Functional neurological disorder (FND), a condition in which individuals experience neurological symptoms (eg, weakness and tremor) without a structural brain lesion, is a complex disorder at the intersection of neurology and psychiatry, involving a dynamic interplay between brain-body processes and the mind [1]. A key explanation concerns a neurobiologically informed model of hierarchical Bayesian inference [2], which describes perception as a process of combining prior expectations and sensory input to minimize prediction errors [3]. FND symptoms arise when these prediction errors are not properly updated over time, leading to dysfunction in Bayesian perception and action [3]. Maladaptive, highly weighted prior expectations, influenced by bodily attention, are central to this process [2]. Empirical evidence supports this model, such as studies on altered predictive mechanisms in functional gait disorder [4]. In this context, functional movement disorders (FMDs), a subtype of FND [5], are atypical movement patterns (eg, tremor, dystonia, and weakness) not explained by organic disease, altered by distraction or nonphysiologic maneuvers and distinct from neurologic disease–related movement disorders. FMDs are common in neurology, accounting for 15% of new referrals, second only to headaches. FMDs have a poor prognosis, with 39% of patients remaining the same or worsening over time, often experiencing high physical disability and psychological issues [6]. These disorders present a clinical paradox: while the movements appear voluntary, patients experience them as involuntary and uncontrollable. The prevalence of FMDs among neurology patients in neurology clinics underscores the need for effective therapeutic interventions to address the significant disability associated with these disorders.

Recent research has highlighted the role of sensory attenuation and the sense of agency in the pathophysiology of FMDs [6]. Sensory attenuation refers to the brain’s ability to predict the sensory consequences of voluntary movement that suppress the actual sensory feedback, that is, reafference, a process that is crucial for distinguishing self-produced sensations from external stimuli [7]. In this predictive mechanism, a duplicate of the motor command, known as an efference copy, is postulated to be sent to a forward model that integrates the efference copy with an estimate of the body’s state to predict the sensory consequences of the movement. This predictive mechanism is postulated to be impaired in patients with FMD [8], reducing their ability to accurately predict the sensory consequences, thereby leading to the development of a reduced sense of agency over their self-generated movement [3]. This reduced sense of agency could explain why these patients perceive their movements as involuntary despite their voluntary appearance [9].

The sensory attenuation process is primarily mediated by forward models within the motor control system [7] that predict the sensory consequences of a planned action and modulate the reafference accordingly [10]. When an action is self-generated, the forward model predicts the resulting sensory input and attenuates the sensory experience to prevent self-caused sensations from interfering with the processing of external stimuli. Sensory attenuation can be measured using the force-matching task, where participants match a force applied to their finger either directly using their opposite hand (direct condition) or indirectly using a potentiometer and a torque motor (slider condition) [11]. Healthy individuals typically overestimate the force in the direct condition, producing greater matched force due to sensory attenuation. This phenomenon is reduced in individuals with FMD, potentially due to altered predictive mechanisms affecting their sensory attenuation process [8].

The force-matching task needs a well-designed haptic device, developed using robust methodological considerations [11], that can be combined with wearable brain imaging targeting key brain regions [3] for brain-behavior analysis. The primary motor cortex (M1) is involved in the execution of voluntary movements and plays a critical role in the sensory attenuation process [12]. Using the theta burst stimulation (TBS) approach, Voss et al [12] applied continuous TBS and intermittent TBS over the M1 to alter its excitability. The authors found that continuous TBS, which decreases cortical excitability, reduced sensory attenuation and improved performance in a force-matching task, suggesting a disruption in the predictive process. By contrast, intermittent TBS had no effect on sensory attenuation. The cerebellum also plays a crucial role in fine-tuning motor actions and predicting their sensory consequences [13]. The study by Blakemore et al [13] demonstrated that the cerebellum predicts the sensory outcomes of motor actions and signal discrepancies between expected and actual (afferent) sensory feedback. Positron emission tomography scans of participants performing a delayed tactile task showed increased cerebellar activity with longer delays, confirming its role in fine-tuning motor actions and motor learning by comparing intended and achieved movements. Within the cerebello-thalamo-cortical network, the thalamus acts as a subcortical relay station. Most afferents to the thalamus, whether ascending or cortical, are collateral branches of axons targeting lower motor centers, suggesting that thalamocortical pathways primarily monitor ongoing motor commands, with modulatory inputs adjusting information flow based on attentional demands [14]. The thalamus exhibits reduced activity during self-generated sensations, indicating its (relay) attenuation [15]. Hua et al [15] applied dynamic causal modeling analysis that indicated top-down modulation from the right inferior parietal lobe during sensory attenuation of an auditory stimulus. In this context, sensory attenuation is evident not only in somatosensory domains but also in auditory and visual modalities; for instance, self-generated sounds and visual stimuli are perceived as less intense than those generated externally, highlighting the broad applicability of sensory attenuation mechanisms across different sensory modalities. In FMDs, impaired sensory attenuation during movement may disrupt the cortico-cerebello-thalamo-cortical loop (a disruption referred to as dysrhythmia) [16], resulting in exaggerated or reduced movements, with both reflecting a reduced sense of agency over one’s own actions [17]. On the basis of a literature review of predictive coding and adaptive control in human-machine interfaces (HMIs; the systems or devices that enable interactions between users [patients] and robotic platforms or rehabilitation devices), Dutta [3] proposed the somato-cognitive action network (SCAN; a proposed brain network of intereffector regions within M1 that integrates motor planning with cognitive, physiological, and whole-body action control) as a key M1-based system linking effector and intereffector regions to support action intention, sensorimotor integration, and self-agency (or “awareness of being the agent of one’s own actions” [18]) by closing the cortico-cerebello-thalamo-cortical loop [16].

