [5]. In addition, many treatment protocols use fixed parameters and do not adapt to the patient’s changing neural state. Objective monitoring tools are therefore needed to support individualized rehabilitation.

Neuroimaging is an important approach for post-stroke functional reorganization [6]. Functional near-infrared spectroscopy (fNIRS) measures relative changes in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) through the modified Beer–Lambert Law [7]. Compared with functional magnetic resonance imaging (fMRI) and electroencephalography (EEG), fNIRS is more portable, relatively motion tolerant, and well suited to monitoring superficial cortical metabolic responses during rehabilitation [8,9]. Because stroke is a vascular disease, this link to neurovascular activity is especially relevant clinically [10].

Several issues still limit broader clinical use of fNIRS. Studies vary widely in assessment paradigms, analytical methods, and uses of fNIRS within closed-loop interventions. In experimental design, substantial heterogeneity remains. Although models such as inter-hemispheric competition stress the balance between affected and unaffected hemispheres [11], no broad consensus on regions of interest (ROIs) selection for stroke assessment. Channel placement and cortical coverage therefore differ markedly across studies [12], and inconsistent probe-task mapping hampers baseline comparison and standard development.

Similar limitations appear in data analysis. Most fNIRS studies still use the general linear model (GLM) to quantify regional activation, which can’t fully exploit the physiological information in the signals [13]. Functional connectivity (FC) [14] and brain network analysis [15] have characterized inter-regional interactions, but they are often built on static Pearson correlation alone, with limited attention to dynamic coupling. More robust analytical frameworks are needed to capture the high-dimensional features of post-stroke neural reorganization.

Intervention studies show similar breadth and complexity. Stroke rehabilitation includes peripheral approaches such as robotic exoskeleton training [16] and electrical stimulation [17], as well as central neuromodulation methods such as transcranial magnetic stimulation (TMS) and transcranial direct current stimulation (tDCS) [18]. fNIRS can objectively read out the hemodynamic responses evoked by these interventions. However, the literature still lacks a clear synthesis of their physiological differences. A more systematic review is needed to organize these pathways and clarify how fNIRS can support closed-loop rehabilitation strategies.

This review examines how fNIRS-derived information is used in post-stroke rehabilitation. Here, fNIRS-driven describes a progression from quantitative assessment and treatment monitoring during recovery to the potential use of validated markers as feedback variables. This review proceeds from the influence of ROI and task design on signal acquisition, through the conversion of signals into quantitative markers, to the interpretation of intervention responses and their potential role in adaptive rehabilitation.

2 Fundamentals of ROI Selection and Experimental Paradigms

Reliable functional assessment depends on aligning anatomical targets with task design. Current fNIRS stroke studies still vary substantially in sensor layout and experimental paradigm, limiting reproducibility and cross-study comparison [19]. This section links the anatomical rationale for ROI selection with the sensorimotor, cognitive and language, and non-execution paradigms used to assess post-stroke function.

2.1 Rationale for ROI selection

The cerebral cortex contains functionally distinct regions, commonly described with Brodmann’s cytoarchitectural map [20], as shown in Fig. 1A. For fNIRS studies, these anatomical boundaries must be translated into practical sensor layouts. Because optode placement usually follows the international 10-10 system, scalp landmarks must be mapped accurately to the underlying cortex [21]. Fig. 1B illustrates this correspondence, and Table 1 links the cortical regions addressed below to their functional roles in post-stroke assessment.

The anterior frontal regions, particularly the frontopolar cortex (FPC) and dorsolateral prefrontal cortex (DLPFC), are closely linked to executive functions such as working memory, decision-making, and attentional control, and are therefore commonly targeted in cognitively demanding paradigms [15,22]. The posterior frontal regions form the main cortical network for motor control. The premotor cortex (PMC) contributes to motor planning and movement sequencing, the supplementary motor area (SMA) is involved in movement initiation and coordination, and primary motor cortex (M1) is most directly related to motor execution. Motor assessment studies often assess all these regions [23,24]. The parietal cortex integrates somatic sensory input, including touch, pain, and proprioception. Somatosensory cortex (S1) is a common target in sensory dysfunction [25]. Since S1 and M1 are anatomically close and functionally related, many studies refer to them together as the sensorimotor cortex (SMC) [26]. This region also follows a clear somatotopic arrangement, with lower-limb and foot representations located around C1/C2 [27] and upper-limb and hand representations around C3/C4 [28].

Language assessment mainly targets the temporal lobe and adjacent cortical regions, including Broca’s area for speech production and Wernicke’s area for semantic comprehension [19], which usually show clear left-hemispheric lateralization during language tasks [29,30]. The occipital lobe is the main cortical region for visual processing. It is less frequently used as a primary target, but it is often included in paradigms involving visual stimulation or visual feedback [31].

2.2 Design of experimental paradigms

Because the hemodynamic response is delayed, most fNIRS studies use block designs alternating task and rest to quantify relative hemoglobin changes. Task blocks commonly last about 10–30 s and are followed by rest periods of about 20–40 s, allowing the HbO/HbR response to rise and partially return toward baseline. For post-stroke assessment, the paradigms considered below are organized according to the function being probed and the feasibility of overt task performance: sensorimotor, cognitive and language, and non-execution. Fig. 2 illustrates these categories. Task design should nevertheless be matched to the clinical or experimental question. For example, Formica et al. [32] combined gripping, tapping, and tactile exploration (Fig. 2D) to evaluate the multidimensional effects of mirror therapy.

2.2.1 Sensorimotor paradigm

Motor assessment paradigms mainly assess hand function, upper-limb movement, gait, balance, and swallowing. Hand function is commonly assessed with three tasks. Grip task (Fig. 2A [23]) relates grip strength to cortical activation [33]. Finger tapping (Fig. 2B [34]) probes fine motor control and finger individuation. Positioning tasks (Fig. 2C [35]) require precise object placement and are often used to assess hand-eye coordination and spatial planning. Upper-limb paradigms mainly use three movement types. Single-joint tasks, such as isolated elbow flexion (Fig. 2G [36]), isolate specific muscle groups. Task-oriented training (Fig. 2H [37]) emphasizes coordinated, goal-directed multi-joint movement and is widely used because it better reflects activities of daily living [38–41]. Bilateral arm training (Fig. 2F [21]) examines inter-limb coordination and inter-hemispheric interaction during synchronous or alternating movement.

Locomotor function is usually assessed with walking tasks performed over a set duration or distance. Common options include overground walking (Fig. 2E [27]) and treadmill walking (Fig. 2I [42]). Comparative studies suggest that both can produce similar cortical activation patterns [43]. Balance is often examined with two simple paradigms. The eyes-open/close task (Fig. 2J [13]) probes stability under changing visual input, whereas the sit-to-stand task (Fig. 2K [44]) evaluates postural control and dynamic weight shifting during transitional movement.

Swallowing tasks are comparatively simple and usually involve swallowing saliva or liquid (Fig. 2L [45]). Some recent studies have added sensory modulation, such as sour taste stimulation, which appears to increase activation in the DLPFC, PMC, and S1, although its clinical value still needs further validation [25].

2.2.2 Cognitive and language paradigm

Cognitive function is commonly evaluated with three paradigms. The verbal fluency task (VFT) (Fig. 2O [46]) requires word generation under semantic or phonemic constraints and is often used to probe lexical access and executive control. The repetition task (Fig. 2P [30]) assesses the phonological loop and working memory. The sorting task (Fig. 2M [22]) examines cognitive flexibility by asking participants to classify items according to predefined rules. In post-stroke cognitive impairment (PSCI), the hemodynamic responses evoked by these tasks are often compared with standardized scales such as the MoCA and MMSE [46–49].

An important extension of stroke assessment is the dual-task paradigm (Fig. 2N [50]), which combines motor and cognitive demands to quantify cognitive-motor interference. Compared with single-task conditions, dual-task performance usually requires greater prefrontal recruitment, which likely reflects the added cost of divided attention [51]. Since fNIRS can be used during natural movement, it is well suited to dual-task assessments involving walking, balance, or other rehabilitation-related activities.

Assessment of post-stroke aphasia (PSA) overlaps partly with cognitive testing, since tasks such as VFT and repetition can also probe lexico-semantic processing. At the same time, more language-specific tasks, especially confrontation naming (Fig. 2Q [30]), are used to isolate lexical retrieval deficits. Together, these paradigms provide a broader view of speech production, comprehension, and semantic access [52].

2.2.3 Non-execution paradigm

Non-execution paradigms are especially useful for patients with severe motor deficits or flaccid paralysis. Resting-state measurement provides a passive option for patients unable to complete voluntary tasks by focusing on spontaneous hemodynamic fluctuations, intrinsic FC, and baseline neurovascular status [53]. Motor imagery (MI) offers another alternative, recruiting networks that partly overlap with those involved in actual movement [54].

3 Analytical Frameworks and Metrics for Assessment

Quantitative assessment links fNIRS measurement with rehabilitation decision-making. This section reviews the main fNIRS frameworks used in the field, including the GLM, laterality index (LI), functional and effective connectivity (FC/EC), graph-theoretical analysis, and multimodal coupling. Together, these measures support the evaluation of functional status and treatment response and, after sufficient validation, may also provide state variables for adaptive rehabilitation.

3.1 General linear model (GLM)

GLM is the standard framework for decomposing continuous hemodynamic time series into task-related components. In this model, the observed signal Y is represented as a linear combination of explanatory variables X plus an error term \epsilon:

(1)Y=X\beta +\epsilon

where Y represents the time-series vector of hemoglobin concentration changes; X denotes the design matrix, constructed by convolving the experimental event timing with a canonical HRF; \epsilon is the residual noise; and \beta represents the magnitude of the hemodynamic response under each condition.

Longitudinal changes in \beta can be used to track neuroplasticity. For example, increased ipsilesional \beta values during motor or cognitive tasks are often associated with better functional scores and may reflect recovery-related recruitment of neural resources [32,35]. GLM-based measures have been used in studies of motor execution, swallowing [45], and language recovery [49,52]. It can also be relevant for closed-loop paradigms that adjust intervention intensity according to cortical responses. Standard implementations are available in widely used toolboxes such as Homer3 [55] and MNE-NIRS [56].

3.2 Laterality index (LI)

LI is a quantitative metric designed to assess the functional asymmetry between homologous brain regions, serving as a proxy for inter-hemispheric balance. It is typically calculated as:

(2)\text{LI}=\frac{A_{\text{left}}-A_{\text{right}}}{A_{\text{left}}+A_{\text{right}}}

where A represents the magnitude of the hemodynamic response within the ROI, quantified as mean HbO concentration or GLM \beta-values. The index ranges from -1 to 1, with positive values indicating left-hemispheric dominance and values near zero reflecting bilateral symmetry.

Many patients show a shift toward the contralesional hemisphere, either because of compensatory recruitment or reduced transcallosal inhibition [57]. Over time, a return toward greater symmetry is often interpreted as a sign of adaptive reorganization and functional recovery. Some studies calculate LI using GLM coefficients [58], wavelet-based amplitude or phase features [59], making it possible to quantify inter-hemispheric balance over time.

3.3 Functional and effective connectivity (FC/EC)

Connectivity analysis is central to studying post-stroke network reorganization. FC describes statistical dependence between regions and is usually interpreted as undirected communication. Clinically, reduced inter-hemispheric FC may indicate impaired communication between bilateral motor or cognitive networks, whereas increased contralesional or frontoparietal connectivity may reflect compensatory recruitment during recovery. EC, by contrast, aims to infer the direction and causal structure of information flow from ipsilesional motor regions to downstream networks is weakened or whether alternative contralesional pathways are being recruited. Although both are commonly represented as connectivity matrices, they capture different aspects of neural organization, as summarized in Table 2.

Within FC analysis, Pearson correlation remains the most common measure of linear synchronization between channels. Fisher’s Z-transformation is often applied before group analysis. Time-frequency methods such as wavelet phase coherence and phase locking value provide complementary information about oscillatory synchronization. In stroke studies, reduced inter-hemispheric FC is a common finding and is often interpreted as network disruption.

For EC analysis, Granger causality remains a common starting point. Under this framework, one signal is said to exert directional influence if its past values improve prediction of another signal. Partial directed coherence extends this idea into the frequency domain. Dynamic Bayesian inference is used when time-varying causal interactions are of interest. Clinically, these measures help describe how stroke disrupts large-scale brain organization and whether alternative contralesional pathways are being recruited during recovery.

3.4 Brain network analysis

Brain network analysis describes the topology of stroke-related network reorganization by treating channels or ROIs as nodes and functional or effective connections as edges. After thresholding the connectivity matrix, measures of network integration, segregation, and nodal importance can be extracted from the resulting sparse network, although their values may depend on the thresholding strategy. The graph-theoretical metrics used for these purposes are summarized in Table 3.

Global network metrics mainly describe how efficiently information can be exchanged across the whole brain. Global efficiency reflects long-range integration and parallel information transfer, and reductions in this measure are often interpreted as evidence of impaired large-scale coordination after stroke [80]. Local efficiency and the clustering coefficient describe local segregation and fault tolerance, and therefore capture the density of local information processing [81]. Small-worldness reflects the balance between integration and segregation relative to a random network. When this balance is altered after stroke, the network may become less efficient overall.

Nodal centrality measures identify network hubs. Betweenness centrality captures how often a node lies on shortest paths between other nodes and therefore indicates potential bridge regions. Eigenvector centrality weights a node according to the importance of the nodes to which it connects. Tracking these measures over time can help identify compensatory hubs in contralesional or perilesional regions.

Together, these metrics extend assessment beyond local activation to whole-network organization. In stroke rehabilitation, reduced global efficiency may indicate impaired integration, whereas shifts in local topology or nodal centrality may help identify adaptive reorganization in contralesional or perilesional regions.

3.5 Multimodal coupling analysis

Multimodal physiological coupling analysis combines fNIRS with electrophysiological or muscular signals to examine coordinated activity across physiological systems. fNIRS has also been integrated with other imaging and physiological techniques; within the post-stroke rehabilitation literature synthesized here, the coupling evidence is concentrated in two frameworks: neurovascular coupling (NVC) assessed with fNIRS and EEG, and corticomuscular coupling (CMC) assessed with fNIRS and surface electromyography (sEMG). Accurate temporal synchronization is essential in both cases because the signals arise from different physiological processes and operate on different time scales [9,82].