The 2 key theories of the sense of agency are the comparator model and the theory of apparent mental causation [19]. The right temporoparietal junction (rTPJ; a brain region implicated in the sense of agency and self-other distinction in movement control) is postulated to be a key node in the agency network for neuromodulation in FND [20]. However, the coordinates of the mean peak of activity in the temporoparietal junction (TPJ) have varied across studies [21], showing activation in the angular gyrus, middle temporal gyrus, parietal operculum, and supramarginal gyrus. This variation reflects the multidimensional processes at the TPJ contributing to the sense of agency, such as mismatch detection, action awareness, and sensory-motor conflicts. The angular gyrus is involved in intersensory mismatch detection, action awareness, and multisensory integration, while the supramarginal gyrus processes sensory-motor conflicts. Changes in perspective, mentalizing, and deception specifically activate the TPJ and posterior superior temporal sulcus, highlighting the role of perspective [22] in assigning motor actions to oneself or others. Overall, different neural mechanisms and specific brain areas within the TPJ contribute to the formation of agency when motor control is disrupted [3]. Functional neuroimaging has shown reduced activity and impaired connectivity of the rTPJ during functional tremor, indicating that altered rTPJ connectivity may play a contributing role. An impaired self-agency can be tested using intentional binding (motor intention is temporally matched with sensory feedback [23]) and sensory attenuation paradigms presented by Maurer et al [17]. The authors found decreased functional connectivity in patients with FMD between the rTPJ—crucial for self-agency—and bilateral sensorimotor regions, including the right sensorimotor cortex, bilateral cerebellum, bilateral supplementary motor area (SMA) [24], and right insula.

In contrast to sensory attenuation, sensory amplification occurs when the perceived intensity of a sensory stimulus is enhanced due to the influence of selective attention [25], for example, after error commission [26]. This effect is particularly noticeable when an individual directs their visual attention to a specific touch event, thereby increasing the perceived intensity of the touch. This phenomenon is related to the visual enhancement of the touch effect, where simply viewing (or even mentally imagining [27]) a body part can change tactile acuity for that part. Frontal and parietal cortices are involved in the modulation of sensory precision by attention that involves neural hierarchies [25]. Increased activity in the posterior parietal cortex is associated with enhanced sensory perception when visual attention is directed toward a stimulus, supporting the idea of optimizing sensory precision through attention. Empirical studies have shown that visual attention to a touch event, whether active or passive, amplifies the perceived intensity of the touch. This sensory amplification effect is significantly larger than the sensory attenuation effect, underscoring the powerful influence of attention on sensory perception. Overall, the interplay between sensory attenuation and amplification highlights the complex nature of sensory processing, where the brain dynamically adjusts the perceived intensity of stimuli based on predictive models and attentional focus [10,25]. Understanding these sensory mechanisms provides valuable insight into how sensory experiences are modulated in various contexts, including during the observation of others’ actions and in conditions involving altered sensory processing, such as FMD [8]. In understanding sensory mechanisms, it is important to distinguish between interoception, the sense of internal bodily states, and exteroception, which relates to the perception of the external environment through senses. Pain, for instance, can be interoceptive when it arises from within the body; and, in some cases, dysfunctional sensory amplification due to the overweighting of nociceptive input can lead to exaggerated prediction errors, reinforcing and intensifying the perception of pain [28], a common feature in FND-related pain symptoms and diagnoses [29].

We aim to leverage the interplay between sensory attenuation and amplification in exteroception, which is crucial for action in the external environment, by applying an operant conditioning–based implicit learning approach [30]. While this method can also be applied to interoception, this study focuses specifically on external action. This approach has shown therapeutic potential in addressing learned nonuse, a condition thought to arise from a maladapted forward model, in individuals with chronic stroke [31]. Our operant conditioning technique reinforces desired motor behaviors using visual feedback by dynamically adjusting gain and noise, with the goal of reweighting sensory attenuation and amplification [30]. This multilevel stimulation process adjusts prediction errors by aligning predicted and actual sensory input within the brain’s predictive coding framework [30]. Kumar et al [31] showed the feasibility of the virtual reality (VR)–based balance training system that used the operant conditioning approach proposed by Dutta et al [30], where higher weighting of the center of pressure from the paretic limb for visuomotor control implicitly encouraged patients with hemiplegic stroke to engage the affected limb during weight-shifting gaming tasks. Intermediate catch trials provided unbiased measures of standing balance capability. This implicit learning approach incorporated individualized weight distribution for VR tasks as well as subtle reinforcement (including suggestions) during VR gaming as an operant factor [32] that improved paretic limb behavior.

Figure 1 shows the neurocomputational framework of agency and motor control [3] illustrated for fundamentals of laparoscopic surgery (FLS) training with a camera-equipped FLS trainer box [33]. It depicts how voluntary movement and the sense of agency are generated through a brain-body loop involving a prospective sense of agency leading to intention, prediction, and feedback. The prefrontal cortex (PFC) sets goals based on a prospective sense of agency, which are evaluated by the basal ganglia for cost and reward before being translated into motor commands via inverse and forward models. The cerebellum predicts the sensory consequences of action, and delayed reafferent signals from the body (eg, touch and position) are compared against these predictions. Discrepancies (prediction errors) update the internal state of the parietal cortex and feed into the SCAN for retrospective agency judgment. We found that portable brain imaging can capture error-related brain-behavior relationships during complex motor tasks [33], as demonstrated in Figure 2, by revealing distinct microstate transitions between expert and novice groups within the Bayesian predictive coding framework. Together, the prospective and retrospective senses of agency support metacognition and are crucial for rehabilitation using extended reality (XR; an umbrella term encompassing VR, augmented reality, and mixed reality technologies), where reestablishing predictive control and agency is key, particularly in disorders such as FMD.