3.5.1 fNIRS-EEG integration for neurovascular coupling (NVC)

NVC describes the relationship between neuronal electrical activity and the accompanying hemodynamic response by integrating EEG and fNIRS [83]. Because the two modalities are closely linked in time [26], this combination is well suited to quantifying NVC and examining how neural activity relates to vascular response [84].

More recent work has extended this approach. Lin et al. used cross-layer EEG-fNIRS connections to derive NVC-related measures of neural reorganization [85]. fNIRS-informed EEG source imaging is used to improve source localization by introducing hemodynamic priors [86]. In addition, Li et al. combined multimodal signals with a linear time-invariant model to study brain network controllability [23]. These approaches may be useful for interventions that need real-time estimates of cortical state, including non-invasive brain stimulation and brain-computer interfaces (BCIs).

3.5.2 fNIRS-sEMG integration for corticomuscular coupling (CMC)

CMC reflects the functional interaction between cortical motor hemodynamics and peripheral muscle activity, which allows central-peripheral relationships to be quantified during motor tasks [21]. Task-based studies suggest that some stroke-related impairments may reflect reduced central-peripheral transmission efficiency rather than complete loss of motor pattern organization [87,88].

To quantify this transmission, transfer entropy (TE) is widely favored over traditional coherence methods [89,90]. Clinical applications show that TE strength from the ipsilesional motor cortex to the paretic limb is significantly reduced in stroke, reflecting weakened central drive and impaired information transfer between the central and peripheral nervous systems [79,91].

By quantifying cortex-muscle interaction, CMC measures such as TE provide a practical way to assess the descending motor pathway. They may also help drive peripheral closed-loop interventions, for example by adjusting exoskeleton assistance or functional electrical stimulation (FES) timing according to the patient’s current CMC state.

4 fNIRS-Informed Rehabilitation and Interventions after Stroke

Rehabilitation promotes stroke recovery through repeated, task-oriented practice that stimulates neural activity and supports circuit reorganization [43,66,92]. In relation to their role in the rehabilitation pathway, the interventions examined below are organized as peripheral interventions and stimulation, non-invasive brain stimulation, and feedback-based therapies, as summarized in Fig. 3. Across these groups, fNIRS is used mainly to measure treatment-related cortical changes; it enters the control loop only when its features trigger or adjust feedback. Before interpreting stimulation-induced fNIRS responses, intervention paradigms should generally include sham or control conditions to reduce potential confounding from non-specific cortical, peripheral vascular, or scalp-related hemodynamic responses.

4.1 fNIRS assessment of peripheral rehabilitation interventions

4.1.1 Robot-assisted exoskeleton rehabilitation

Robot-assisted exoskeleton rehabilitation is widely used in stroke rehabilitation because it can provide intensive, standardized, and quantifiable sensorimotor training [16]. Although robotic assistance may initially increase cognitive load [93], it can also provide consistent proprioceptive input that supports functional recovery. fNIRS studies generally show that exoskeleton training increases cortical activation in the affected hemisphere and supports functional reorganization [94–96].

One advantage of exoskeleton systems is that training mode can be adjusted to different impairments [71]. In bimanual paradigms, the unaffected hand can assist the paretic hand through a rehabilitation glove, and fNIRS studies suggest that this may enhance inter-hemispheric coupling [97]. Lower-limb exoskeleton-assisted gait training recruits the PMC and SMA [98], while static isotonic upper-limb robot training may produce broader cortical activation than dynamic training [99].

Training intensity should also match impairment severity. Assisted or passive modes are often necessary for severe deficits [73], whereas active participation is preferred for mild or moderate impairment. fNIRS studies show that voluntary effort tends to evoke stronger cortical activation and is generally associated with better rehabilitation outcomes [100]. For patients who cannot perform overt movement, MI can evoke activation patterns similar to active movement [36].

4.1.2 Electrical stimulation

Neuromuscular electrical stimulation (NMES) uses surface electrodes to activate motor nerves and induce muscle contraction, thereby increasing proprioceptive input. fNIRS studies suggest that this input can alter cortical excitability and connectivity. In PSA, median-NMES can enhance FC in language-related networks [72]. In closed-loop designs, stimulation of the paretic hand can be triggered by activity of the unaffected hand, which may further strengthen inter-hemispheric coupling, particularly in the SMC [63].

FES can coordinate muscle contractions to restore functional movements such as gait. fNIRS reveals that stroke patients engaging in FES-assisted walking require the compensatory recruitment of the SMC–SMA–PMC cortical network [27]. However, the hemodynamic response varies across protocols; while some enhance activation in the contralateral SMA and PMC, others may reduce M1 activation due to improved neural efficiency [101]. These fNIRS findings support fNIRS-based optimization of stimulation parameters and future protocols that adapt FES or NMES intensity to cortical state.

4.1.3 Acupuncture-based interventions

Acupuncture is also used as an adjunct to stroke rehabilitation. fNIRS studies suggest that specific acupoints can evoke measurable cortical hemodynamic responses [75]. In PSCI, scalp acupuncture is associated with frontal and parietal activation and improved MoCA scores [102]. In PSA, stimulation over the projection of Broca’s area has been linked to local activation and language improvement [103]. In motor recovery, both body and auricular acupuncture are associated with increased M1 activation [76,104], while electro-acupuncture may strengthen PMC and in swallowing-related networks [105]. These findings suggest that fNIRS may be useful for monitoring response to acupuncture-based interventions.

4.2 fNIRS monitoring of non-invasive brain stimulation

4.2.1 Transcranial direct current stimulation (tDCS)

tDCS modulates cortical excitability by applying weak currents that alter resting membrane potentials. Anodal stimulation generally increases excitability, whereas cathodal stimulation reduces it [106]. fNIRS can track the associated hemodynamic response over time. Longitudinal studies suggest that cortical activation may rise early and then decline, with stable after-effects appearing after sufficient stimulation [74].

fNIRS has also been used to assess the network-level effects of specific tDCS protocols. In PSCI, anodal stimulation over the DLPFC at 2 mA for 30 minutes has been associated with increased activation in the left superior temporal gyrus and stronger FC, alongside improved cognitive performance [46]. To improve spatial focality, high-definition tDCS (HD-tDCS) uses a multi-electrode array. Studies monitored with fNIRS suggest that HD-tDCS over the ipsilesional PMC at 1 mA for 20 minutes can alter local hemoglobin concentrations and may help restore inter-hemispheric balance [34].

These findings indicate that tDCS affects not only local activation but also broader functional networks. They also suggest that fNIRS monitoring may help refine stimulation parameters in future adaptive or closed-loop protocols.

4.2.2 Transcranial magnetic stimulation (TMS)

Repetitive transcranial magnetic stimulation (rTMS) modulates neuronal activity through electromagnetic induction, and its effects depend on stimulation frequency. Low-frequency stimulation (below 1 Hz) is generally inhibitory, whereas high-frequency stimulation (above 5 Hz) is excitatory [107]. fNIRS has been used to characterize these responses and to support individualized targeting [108]. In motor recovery, high-frequency stimulation of ipsilesional M1, especially with bilateral arm training, has been associated with increased local hemoglobin concentration and improved inter-hemispheric balance [69,78]. Conversely, low-frequency stimulation of contralesional M1 may suppress maladaptive hyperactivation [85]. Individual responses remain variable and may depend on neurovascular characteristics [70].

Outside motor rehabilitation, fNIRS has also been used to study cognitive and language networks. High-frequency rTMS over the left DLPFC has been linked to stronger frontoparietal connectivity and improved working memory [109], whereas stimulation of Broca’s area may recruit language-related networks in aphasia [19,29].

Intermittent theta-burst stimulation (iTBS) can induce plastic changes within a shorter period [64]. fNIRS studies suggest that iTBS produces activation patterns similar to high-frequency rTMS but with greater time efficiency. fNIRS has also been used to compare stimulation schemes such as combined cerebellum–SMA stimulation for balance-related networks [67]. These results support fNIRS monitoring during protocol optimization.

4.3 fNIRS in feedback-based rehabilitation

4.3.1 Visual feedback

Visual feedback forms a behavioral loop in which patients adjust movement using visual information about task performance. In most fNIRS studies, this feedback is not generated from fNIRS signals; fNIRS instead measures the accompanying cortical response. Increased ipsilesional M1 activation and stronger connectivity with prefrontal regions suggest recruitment of motor, attentional, and error-monitoring processes [61]. Visual context may also modulate these responses [40], but fNIRS-derived markers are rarely used to adapt feedback online.

4.3.2 Mirror therapy

Mirror therapy uses the visual illusion of paretic limb movement to compensate partly for impaired proprioception [32]. fNIRS studies show that observing mirrored movement can evoke cortical responses resembling active execution [110], particularly in ipsilesional M1 and may help restore inter-hemispheric balance without paretic-limb movement [58]. Combining mirror therapy with exoskeleton assistance may produce stronger cortical effects than passive mobilization [31].

4.3.3 Brain-computer interface (BCI)

BCI represents the pinnacle of central-intervention strategies, decoding MI signals to drive external assistive devices. By bypassing damaged efferent pathways to mechanically execute the intended movement, BCI reinforces the “intention-action” link via Hebbian learning [111].

Within BCI-based rehabilitation, fNIRS can assess training effects or serve as a signal source. fNIRS studies indicate that BCI-triggered training often produces stronger and more focused cortical activation than open-loop MI or passive training, together with better motor scores [112]. As a signal source, fNIRS-based BCI uses hemodynamic features such as the slope and amplitude of HbO signals to classify movement intention [54]. Some studies suggest that successful BCI use is accompanied by strengthening of frontopolar control networks [49], implying that functional gains reflect genuine network reorganization rather than only peripheral adaptation. However, existing fNIRS-based BCI studies have rarely focused on real-time closed-loop decoding because of the intrinsic hemodynamic delay of fNIRS.

4.3.4 Muscle-computer interface (MCI)

In MCI, sEMG closes the feedback loop by converting muscle activity into visual or tactile cues. fNIRS assesses the associated cortical response. Combined fNIRS-MCI studies suggest that accurate biofeedback can increase contralateral M1 and PFC activation, linking peripheral training with central motor planning [91]. Longitudinal work also report improved inter-hemispheric balance and reduced antagonist co-contraction [113]. Inaccurate feedback may instead dampen cortical activation and interfere with learning [114], emphasizing the need for reliable signal processing.

5 Discussion

5.1 Methodological considerations for fNIRS-based assessment

The results of fNIRS are significantly influenced by the selection before signal analysis. ROI selection determines the cortical area for sampling, while task design determines the functional process represented by the hemodynamic response. Motor tasks may recruit different cortical regions depending on movement complexity, effort, and compensatory behavior, so the same experimental paradigm may provide different information across patients. Resting-state and MI paradigms can be extended to patients who cannot perform explicit movement, but their results describe different aspects of cortical function.

The analytical methods reviewed in this work also provide related but non-equivalent information. GLM describes regional task responses, whereas LI summarizes interhemispheric asymmetry. FC/EC and brain network analysis focus on interactions and brain organization. An increase in cortical activation doesn’t necessarily indicate better recovery, as this may reflect additional neural recruitment, greater task effort, or inefficient processing. Similarly, a shift toward hemispheric symmetry may be consistent with functional reorganization, but its significance depends on lesion location, recovery stage, and task performance. Therefore, when multiple indicators present a consistent pattern and imaging changes are examined simultaneously with behavioral performance, the results will be more reliable.

Methodological standardization remains necessary for comparison across studies. Common procedures for optode localization, channel-quality assessment, preprocessing, and task reporting can reduce avoidable variability. At the same time, differences in head anatomy, lesion distribution, and residual function may require individualized cortical coverage or task difficulty. Such adjustments should be predetermined and recorded rather than introduced after examination of the results. For longitudinal studies, consistent probe placement and task execution are particularly important, because small differences between sessions may resemble physiological change. Using the same subject design and repeated measurements can improve interpretation, but their value depends on whether the recorded channels sample comparable cortical regions over time.

5.2 Clinical implications of fNIRS in stroke rehabilitation

In stroke rehabilitation, fNIRS is mainly used to assess treatment-related cortical change. Across robot-assisted training, electrical stimulation, acupuncture, non-invasive brain stimulation, and feedback-based therapies, it can show whether an intervention is accompanied by changes in regional activation, hemispheric balance, or FC. This information may help explain why patients with similar behavioral improvement exhibit different patterns of cortical reorganization. It may also identify physiological changes that are not fully represented by a single clinical scale. However, the results of fNIRS should be regarded as complementary to, rather than a replacement for, established clinical assessments. Cortical responses alone cannot establish whether an intervention has produced a clinically meaningful therapeutic effect.

Several practical features support its use in rehabilitation. fNIRS is portable, non-invasive, relatively tolerant of motion, and is suitable for upper-limb training, gait, balance tasks, and robotic assistance. During the rehabilitation process, measurements can be repeated without being limited by space and operational procedures as in fMRI. Compared with EEG, fNIRS provides spatial information about cortical hemodynamics and is less affected by electromagnetic interference, which facilitates measurements during electrical or magnetic stimulation. These features allow researchers to study cortical responses under conditions closer to conventional treatment. However, relative movement tolerance cannot eliminate motion artifacts, and higher practicality alone is not sufficient to ensure that the measurement results have clinical significance.

The significance of fNIRS changes depends on its relationship with patient characteristics and functional outcomes. Increased activation after training may indicate an enhanced recruitment of the affected hemisphere, which may also indicate increased effort. Reduced activation may represent limited engagement or improved neural efficiency. These situations cannot be differentiated solely based on hemodynamic data. Therefore, the characteristics of the lesion, impairment severity, time after stroke, task completion, and changes in clinical scores should be considered together. Currently, most research evidence is based on group comparisons, while rehabilitation decisions involve individual patients. More consistent associations between fNIRS measures and clinically meaningful change are needed before these measures can support patient-level treatment selection or adjustment. At present, the most appropriate use is to strengthen treatment monitoring and physiological interpretation alongside established clinical outcomes.

5.3 Challenges and future directions

Technical and physiological factors limit the interpretation of fNIRS. The signal is mainly sensitive to superficial cortical tissue, while the effects related to deep lesions are indirectly manifested through changes in connected cortical regions. Stroke-related vascular damage and altered NVC may change the relationship between neural activity and HbO or HbR responses [7,84]. Dynamic tasks introduce motion artifacts, optode displacement, scalp blood-flow changes, and systemic physiological fluctuations. Therefore, reliable longitudinal measurement requires stable channel configurations, reproducible optode placement, documented individual adjustments, and predefined signal-quality criteria [14]. Sham or control conditions are also needed in intervention studies because observed responses may arise from task repetition, sensory input, attention, or spontaneous variation rather than the intended treatment.