The central hypothesis is that a prospective sense of agency, which enables self-initiated movement intentions, can be generated in a top-down manner through “suggestion and imagery” within an XR environment. This process can be monitored using slow movement-related cortical potentials (MRCPs) [34], such as the early readiness potential originating from the SMA complex (Bereitschaftspotential I) to the SCAN regions of the M1 (Bereitschaftspotential II). At the intereffector SCAN regions, MRCP is postulated to begin with a slow negative shift around 1.5-2 seconds before movement onset, peaking negatively at movement onset, which coincides with a sharp increase in event-related desynchronization in the mu and beta bands. While event-related desynchronization is reflected in non-SCAN regions of M1, the intereffector SCAN regions are postulated to support body integrated action planning. Within the XR context, exafferent sensory input will be modulated following self-initiated motor execution to integrate with reafferent feedback through postulated Bayesian fusion [3], thereby influencing prediction error dynamics that arise from a disrupted forward model, as observed in FMD [35]. Following movement execution (postmovement positivity in MRCP), beta event-related synchronization signals the return to a baseline state. In this context, through operant conditioning embedded in XR-based gaming rewards, implicit learning is postulated to drive the recalibration of the forward model, progressively reducing reliance on exafferent input and supporting the restoration of normal motor function (Figure 3).

Voluntary movement is initiated by neural activity that first appears in the rostral cingulate zone within the dorsal anterior cingulate cortex, followed by activation in the pre-SMA and SMA regions and then in the M1 [36]. All these regions send projections to the spinal cord, with the M1 serving as the primary conduit for transmitting motor commands via the corticospinal tract. Concurrently, an efference copy of the motor commands is sent to the primary somatosensory cortex, cerebellum, and striatum, where it supports real-time correction, motor learning, and the inhibition of competing movements. We propose that during the preparatory stage of a voluntary motor task, efferent signaling from the SMA complex [24] sets the initial conditions by modulating the SCAN node within the M1 [3], thereby priming the non-SCAN effector regions for context-appropriate action. We suggest that this contextual modulation of the SCAN node by the SMA is impaired in FMD [24,37], which can lead to persistent fatigue [38]. Moreover, stress-related mechanisms in FMD may arise from overlapping adrenal connectivity and medial motor regions (the SMA and dorsal anterior cingulate cortex) disrupting the integration of motor intention and bodily arousal [36].

Building on the findings of Maurer et al [17] that link impaired self-agency to disrupted connectivity between the rTPJ and sensorimotor regions, we propose using simultaneous mobile brain and body imaging (brain-behavior joint monitoring [33,39]) to monitor intentional binding and efferent signaling. In this context, action-perception coupling monitoring [39] will leverage task-specific dynamic functional connectivity (dFC; a method for assessing time-varying brain network interactions during tasks) [40] between the rTPJ and sensorimotor cortex to assess abnormalities in sensory feedback processing in FMD, specifically, dFC between the rTPJ and the SMA complex [24], which contributes to efferent signaling and intentional binding [41-43] from the pretask to posttask phases, respectively, in the action-perception cycle.

In FMD, increased (localized to the left SMA [37]) and persistent beta synchronization [44] disrupts motor preparation, diverging from the typical role of brief beta bursts in facilitating cognitive control and goal-directed behavior [45], that is, “brain as controller.” We propose that within the sensorimotor cortex, mu rhythms support the brain’s role as an “observer” [22] engaged in action observation, salience detection [46], and contextual gain setting [47], while beta rhythms reflect the brain’s role as a “controller,” governing motor execution and precision [48]. In this context, it is further postulated that mu rhythms are linked to SCAN-like intereffector zones and the cingulo-opercular (salience) network [36], whereas beta rhythms are more specifically tied to non-SCAN effector-centric motor control in the M1. These sensorimotor rhythms alternate in bursts to modulate corticospinal excitability in healthy individuals, that is, motor-evoked potential amplitudes measuring corticospinal excitability peak during the beta rhythm’s peak-to-falling phase and the mu rhythm’s trough-to-rising phase in the sensorimotor cortex [49]. The beta rhythms extend to broader motor-related regions, including the basal ganglia, and play a role in motor control and movement stabilization crucial in “brain as controller.” The dynamic interplay of mu and beta event-related desynchronization and event-related synchronization, which is essential for normal motor function, is disrupted in FMD and may be restored through electroencephalography (EEG; a technique for recording electrical activity of the brain, used here for assessing cognitive and motor states) via biofeedback-driven training, which sequences the execution of an observation model through to the correction step [35,50] in XR-based motor tasks (Figure 1).

The amygdala has functional and anatomical connections to cingulo-opercular network nodes such as the anterior insula and anterior cingulate cortex [51]. FMD is postulated to involve maladaptive cognitive learning from aversive stimuli, leading to overactivity in the amygdala [52]. A heightened fronto-amygdala connectivity can disrupt the left PFC’s role in flexible, top-down cognitive control [53]. Therefore, our operant conditioning approach in XR begins with a relaxation and deep-breathing activity [35] combined with vagus nerve stimulation [54]. This is followed by a motor relearning task using a haptic device, rooted in the behaviorist framework formulated by Skinner [55] and formalized through reinforcement learning [30], designed to deliver implicit rewards and promote the adaptation of the internal forward model through gaming behavior [35]. This process relies heavily on the cerebellum for prediction error coding to facilitate effective motor adaptation to exafferent signaling for reinforcement learning [56]. In this context, portable neuroimaging also enables the investigation of individual activity in the brain regions such as the ventrolateral PFC, superior parietal lobule, supramarginal gyrus, angular gyrus, dorsolateral PFC, frontal eye field, and premotor cortex, and M1 [33,39]. Within the frontoparietal network, a distinction is postulated between attention driven by unexpected stimuli (bottom up) and attention that is controlled and goal directed (top down) [33]. The ventral frontoparietal network, including regions such as the temporoparietal cortex and inferior frontal cortex, is specialized for detecting behaviorally relevant stimuli, particularly when they are salient or unexpected (bottom up). This network acts as a “circuit breaker” for the dorsal system, directing attention to salient events. By contrast, the dorsal frontoparietal network, comprising areas such as the intraparietal sulcus and frontal eye fields, is primarily involved in goal-directed, top-down attentional control. This system is engaged during focused attention on specific targets or tasks. Understanding the interplay between these brain networks using switching state-space models for neuroimaging [57] is crucial during operant conditioning in FMD. In this context, the information flow model for developing the sense of agency [58] involves a leading network for mismatch detection, including the right supramarginal gyrus, left anterior inferior parietal lobule, anterior insula, and rTPJ. Mismatch information is relayed through intermediate areas before reaching the lagging network, which includes the bilateral PFC, cingulate cortex, and bilateral posterior inferior parietal lobule. The role of the PFC in the lagging network suggests its later involvement in retrospective sense of agency processing.

Expertise in the FLS “suturing and intracorporeal knot-tying” task was assessed through error-related brain states using portable, multimodal brain imaging [33]. The study by Walia et al [33] revealed significant differences in EEG microstate dynamics between experts and novices during complex surgical task training, aligning with the microstate taxonomy formulated by Bréchet et al [59]. Microstate 1 (comparable to microstate C [ventral visuospatial attention] in the taxonomy formulated by Bréchet et al [59]) and microstate 3 (comparable to microstate D [focus switching and reorientation]) were dominant across conditions. Experts lacked microstate 2 (comparable to microstate A [exploration processes]) during error epochs, indicating reduced exploratory motor responses and a more efficient exploitative strategy. Experts transitioned frequently to microstate 3 for error correction, while novices relied heavily on microstate 1 and transitioned back and forth between microstates 1 and 3, reflecting ongoing attempts at skill learning (Figure 2). In addition, experts exhibited prefrontal activity (microstates 5 and 6) preceding transitions to microstate 3, while novices showed left-right transitions (microstate 4) to microstate 1. Thus, microstate analysis highlighted that experts use refined attentional dynamics and reduced exploratory activity during errors, consistent with the neurophysiological underpinning described by Bréchet et al [59]. Recently, we used wearable functional near-infrared spectroscopy (fNIRS; a noninvasive brain imaging method using light to measure changes in cerebral blood oxygenation) [60] to assess brain activation during simulated operative dictation in an ecologized enactive framework. The operative dictation of a surgical task with “brain as controller” relies on first-person mental imagery—“brain as observer” [22]—to consciously recreate and manipulate past action experiences as mental representations without physical stimuli (refer to the “action simulation” loop in Figure 1). Simultaneous expert feedback and cues (exafferent external suggestions) can enhance enactment imagination, facilitating beneficial updates to the internal models.

The cerebellum plays a crucial role in reshaping the relationship between motor intention, action, and outcome through its connectivity with the medial PFC, as well as its interactions within the thalamocortical circuitry, including the cerebellum-thalamus-SMA pathway, for sensory gating mechanisms in the correction step (Figure 2) [61]. This dynamic interaction in the “action simulation” loop (Figure 1) likely provides a foundation for hypnosis to modulate the sense of agency [62]. Evidence suggests that the cerebellum contributes to the prediction and coordination of motor actions, integrating sensory feedback to refine the alignment between motor intentions and perceived outcomes [13]. Hypnosis, through its effects on attention and sensory modulation, may influence these cerebellar processes, which can be facilitated with noninvasive brain stimulation [63,64], enhancing the precision of prediction errors [56,65]. The study by Polito et al [66] highlights how hypnotic suggestions can disrupt habitual motor-sensory loops, potentially recalibrating the perception of control over actions. In addition, Blakemore et al [62] demonstrated that hypnosis can modify the temporal binding of intentional actions and their outcomes, further supporting its capacity to reshape the sense of agency through altered cerebellar and SMA connectivity. These insights suggest that hypnosis, by interacting with the neural substrates involved in sensory gating and motor planning, has the potential to refine the brain’s representation of agency, particularly in conditions where the sense of agency is disrupted.