Multimodal measurement may help separate these physiological components. EEG-fNIRS integration links electrical activity with hemodynamic response and provides a basis for assessing NVC [115]. fNIRS-sEMG integration relates cortical hemodynamics to peripheral muscle activity and can characterize changes in CMC [116]. These methods may help distinguish changes in cortical recruitment function from impaired vascular responses or low central-peripheral conduction efficiency. However, their value depends on accurate synchronization, analysis methods suited to different time scales, and maintaining stable coupling metrics in multiple experiments. Therefore, multimodal signals should be integrated around specific physiological problems rather than simply combined for the purpose of increasing available features.

Future directions may involve combined interventions and adaptive rehabilitation systems. Combining robot training or electrical stimulation with tDCS or rTMS may achieve a linkage from lower sensory-motor input to upper cortical regulation [71,100]. Any additional benefit needs to be compared with each component separately, as the improvements after the combined approach do not necessarily imply a synergistic effect. Closed-loop systems also face similar challenges. Currently, most fNIRS applications rely on offline classification or comparisons before and after intervention, whereas true closed-loop systems require the use of physiological information to adjust assistance, stimulation, feedback, or task difficulty during the same session [117]. EEG may support rapid detection of motor intention, while fNIRS may track slow changes in cortical recruitment or fatigue. Progress depends on reproducible individual-level measures, external validation, and evidence that signal-guided adaptation improves meaningful functional outcomes rather than classification performance alone.

6 Conclusions

This review has outlined an fNIRS-informed pathway for stroke rehabilitation. ROI and paradigm design determine what cortical activity is sampled; analytical methods convert hemodynamic signals into measures of regional recruitment, connectivity, and network organization; and intervention studies test whether these measures are sensitive to treatment-related change. fNIRS currently functions mainly as a quantitative cortical readout for assessment and treatment monitoring. Its use as a feedback variable for online parameter adjustment remains limited by biomarker reliability and hemodynamic delay. Multimodal recording and combined central-peripheral interventions may broaden physiological interpretation, but they do not by themselves establish closed-loop control. Progress toward fNIRS-driven rehabilitation therefore depends on validating markers that are reproducible at the individual level, sensitive to clinically meaningful change, and suitable for use within adaptive treatment protocols.

References

[1] Feigin V L , Brainin M , Norrving B . et al. World stroke organization: global stroke fact sheet 2025. International Journal of Stroke, 2025, 20(2): 132–144

[2] Maier M , Ballester B R , Verschure P F M J . Principles of neurorehabilitation after stroke based on motor learning and brain plasticity mechanisms. Frontiers in Systems Neuroscience, 2019, 13: 74

[3] Filler J , Georgakis M K , Dichgans M . Risk factors for cognitive impairment and dementia after stroke: a systematic review and meta-analysis. The Lancet Healthy Longevity, 2024, 5(1): e31–e44

[4] Hart R G , Catanese L , Perera K S . et al. Embolic stroke of undetermined source: a systematic review and clinical update. Stroke, 2017, 48(4): 867–872

[5] Kwakkel G , Stinear C , Essers B . et al. Motor rehabilitation after stroke: european stroke organisation (ESO) consensus-based definition and guiding framework. European Stroke Journal, 2023, 8(4): 880–894

[6] Thiebaut de Schotten M , Forkel S J . The emergent properties of the connected brain. Science, 2022, 378(6619): 505–510

[7] Wang J , Zhou J , Zhu J . et al. Brain remodeling in stroke patients: a comprehensive review of mechanistic and neuroimaging studies. Behavioural Brain Research, 2025, 486: 115548

[8] Crofts A , Kelly M E , Gibson C L . Imaging functional recovery following ischemic stroke: clinical and preclinical fMRI studies. Journal of Neuroimaging, 2020, 30(1): 5–14

[9] Yeung M K , Chu V W . Viewing neurovascular coupling through the lens of combined EEG-fNIRS: a systematic review of current methods. Psychophysiology, 2022, 59(6): e14054

[10] Huo C C , Xu G C , Xie H . et al. Functional near-infrared spectroscopy in non-invasive neuromodulation. Neural Regeneration Research, 2024, 19(7): 1517–1522

[11] Bertolucci F , Chisari C , Fregni F . The potential dual role of transcallosal inhibition in post-stroke motor recovery. Restorative Neurology and Neuroscience, 2018, 36(1): 83–97

[12] Chen Y F , Zhu G Y , Mao M C . et al. Study protocol of a randomized controlled trial for the synergizing effects of rTMS and Tui Na on upper limb motor function and cortical activity in ischemic stroke. Frontiers in Neurology, 2022, 13: 993227

[13] Liu J , Xia Y F , Zhang Q . et al. The correlation between center of pressure and cerebral cortex activation characteristics in patients with stroke: a functional near-infrared spectroscopy study. Brain Research, 2025, 1860: 149648

[14] Xu G C , Huo C C , Yin J H . et al. Test-retest reliability of fNIRS in resting-state cortical activity and brain network assessment in stroke patients. Biomedical Optics Express, 2023, 14(8): 4217–4236

[15] Peng Y , Zheng Y , Yuan Z W . et al. The characteristics of brain network in patient with post-stroke depression under cognitive task condition. Frontiers in Neuroscience, 2023, 17: 1242543

[16] Bonanno L , Cannuli A , Pignolo L . et al. Neural plasticity changes induced by motor robotic rehabilitation in stroke patients: the contribution of functional neuroimaging. Bioengineering, 2023, 10(8): 990

[17] Cui Y , Cong F , Zeng M . et al. Effects and mechanisms of synchronous virtual reality action observation and electrical stimulation on upper extremity motor function and activities of daily living in patients with stroke: a protocol for a randomized controlled trial. Frontiers in Neurology, 2025, 16: 1499178

[18] Chen S Y , Tsou M H , Chen K Y . et al. Impact of repetitive transcranial magnetic stimulation on cortical activity: a systematic review and meta-analysis utilizing functional near-infrared spectroscopy evaluation. Journal of NeuroEngineering and Rehabilitation, 2024, 21(1): 108

[19] Gan L , Huang L T , Zhang Y . et al. Effects of low-frequency rTMS combined with speech and language therapy on broca’s aphasia in subacute stroke patients. Frontiers in Neurology, 2024, 15: 1473254

[20] Strotzer M . One century of brain mapping using Brodmann areas. Clinical Neuroradiology, 2009, 19(3): 179–186

[21] Tang Y , Chen T S R , Xu Y . et al. Investigating the correlation between neural and muscular activities during bilateral arm training in stroke survivors: a cross-sectional study. Neural Regeneration Research, 2026, 21(7): 3026–3034

[22] Huang Y H , Chen W Y , Liu Y H . et al. Mild cognitive impairment estimation based on functional near-infrared spectroscopy. Journal of Biophotonics, 2024, 17(1): e202300251

[23] Li X H , Fang F , Li R H . et al. Functional brain controllability alterations in stroke. Frontiers in Bioengineering and Biotechnology, 2022, 10: 925970

[24] Ni J , Jiang W , Gong X Y . et al. Effect of rTMS intervention on upper limb motor function after stroke: a study based on fNIRS. Frontiers in Aging Neuroscience, 2023, 14: 1077218