A recent study by Bréchet et al [67] examined how intense meditation training modulates EEG microstates, revealing significant postmeditation topographical changes. Two novel microstate topographies emerged after meditation, explaining nearly 50% of the data variance, and were localized to brain areas involved in self-related multisensory experiences, including the insula, supramarginal gyrus, and superior frontal gyrus. Unlike prior studies that observed stability in microstate topographies [68], meditation induced notable deviations from the premeditation and placebo conditions, including the absence of microstate C (postulated in “observation model” in Figure 2) dominance after meditation. These findings align with meditation-induced changes in consciousness, such as enhanced self-observation activity, which may increase during meditation-primed self-hypnosis [22]. The study by Katayama et al [69] explored EEG microstates during different stages of hypnosis, ranging from resting, light hypnosis, deep hypnosis, and recovery, to understand changes in brain activity associated with altered states of consciousness. The “classic” 4 classes (A, B, C, and D) were identified, with microstates A and C increasing in duration, occurrence, and time coverage during deep hypnosis, while microstates B and D decreased. Resting and recovery conditions exhibited intermediate microstate characteristics between light and deep hypnosis, indicating distinct trajectories in the transition to hypnotic states. The findings suggest that deep hypnosis prioritizes specific cognitive processes related to the postulated “brain as observer,” enhancing insight [22] while reducing executive processes related to the postulated “brain as controller.” Comparisons with other states revealed similarities between deep hypnosis and schizophrenia (eg, reduced executive control) and between light hypnosis and meditation (eg, relaxation and heightened awareness). Overall, the study by Katayama et al [69] demonstrates that EEG microstates systematically evolve with hypnotic depth, reflecting shifts in the brain’s cognitive and functional dynamics. In this context, the postulated “brain as observer” approach [22] highlights the functional contributions of the anterior and posterior subsystems of the default mode network (DMN) and their connectivity during future-oriented thought; for example, combining insight meditation (with self-hypnotic induction) to support “brain as observer” and enactment imagination (with suggestions) to enhance “brain as controller” may help modulate FMD symptoms. In the related study by Xu et al [70], participants reported higher levels of vividness and self-projection when reflecting on their future self, highlighting the dominant cognitive function of future-self imagery. A functional dissociation was observed within the DMN: the anterior DMN was more active during present-self reflection, associated with self-referential processing (“brain as observer”), while the posterior DMN was preferentially activated during future-self reflection, linked to mental scene construction (“brain as controller”). These findings highlight the distinct roles of DMN subsystems in future-self oriented thinking to escape the present brain-body FND state, offering insight into the neural basis of the “brain as observer” versus the “brain as controller” functions [22].

“Suggestion” and “imagery” are key tools in psychotherapy. For operant conditioning in XR [35], mental imagery can help patients reimagine movement scenarios with positive outcomes. Suggestions (with or without hypnotic induction [71]) can be used to challenge distorted beliefs about limb weakness with evidence [72] and promote adaptive perspectives. Priming new perspectives [73] can be strengthened with positive sensory feedback using an adaptive XR-based HMI to simulate “normal” movement state transitions [74]. This study introduces an innovative HMI approach by combining top-down and bottom-up attention facilitation to develop a sense of agency in an engaging XR gaming environment; for example, participants may lack task-error perception in the lagging network [58], showing low medial frontal cortex activity. In this context, portable brain imaging of medial frontal cortex activity can guide participant-specific motor control interventions to improve action-perception coupling via intentional binding using a haptic robotic device (HRX-1; a modular haptic robotic device developed by HumanRobotiX Ltd, used in this study for force feedback and biofeedback-based motor training; Figure 3). Our approach also aims to uncover—using brain-behavior monitoring—how the sense of agency develops through action-effect contingencies using tunable haptic feedback and XR force-matching tasks [3,11] under the action-effect binding paradigm [75].

In this study, we examine perception-action binding in FMD within the predictive coding framework [3], focusing on dFC across key brain regions, including the sensorimotor cortex, the pre-SMA and SMA regions (SMA complex [24]), and the rTPJ, all of which are essential for motor preparation, control, and the sense of agency [43,76]. In stimulus-cued movement, postmovement beta synchronization both predicts behavioral stimulus–response binding and is significantly elevated in patients with FMD compared to healthy controls [37]. Midcentral beta rebound reflects the electrophysiological “resetting” of overlapping networks in the motor area and SMA [24], also linked to visual cued motor imagery [77]. Persistent beta synchronization during motor preparation, not just after movement, may reflect abnormal explicit control of movement in FMD [44]. This likely results from excessive attentional focus on the act of moving itself, with attention misdirected away from the movement’s goal and instead fixated on the mechanics of movement. This suggests active monitoring of the state [3] (Figure 1), which is likely driven by the perceived risk and uncertainty within the intermittent control framework [78].

In FMD, the normal intermittency of active feedback control characterized by on-off switching seems to be replaced with active monitoring of the state [3], potentially due to perceived risk and uncertainty reflected in persistent beta synchronization [78]. While directly correlating on-off switching with beta desynchronization-synchronization is complex in real-world settings, this study aims to explore this relationship in a controlled XR environment where stochastic movements may reveal meaningful patterns. In addition, perceived risk and uncertainty can be modulated in an XR environment while probing intermittent control in individuals with FMD versus healthy controls. In this context, we suggest that “action simulation” as active observer originating from SCAN nodes within the M1 (Figure 1) operates in parallel with the motor execution loop, which arises from non-SCAN effector-specific regions in the M1 responsible for motor commands (postmovement beta rebound is generated in the motor cortex [79]). During motor imagery, this execution pathway is inhibited, allowing “action simulation” to dominate. Predictions generated by the “action simulation” active observer are integrated with reafference from body sensors via Bayesian fusion that may not always be optimal [3]; for example, in individuals with FMD, impaired beta synchronization may reflect dysfunction in this fusion process, potentially resulting in the disproportionate weighting of internally generated active observer predictions from the action simulation loop over actual sensory evidence from body sensors. We propose to normalize this fusion process through the integration of top-down motor imagery (“action simulation”) and bottom-up sensory feedback from body sensors modulated with exafferent inputs in the XR gaming environment. Our co-designed XR-based HMI aims to offer an artificial intelligence (AI)—and machine learning (ML)—enabled personalized platform for delivering a game-based intervention for home rehabilitation [35].

Prior studies on intentional binding [76,80,81] suggest that repeated operant experience can strengthen associations, linking it to lower-level, implicit aspects of the sense of agency through automatic associative learning. Prior studies also highlight a distinction between implicit and explicit aspects of the sense of agency. In this context, the conscious expectancy in conscious, reflective judgment of the sense of agency can be developed by showing patients with functional motor symptoms their physical signs, such as the Hoover sign or tremor entrainment [82]. Automatic link formation with associative learning for a prereflective, automatic experience of the sense of agency can be developed via repeated operant experiences as shown in our learned nonuse cases [31]. While the implicit and explicit aspects of the sense of agency are separable, they interact through the same top-down and bottom-up mechanisms [81] that can be artificially modulated in the XR through exafference (Figure 3). Therefore, XR with exafference holds significant potential for advancing the understanding, diagnosis, and treatment of FMD. The recent review by Brouwer et al [74] highlighted how VR can be used to explore neurocognitive mechanisms, such as the sense of agency, attention, and suggestibility, and proposed VR-based interventions that could offer novel therapeutic avenues. The premise is that VR can manipulate and study predictive coding abnormalities to uncover the mechanisms of FMD where the key concepts include attention, the sense of agency, and suggestibility.

A compelling recent study [83] showed that multidisciplinary treatment comprising physiotherapy and psychotherapy can significantly improve the physical component of quality-of-life outcomes in patients with FMD. However, a lack of specialists means that access to these treatments is sparse. Even when available, the cost for patients to attend regular hospital sessions can be financially prohibitive. Lord Darzi’s 2024 report further highlighted the need to bring care closer to home by aligning financial incentives and prioritizing technology [84]. Our proposal is therefore ideally timed [85], aiming to bring cost-effective and accessible VR rehabilitation [74] into the homes of people with FMD. We aim to create a scalable, home-based rehabilitation solution that shifts care from clinic to home with a stepped care model [86,87]. Home care reduces the need for travel; promotes sustainable health care; and offers accessible, personalized rehabilitation. Our approach addresses the growing demand for remote health care and will provide equitable and cost-effective therapies. The proposal also includes a product delivery road map with technoeconomic analyses applied to the stepped care model. This stepped care model proposed for Neurorehabilitation Online [88] can lower health care costs, encourage adoption, and expand the solution’s use to other rehabilitation needs.

The predesign study, part of patient engagement and technology coproduction, was classified as service development and quality improvement, not requiring full National Health Service (NHS) ethics review. All participants gave written informed consent, and the study followed the Declaration of Helsinki principles. For this study, ethics approval will be obtained from the relevant institutional review boards and the health research ethics committee at Lancashire Teaching Hospitals NHS Foundation Trust before commencing each research stage. All participants will provide written informed consent before participation. They will receive accessible materials explaining the project’s goals, methods, and potential risks, emphasizing their right to withdraw at any time. All personal and biometric data will be anonymized and encrypted. Data management protocols adhering to ISO (International Organization for Standardization) 27001 standards will be implemented to ensure confidentiality and secure storage, in compliance with the General Data Protection Regulation. Participants involved in narrative interviews will review and approve video content intended for research dissemination. They retain the right to withdraw their consent for the use of their video at any stage.

This study will recruit adults aged ≥18 years with a confirmed diagnosis of functional limb weakness, a well-characterized subtype of FMD. Functional limb weakness is among the most common presentations of FMD and is particularly suited for XR-based motor retraining due to the clarity of its diagnostic criteria (eg, the Hoover sign [72]), the reproducibility of motor tasks in VR environments [35], and alignment with clear rehabilitative goals [89]. This focus aligns with emerging evidence supporting the feasibility, usability, and potential of hypnotherapy in this subgroup [90]. All participants will be assessed by a consultant neurologist using positive diagnostic signs consistent with DSM-5 (Diagnostic and Statistical Manual of Mental Disorders [Fifth Edition]) criteria and established clinical features (eg, the Hoover sign and variability with distraction). This study will emphasize safety, usability, and the personalization of XR-based interventions in the home setting.

The inclusion and exclusion criteria are presented in Textbox 1.

Participants will be recruited through neurology outpatient clinics and the specialized FND service affiliated with the Royal Preston Hospital in Preston, Lancashire, England. Treating clinicians will identify eligible individuals during routine care, and interested participants will be referred to the study team for consent and screening. Partnerships with FND charities and patient advocacy groups will support broader outreach and ensure inclusive recruitment.

The research domain criteria (RDoC) framework [91], developed by the National Institute of Mental Health, offers a dimensional approach to understanding shared pathways, which can guide the development of targeted neurotechnological innovations. Unlike traditional diagnostic methods that focus on symptom-based categories, the RDoC framework seeks to understand disorders by integrating multiple levels of information, including neuroscience and behavioral science. The adoption of a dimensional, neuroscience-based framework is proposed to facilitate better phenotyping of heterogeneity and the identification of biomarkers, potentially leading to more precise and effective rehabilitation technologies [92].