[25] Kang J L , Lu J Y , Gu M B . et al. Cerebral hemodynamics and functional connectivity changes in stroke patients with dysphagia under acidic taste stimulation: a preliminary study. Frontiers in Neurology, 2025, 16: 1533099

[26] Premchand B , Zhang Z , Ang K K . et al. A personalized multimodal BCI-soft robotics system for rehabilitating upper limb function in chronic stroke patients. Biomimetics, 2025, 10(2): 94

[27] Xu S , Zhu S Z , Li M Y . et al. Altered cortical activation patterns in post-stroke patients during walking with two-channel functional electrical stimulation: a functional near-infrared spectroscopy observational study. Frontiers in Neurology, 2025, 15: 1449667

[28] Peng C , Wang Z Y . Diagnosis of motor function injury based on near-infrared spectroscopy brain imaging (fNIRS) technology. Preventive Medicine, 2023, 174: 107641

[29] Chang W K , Park J , Lee J Y . et al. Functional network changes after high-frequency rTMS over the most activated speech-related area combined with speech therapy in chronic stroke with non-fluent aphasia. Frontiers in Neurology, 2022, 13: 690048

[30] Li H Z , Liu J J , Tian S . et al. Language reorganization patterns in global aphasia—evidence from fNIRS. Frontiers in Neurology, 2023, 13: 1025384

[31] Liu P J , Han F L , Yang X Y . et al. Design and clinical observation of a multisensory stimulation soft hand rehabilitation system. IEEE/ASME Transactions on Mechatronics, 2025, 30(6): 7489–7499

[32] Formica C , De Salvo S , Muscarà N . et al. Applications of near infrared spectroscopy and mirror therapy for upper limb rehabilitation in post-stroke patients: a brain plasticity pilot study. Journal of Clinical Medicine, 2024, 13(21): 6612

[33] Ma Y Q , Xie D Y , Yu Y . et al. Differences in brain activation and connectivity during unaffected hand exercise in subacute and convalescent stroke patients. Neuroscience, 2025, 565: 10–18

[34] Kim H , Kim J , Lee G . et al. Task-related hemodynamic changes induced by high-definition transcranial direct current stimulation in chronic stroke patients: an uncontrolled pilot fNIRS study. Brain Sciences, 2022, 12(4): 453

[35] Chen S Y , Mao M C , Zhu G Y . et al. Cortical activity in patients with high-functioning ischemic stroke during the Purdue Pegboard test: insights into bimanual coordinated fine motor skills with functional near-infrared spectroscopy. Neural Regeneration Research, 2024, 19(5): 1098–1104

[36] Li H , Fu X F , Lu L J . et al. Upper limb intelligent feedback robot training significantly activates the cerebral cortex and promotes the functional connectivity of the cerebral cortex in patients with stroke: a functional near-infrared spectroscopy study. Frontiers in Neurology, 2023, 14: 1042254

[37] Huo C C , Xu G C , Sun A P . et al. Cortical response induced by task-oriented training of the upper limb in subacute stroke patients as assessed by functional near-infrared spectroscopy. Journal of Biophotonics, 2023, 16(3): e202200228

[38] Chu Q T , Guo X , Zhang T Y . et al. Stroke-related alterations in the brain's functional connectivity response associated with upper limb multi-joint linkage movement. Brain Sciences, 2023, 13(2): 338

[39] Zhang Y Z , Wang D , Wang D Y . et al. Motor network reorganization in stroke patients with dyskinesias during a shoulder-touching task: a fNIRS study. Journal of Innovative Optical Health Sciences, 2023, 16(6): 2340003

[40] Qu J , Bu L G , Chen Z X . et al. ArmVR: innovative design combining virtual reality technology and mechanical equipment in stroke rehabilitation therapy. IEEE Transactions on Visualization and Computer Graphics, 2025, 31(5): 2288–2298

[41] Zhang G H , Zhu G Y , Hu F . et al. Dynamic neural mechanisms underlying sustained motor training post-stroke: an fNIRS investigation employing time-varying ALFF and functional connectivity analysis. Journal of Biophotonics, 2025, 18(3): e202400491

[42] Xie Y J , Chen K X , Ma X D . et al. Cortical activity during balance and walking tasks in stroke patients: functional near-infrared spectroscopy neuroimaging research. Brain Research, 2025, 1864: 149808

[43] Clark D J , Rose D K , Butera K A . et al. Rehabilitation with accurate adaptability walking tasks or steady state walking: a randomized clinical trial in adults post-stroke. Clinical Rehabilitation, 2021, 35(8): 1196–1206

[44] Yang Z , Ye L , Yang L N . et al. Early screening of post-stroke fall risk: a simultaneous multimodal fNIRS-EMG study. CNS Neuroscience & Therapeutics, 2024, 30(9): e70041

[45] Ma X C , Peng Y , Zhong L D . et al. Hemodynamic signal changes during volitional swallowing in dysphagia patients with different unilateral hemispheric stroke and brainstem stroke: a near-infrared spectroscopy study. Brain Research Bulletin, 2024, 207: 110880

[46] Yang C H , Zhang T Y , Huang K Q . et al. Increased both cortical activation and functional connectivity after transcranial direct current stimulation in patients with post-stroke: a functional near-infrared spectroscopy study. Frontiers in Psychiatry, 2022, 13: 1046849

[47] Liu Y Y , Zhong Z Y , Chen J . et al. Brain activation patterns in patients with post-stroke cognitive impairment during working memory task: a functional near-infrared spectroscopy study. Frontiers in Neurology, 2024, 15: 1419128

[48] Zou J H , Yin Y Y , Lin Z F . et al. The analysis of brain functional connectivity of post-stroke cognitive impairment patients: an fNIRS study. Frontiers in Neuroscience, 2023, 17: 1168773

[49] Tetsuka M , Sakurada T , Matsumoto M . et al. Higher prefrontal activity based on short-term neurofeedback training can prevent working memory decline in acute stroke. Frontiers in Systems Neuroscience, 2023, 17: 1130272

[50] Hermand E , Compagnat M , Dupuy O . et al. Functional status is associated with prefrontal cortex activation in gait in subacute stroke patients: a functional near-infrared spectroscopy study. Frontiers in Neurology, 2020, 11: 559227

[51] He J , Gong C , Zhu X . et al. Cortical activation changes in supratentorial stroke patients during posture-cognition dual task. Frontiers in Neurology, 2025, 16: 1521687

[52] Gilmore N , Yücel M A , Li X G . et al. Investigating language and domain-general processing in neurotypicals and individuals with aphasia—a functional near-infrared spectroscopy pilot study. Frontiers in Human Neuroscience, 2021, 15: 728151

[53] Arun K M , Smitha K A , Sylaja P N . et al. Identifying resting-state functional connectivity changes in the motor cortex using fNIRS during recovery from stroke. Brain Topography, 2020, 33(6): 710–719

[54] Dai Y , Huang F , Zhu Y F . Clinical efficacy of motor imagery therapy based on fNIRS technology in rehabilitation of upper limb function after acute cerebral infarction. Pakistan Journal of Medical Sciences, 2022, 38(7): 1980–1985

[55] Huppert T J , Diamond S G , Franceschini M A . et al. HomER: a review of time-series analysis methods for near-infrared spectroscopy of the brain. Applied Optics, 2009, 48(10): D280–D298