While the RDoC framework offers a comprehensive approach to explaining the diagnosis and developing more personalized rehabilitation technologies, Spagnolo et al [92] acknowledge practical challenges, particularly the complexity of integrating multiple levels of analysis into the clinical care pathways. To address this, we propose using experience-based co-design (EBCD; a participatory method involving patients, caregivers, and clinicians in designing health care solutions based on lived experience) [93] adapted for technology integration with the proposed optimum clinical pathway for adults with FND [94]. Rehabilitation technologies encompassing HMIs that access, monitor, investigate, assess, manipulate, and emulate neural systems present significant ethical concerns due to their profound impact on human identity, autonomy, privacy, and well-being. In this context, engaging patients and stakeholders is essential for conducting health care economic analyses and addressing ethical issues in implementing rehabilitation technologies for FND.

This study aims to coproduce an at-home rehabilitation solution using XR biofeedback training under operant conditioning for functional limb weakness, integrated into the FND stepped care framework [86] for multidisciplinary treatment and rehabilitation. The goal also ensures that at-home rehabilitation solutions are not only precise and effective but also sustainable, economically viable, and ethically sound. Our predesign study [35], funded by Engineering and Physical Sciences Research Council (EPSRC) Network+ funding, led to the development of this study, beginning with activity 1 (engagement with and preferences of people with lived experience and broader stakeholders). Activity 1 will focus on understanding and innovating at-home rehabilitation XR solutions for integration into the adult clinical pathway, starting with recruitment and consent once ethics approval is obtained. While invention refers to the act of creating something entirely original, innovation involves enhancing existing technologies in new ways to use them. Our innovation effort will ensure that the co-design is informed by people with lived experience and broader stakeholder engagement and preferences, leading to more effective and meaningful implementation. The EBCD process involves 3 phases (Figure 4): capture the experience, understand the experience, and improve the experience (with the final phase being addressed in activity 2). Activity 2 will translate at-home rehabilitation XR technology requirements from activity 1 into actionable items using the predesign framework [35], considering broader stakeholder input and health care economics. We anticipate diverse views on AI regulations and the cost-effectiveness of XR technologies in FND, with concerns about equity and economic viability, particularly in resource-limited settings [94]. Cost-effectiveness analyses (CEAs) often overlook ethical considerations, potentially perpetuating disparities, especially in conditions such as FND [95]. To address the predisposing, precipitating, and perpetuating factors in FND [96], we will integrate rehabilitation technologies into a disease severity–based stepped care framework for FND management [86]. This activity, focused on activity 2 (requirement translation of XR technologies in FMD), will deliver actionable items and prioritized steps to close the loop, incorporating equity weights for CEAs based on patient disease severity [97] within the stepped care framework [86].

Collaboration with FND charities (eg, FND North and FND Action) and patient advocates will support recruitment, ensuring diverse participation through purposive sampling. EBCD processes will be used to coproduce the XR platform with participants, aligning the development with stakeholder needs.

Initial studies will take place in controlled laboratory settings at the Lancashire Clinical Research Facility to prioritize safety. Studies involving patients with FMD will be conducted under clinical supervision.

Minimal physical risks are anticipated with XR use. Co-design efforts will address known discomforts such as eyestrain, and iterative feedback will ensure participant safety and well-being during XR tasks. Recognizing the vulnerability of patients with FMD, therapeutic environments within the XR platform will be carefully co-designed to prevent distress, guided by feedback from the EBCD process. Transparent consent processes, stringent deidentification protocols, and active stakeholder collaboration will mitigate ethical risks associated with the research.

This coproduction study is projected to be completed within 4 years, with results anticipated for publication in 2029.

The REACT2HOME proposal lays the groundwork for a personalized, home-based XR rehabilitation approach tailored to individuals with functional limb weakness [89]. This subtype of FMD has a clearly defined semiology and lends itself well to structured motor retraining in XR. Our approach leverages operant conditioning and dynamic feedback loops to recalibrate predictive motor models, targeting the neural circuits implicated in agency and sensory attenuation. While full-scale clinical efficacy remains to be demonstrated, early indicators from usability testing and prior XR and VR studies suggest that targeted digital biofeedback could serve as a promising adjunct to multidisciplinary FND treatment [35,74,104]. Given the variability in FMD presentation, we expect recovery trajectories to be most significant in patients with functional weakness who retain volitional motor capacity but show impairment in motor planning, agency, or learned nonuse patterns [90].

The predesign study demonstrated the potential of XR technology as a transformative tool for FND rehabilitation [35]. The perception-action cycle, which is central to human adaptation, is disrupted in FND due to maladaptive learning processes that diminish agency and disrupt motor control. By integrating biofeedback training with immersive XR environments, this approach addresses both motor and sensory dysfunctions, presenting a novel therapeutic avenue. The findings align with the results of our Delphi survey (a structured expert consensus method involving multiple rounds of questionnaires to refine opinions or priorities) on VR haptic biofeedback [105], which highlighted the relevance of XR technology in improving motor control for patients with FMD. Participants in both the predesign and Delphi studies emphasized the immersive and customizable nature of XR environments as a critical factor for engagement and recovery. However, barriers such as cost, accessibility, and motion sickness need to be addressed to enable widespread adoption, as reflected in both studies.

The usability testing revealed higher satisfaction with VR relaxation and XR position feedback tasks, aligning with the Delphi findings that underscored the importance of comfort and accessibility [35]. However, the variability in XR force feedback usability, particularly among participants with conditions such as functional dystonia, emphasizes the need for personalized approaches. These results support the Delphi survey recommendation for task-specific customizations and accessibility enhancements to ensure broader usability across diverse patient needs. The predesign study also underscored the importance of coproduction in designing rehabilitation technologies, as demonstrated by the mixed methods approach and Delphi framework. Engaging patients with FND and clinicians in the design process ensures that the technology addresses real-world challenges, such as motion sickness and sensory overload, highlighted in both the predesign and Delphi studies.

This study builds on prior predesign work in FND rehabilitation [35] and integrates key pathophysiological insights specific to functional limb weakness, namely deficits in sensory attenuation, motor intention encoding, and agency disruption, to refine the XR platform for this targeted clinical application. By aligning the XR interventions with these mechanistic targets, such as prediction error recalibration and intention-action coupling, the platform aims to deliver therapeutically relevant biofeedback experiences. Patient-specific adjustments, including calibration for individual motor impairments and optimization of sensory inputs, are incorporated to enhance comfort, usability, and therapeutic responsiveness. Using a structured EBCD process, our approach supports meaningful engagement with patients and caregivers, ensuring that the technology remains responsive to their evolving needs. The platform’s modular design, guided by insights from both iterative usability testing and Delphi surveys, positions it for broader integration into a stepped care FND model, facilitating sustained improvements in functional outcomes and patient experience from clinic to home.

While several VR interventions have been explored for FND [74], most existing tools are based on cognitive behavioral therapy principles, exposure-based paradigms, or distraction techniques aimed at symptom habituation or anxiety reduction; for example, VR protocols often focus on mirror visual feedback and exposure therapy to reduce symptom expression during tasks [104]. By contrast, our XR platform introduces a mechanistically grounded operant conditioning framework designed to recalibrate sensorimotor prediction errors and restore the sense of agency. This is achieved through the real-time, implicit reinforcement of volitional motor behaviors using personalized visual and haptic feedback. Rather than focusing solely on cognitive reframing, the platform targets disrupted forward models and sensory attenuation mechanisms, engaging dynamic brain-behavior loops through closed-loop feedback for implicit learning [3]. In addition, our use of wearable neuroimaging and dFC analysis to guide task adaptation and monitor agency-related brain networks (eg, SMA-rTPJ connectivity) represents a significant technical and conceptual advancement over prior VR applications in FND rehabilitation. This biologically informed, adaptive XR approach fills a critical gap between top-down cognitive interventions and bottom-up motor retraining strategies. However, we recognize that FMD encompasses a broad spectrum of presentations, including functional tremors and dystonia, which may respond differently to XR-based interventions. Therefore, while the current design is optimized for patients with functional limb weakness, future iterations of the platform could include additional modules tailored to other FMD subtypes; for example, tremor-specific VR scenarios could be introduced using wearable sensors and visual feedback techniques.

As AI- and ML-enabled XR technologies continue to evolve, their adaptability offers significant promise for addressing the heterogeneity of FMD, for example, by using physical AI to address real-time force feedback in functional dystonia [35]. Advances in wearable brain imaging, AI- and ML-enabled personalization, and modular haptic feedback design will facilitate expansion beyond single subtypes. By grounding these tools in robust clinical frameworks and coproduction with patients, AI- and ML-enabled XR platforms can support both therapeutic engagement and mechanistic understanding of symptom recovery in FND. In this context, our long-term vision is to incorporate this technology within a stepped care model [86], enabling equitable, scalable, and home-based rehabilitation for a broader population with FMD. Ongoing stakeholder engagement and iterative validation will be key to achieving this translational pathway. To ensure long-term sustainability beyond the 4-year study timeline, we are developing an implementation road map that aligns with the priorities of the NHS long-term plan for digital health and community-based care. Upon the successful demonstration of clinical efficacy and usability, we aim to transition the XR platform into validated digital therapeutics that can be adopted within existing care pathways for FND. We plan to partner with integrated care systems, NHS Trusts, and neurorehabilitation clinics to pilot deployment within stepped care service models. In parallel, we will engage early with commissioners, health economists, and digital health assessment frameworks (eg, the National Institute for Health and Care Excellence evidence standards framework for digital health technologies [106]) to ensure readiness for regulatory approval and health technology assessment. Reimbursement strategies will include positioning the platform under NHS tariff codes for remote rehabilitation and pursuing designation as an “innovation technology payment” candidate [107]. In addition, we will explore partnerships with insurers and digital health platforms to extend reach into community and private care settings. Open-source models for modular hardware components and software algorithms will also be evaluated to support broader dissemination and integration. These combined strategies will help embed this XR platform into routine FND care, making it a scalable and sustainable solution beyond the initial funding period.

The REACT2HOME study presents a novel, interdisciplinary approach to at-home neurorehabilitation for functional limb weakness. Grounded in predictive coding and agency-related neuroscience, the XR platform combines operant conditioning with adaptive feedback to restore volitional motor control and recalibrate disrupted sensorimotor processes. The methodology integrates EBCD, QFD, and Agile engineering sprints to ensure that the technology is user centered, technically robust, and clinically meaningful. Personalization is achieved through ML algorithms that dynamically adjust task difficulty and haptic feedback based on individual behavioral and neurophysiological responses. A rigorous risk management strategy aligned with ISO 14971 and ISO/IEC (International Electrotechnical Commission) 5927:2024 ensures safety and usability, while economic evaluation under the GRACE framework addresses long-term value, equity, and risk-adjusted cost-effectiveness. Importantly, the study establishes a road map for sustainable adoption through partnerships with health care providers, regulators, and payers, positioning XR-based rehabilitation as a scalable solution within stepped care models for FND. Through this protocol, we aim to transform how functional motor symptoms are managed, delivering evidence-based, accessible, and adaptive therapy directly into patients’ homes.