[56] Luke R , Larson E E , Shader M J . et al. Analysis methods for measuring passive auditory fNIRS responses generated by a block-design paradigm. Neurophotonics, 2021, 8(2): 025008

[57] Kohli S , Fitzgibbon-Collins L K , Luan S Y . et al. Exploring the relationship between prefrontal cortex activation, standing balance, and fatigue in people post-stroke: a fNIRS study. NeuroRehabilitation, 2025, 57(3): 352–362

[58] Kim D H , Lee K D , Bulea T C . et al. Increasing motor cortex activation during grasping via novel robotic mirror hand therapy: a pilot fNIRS study. Journal of NeuroEngineering and Rehabilitation, 2022, 19(1): 8

[59] Xu G C , Chen T D , Yin J H . et al. Lateralization of cortical activity, networks, and hemodynamic lag after stroke: a resting-state fNIRS study. Journal of Biophotonics, 2024, 17(7): e202400012

[60] Kong Y , Peng W N , Li J . et al. Alteration in brain functional connectivity in patients with post-stroke cognitive impairment during memory task: a fNIRS study. Journal of Stroke and Cerebrovascular Diseases, 2023, 32(9): 107280

[61] Liu P J , Yang X Y , Han F L . et al. Brain activation pattern caused by soft rehabilitation glove and virtual reality scenes: a pilot fNIRS study. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2024, 32: 3848–3857

[62] Liu L Y , Jin M X , Zhang L G . et al. Brain-computer interface-robot training enhances upper extremity performance and changes the cortical activation in stroke patients: a functional near-infrared spectroscopy study. Frontiers in Neuroscience, 2022, 16: 809657

[63] Guo C , Sui Y X , Xu S . et al. Contralaterally controlled neuromuscular electrical stimulation-induced changes in functional connectivity in patients with stroke assessed using functional near-infrared spectroscopy. Frontiers in Neural Circuits, 2022, 16: 955728

[64] Xu S , Chen S F , Chen N L . et al. Immediate neuroplastic changes in the cortex after iTBS on the cerebellum of stroke patients: a preliminary fNIRS study. Neural Plasticity, 2025, 2025(1): 1362222

[65] Sui Y X , Kan C J , Zhu S Z . et al. Resting-state functional connectivity for determining outcomes in upper extremity function after stroke: a functional near-infrared spectroscopy study. Frontiers in Neurology, 2022, 13: 965856

[66] Sheng Y L , Han J . Biomechanical characteristics and neuromuscular action control mechanism of single-dual-task walking-conversion training in stroke patients. Journal of Back and Musculoskeletal Rehabilitation, 2025, 38(3): 576–592

[67] Xia Y F , Tang X W , Hu R P . et al. Cerebellum-cerebrum paired target magnetic stimulation on balance function and brain network of patients with stroke: a functional near-infrared spectroscopy pilot study. Frontiers in Neurology, 2022, 13: 1071328

[68] Chen L L , Meng F Y , Huo C C . et al. Effects of tactile feedback in post-stroke hand rehabilitation on functional connectivity and cortical activation: an fNIRS study. Biomedical Optics Express, 2025, 16(2): 643–656

[69] Huo C C , Xu G C , Xie H . et al. Effect of high-frequency rTMS combined with bilateral arm training on brain functional network in patients with chronic stroke: an fNIRS study. Brain Research, 2023, 1809: 148357

[70] Xie H , Li X , Xu G C . et al. Effects of transcranial magnetic stimulation on dynamic functional networks in stroke patients as assessed by functional near-infrared spectroscopy: a randomized controlled clinical trial. Cerebral Cortex, 2023, 33(24): 11668–11678

[71] Xie H , Li X , Huang W H . et al. Effects of robot-assisted task-oriented upper limb motor training on neuroplasticity in stroke patients with different degrees of motor dysfunction: a neuroimaging motor evaluation index. Frontiers in Neuroscience, 2022, 16: 957972

[72] Xie H , Jing J , Ma Y P . et al. Effects of simultaneous use of m-NMES and language training on brain functional connectivity in stroke patients with aphasia: a randomized controlled clinical trial. Frontiers in Aging Neuroscience, 2022, 14: 965486

[73] Zou T , Liu N , Wang W F . et al. Longitudinal assessment of the effects of passive training on stroke rehabilitation using fNIRS technology. International Journal of Human-Computer Studies, 2024, 183: 103202

[74] Lee G , Lee J , Kim J . et al. Whole brain hemodynamic response based on synchrony analysis of brain signals for effective application of HD-tDCS in stroke patients: an fNIRS study. Journal of Personalized Medicine, 2022, 12(3): 432

[75] Qu J , Du Y J , Jing J . et al. Short-term longitudinal study on brain network informatics of stroke patients under acupuncture and motor imagery intervention. IEEE Journal of Biomedical and Health Informatics, 2025, 29(5): 3356–3365

[76] Yuan Z W , Mao F , Liu Z M . et al. Characteristic of motor-related cortical network during stimulating at the affected or bilateral limbs of stroke patients by acupuncture. Brain Research Bulletin, 2025, 227: 111393

[77] Xu G C , Huo C C , Yin J H . et al. Effective brain network analysis in unilateral and bilateral upper limb exercise training in subjects with stroke. Medical Physics, 2022, 49(5): 3333–3346

[78] Li W H , Cao P , Wei R . et al. Effect of sequential repetitive transcranial magnetic stimulation with bilateral arm training on the brain effective connectivity in chronic stroke. Journal of Biophotonics, 2025, 18(5): e202400508

[79] Wang X H , Li W D , Song R . et al. Corticomuscular coupling alterations during elbow isometric contraction correlated with clinical scores: an fNIRS-sEMG study in stroke survivors. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2025, 33: 696–704

[80] Li R H , Rui G X , Zhao C L . et al. Functional network alterations in patients with amnestic mild cognitive impairment characterized using functional near-infrared spectroscopy. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2020, 28(1): 123–132

[81] Wang Y L , Zhao X , Zhou W . et al. Dynamic weighted “small-world” graphical network establishment for fNIRS time-varying brain function analysis. Biomedical Signal Processing and Control, 2021, 69: 102902

[82] Zhou L G , Xiao F , Shen N . et al. Protocol for simultaneous EEG and fNIRS measurement in characterizing brain state. Phenomics, 2025, 5(2): 219–237

[83] Sun X L , Dai C Q , Wu X B . et al. Current implications of EEG and fNIRS as functional neuroimaging techniques for motor recovery after stroke. Medical Review, 2024, 4(6): 492–509

[84] Iadecola C . The neurovascular unit coming of age: a journey through neurovascular coupling in health and disease. Neuron, 2017, 96(1): 17–42

[85] Lin J N , Jin S X , You Y G . et al. EEG-fNIRS multilayer brain network analysis revealed functional neural reorganization of rTMS with motor training in stroke. IEEE Transactions on Biomedical Engineering, 2026, 73(1): 269–280

[86] Li R H , Li S , Roh J . et al. Multimodal neuroimaging using concurrent EEG/fNIRS for poststroke recovery assessment: an exploratory study. Neurorehabilitation and Neural Repair, 2020, 34(12): 1099–1110

[87] Ye S J , Tao L , Gong S . et al. Upper limb motor assessment for stroke with force, muscle activation and interhemispheric balance indices based on sEMG and fNIRS. Frontiers in Neurology, 2024, 15: 1337230

[88] Sheng Y X , Wang J X , Tan G S . et al. Muscle synergy plasticity in motor function recovery after stroke. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2024, 32: 1657–1667

[89] Liu J B , Sheng Y X , Liu H H . Corticomuscular coherence and its applications: a review. Frontiers in Human Neuroscience, 2019, 13: 100

[90] Chang H , Xin R , Sheng Y X . et al. Post-stroke corticomuscular coupling assessment based on bilateral cerebral hemisphere difference. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2025, 33: 1817–1826

[91] Jian C Y , Liu H H , Deng L C . et al. Stroke-induced alteration in multi-layer information transmission of cortico-motor system during elbow isometric contraction modulated by myoelectric-controlled interfaces. Journal of Neural Engineering, 2021, 18(4): 0460e1

[92] Saita K, Morishita T, Arima H, et al. Qualitative and comparative cortical activity data analyses from a functional near-infrared spectroscopy experiment applying block design. Journal of Visualized Experiments, 2020(166): e61836

[93] Caliandro P , Molteni F , Simbolotti C . et al. Exoskeleton-assisted gait in chronic stroke: an EMG and functional near-infrared spectroscopy study of muscle activation patterns and prefrontal cortex activity. Clinical Neurophysiology, 2020, 131(8): 1775–1781

[94] Song K J , Chun M H , Lee J . et al. The Effect of robot-assisted gait training on cortical activation in stroke patients: a functional near-infrared spectroscopy study. NeuroRehabilitation, 2021, 49(1): 65–73

[95] Yoo M , Chun M H , Hong G R . et al. Effects of training with a powered exoskeleton on cortical activity modulation in hemiparetic chronic stroke patients: a randomized controlled pilot trial. Archives of Physical Medicine and Rehabilitation, 2023, 104(10): 1620–1629

[96] Huo C C , Shao G J , Chen T D . et al. Effectiveness of unilateral lower-limb exoskeleton robot on balance and gait recovery and neuroplasticity in patients with subacute stroke: a randomized controlled trial. Journal of NeuroEngineering and Rehabilitation, 2024, 21(1): 213

[97] Wei Y, Chen J L, Fang R, et al. Investigating the effect of different types of exercise on upper limb functional recovery in patients with right hemisphere damage based on fNIRS. Journal of Visualized Experiments, 2024(204): e65996

[98] Li X , Zhang H H , Zhang W Y . et al. Neural mechanisms underlying the improvement of gait disturbances in stroke patients through robot-assisted gait training based on QEEG and fNIRS: a randomized controlled study. Journal of NeuroEngineering and Rehabilitation, 2025, 22(1): 136

[99] Zheng J Y , He W Y , Ma Q Q . et al. Cortical activation in robot-assisted dynamic and static resistance training combining VR interaction: an fNIRS based pilot study. NeuroRehabilitation, 2023, 52(3): 413–423

[100] Yu Y , Huang W H , Tuerxun H . et al. Enhanced neuroplasticity and gait recovery in stroke patients: a comparative analysis of active and passive robotic training modes. BMC Neurology, 2025, 25(1): 239

[101] He X K , Lei L , Yu G . et al. Asymmetric cortical activation in healthy and hemiplegic individuals during walking: a functional near-infrared spectroscopy neuroimaging study. Frontiers in Neurology, 2023, 13: 1044982

[102] Chen J , Li H , Zeng C . et al. Evaluation of the recovery outcome of poststroke cognitive impairment after cluster needling of scalp acupuncture therapy based on functional near-infrared spectroscopy. Brain and Behavior, 2020, 10(8): e01731

[103] Lin B B , Ni J L , Xiong X . et al. Language function improvement and cortical activity alteration using scalp acupuncture coupled with speech-language training in post-stroke aphasia: a randomised controlled study. Complementary Therapies in Medicine, 2025, 89: 103137

[104] Zhang J , Huang W H , Chen Z H . et al. Effect of auricular acupuncture on neuroplasticity of stroke patients with motor dysfunction: a fNIRS study. Neuroscience Letters, 2023, 812: 137398

[105] Fu X F , Li H , Yang W . et al. Electroacupuncture at HT5 + GB20 promotes brain remodeling and significantly improves swallowing function in patients with stroke. Frontiers in Neuroscience, 2023, 17: 1274419

[106] Li C , Chen Y , Tu S T . et al. Dual-tDCS combined with sensorimotor training promotes upper limb function in subacute stroke patients: a randomized, double-blinded, sham-controlled study. CNS Neuroscience & Therapeutics, 2024, 30(4): e14530

[107] Chen N , Qiu X , Hua Y . et al. Effects of sequential inhibitory and facilitatory repetitive transcranial magnetic stimulation on neurological and functional recovery of a patient with chronic stroke: a case report and literature review. Frontiers in Neurology, 2023, 14: 1064718

[108] Chang P W , Lu C F , Chang S T . et al. Functional near-infrared spectroscopy as a target navigator for rTMS modulation in patients with hemiplegia: a randomized control study. Neurology and Therapy, 2022, 11(1): 103–121

[109] Liu Y W , Ai Y N , Cao J . et al. High-frequency rTMS broadly ameliorates working memory and cognitive symptoms in stroke patients: a randomized controlled trial. Neurorehabilitation and Neural Repair, 2024, 38(10): 729–741

[110] Cheng C F , Liu T T , Zhang B B . et al. Effects of robot-assisted hand function therapy on brain functional mechanisms: a synchronized study using fNIRS and sEMG. Frontiers in Medicine, 2024, 11: 1411616

[111] Ji X , Lu X , Xu Y . et al. Effects and neural mechanisms of a brain-computer interface-controlled soft robotic glove on upper limb function in patients with subacute stroke: a randomized controlled fNIRS study. Journal of NeuroEngineering and Rehabilitation, 2025, 22(1): 171

[112] Mihara M , Fujimoto H , Hattori N . et al. Effect of neurofeedback facilitation on poststroke gait and balance recovery: a randomized controlled trial. Neurology, 2021, 96(21): e2587–e2598

[113] Alnajjar F , Ozaki K I , Itkonen M . et al. Self-support biofeedback training for recovery from motor impairment after stroke. IEEE Access, 2020, 8: 72138–72157

[114] Jeong H , Song M S , Jang S H . et al. Investigating the cortical effect of false positive feedback on motor learning in motor imagery based rehabilitative BCI training. Journal of NeuroEngineering and Rehabilitation, 2025, 22(1): 61

[115] Burma J S , Johnson N E , Oni I K . et al. A multimodal neuroimaging study of cerebrovascular regulation: protocols and insights of combining electroencephalography, functional near-infrared spectroscopy, transcranial doppler ultrasound, and physiological parameters. Journal of Neural Engineering, 2025, 22(1): 016003

[116] Wang X H , Luo Z C , Zhang M X . et al. The interaction between changes of muscle activation and cortical network dynamics during isometric elbow contraction: a sEMG and fNIRS study. Frontiers in Bioengineering and Biotechnology, 2023, 11: 1176054

[117] Liu J B , Tan G S , Wang J X . et al. Closed-loop construction and analysis of cortico-muscular-cortical functional network after stroke. IEEE Transactions on Medical Imaging, 2022, 41(6): 1575–1586

Rights & permissions

The Author(s) 2026. This article is published by Higher Education Press.