Redefining stroke rehabilitation: Mobilizing the embodied goal-oriented brain

Advancements in stroke rehabilitation remain limited and call for a reorientation. Based on recent results, this study proposes a network-centric perspective on stroke, positing that it not only causes localized deficits but also affects the brain ’ s intricate network of networks, transiting it into a pathological state. Translating these system-level insights into interventions requires brain theory, and the Distributed Adaptive Control (DAC) theory offers such a framework. When applied in the rehabilitation gaming system, these principles demonstrate superior results over conventional methods. This impact stems from activating extensive brain networks, particularly the executive control network, focused motor learning


Introduction
Stroke is the third leading cause of death and disability worldwide, accounting for 56% of the burden of neurological disease [1].Stroke results from neural tissue necrosis following oxygen loss after occlusion or hemorrhage of vessels supplying blood to the brain, potentially causing debilitating symptoms ranging from motor and affective deficits to impaired higher-order functions such as cognition, consciousness, and memory, combined with long-term side effects such as seizures, pain, depression, and fatigue (for a review, see [2]).Despite substantial progress in treatment during acute stages of stroke such as thrombolysis and thrombectomy, prevention, diagnostics, rehabilitation, and prognostics of stroke are lagging [3].Yet, with increased stroke survival rates, more patients face impairments and need long-term care and rehabilitation.The longterm outlook of these survivors is negative, with high mortality, low quality of life, and reduced instrumental activities of daily living (iADL).Stroke is a significant and growing threat to society and already overburdened health systems.
Several taxonomies have been proposed to create structure in the plethora of stroke symptoms, comorbidities, and their complex interplay.The WHO's International Classification of Functioning, Disability, and Health (ICF) considers the body's impairment, the person's activities, and participation in the social environment.In contrast, the Action Plan for Stroke in Europe 2018e2030 (APSE) considers three poststroke deficit categories: motor, cognitive, and communication.In addition, the etiology of symptoms follows a distinct chronicity, which is believed to be defined by specific mechanisms of plasticity, adaptation, and behavioral and functional change.Five poststroke phases are differentiated: hyperacute (first 24 hours), acute (up to 7 days), early subacute (up to 3 months), late subacute (4e6 months), and chronic (beyond six months).The behavioral change up to the subacute phase is believed to express the potential for "true recovery," that is the restoration of movement patterns similar to those displayed before a stroke.Later stages putatively show adaptation of an affected limb's use or compensation through using initially inactive body parts.The cognitive and affective analogs of these discrete stages are not defined.Underlying these behavioral changes are processes of spontaneous recovery, plasticity, restoration of excitatory-inhibitory balance, and neuronal remodeling akin to the mechanisms involved in development, learning, and memory [4].Although the precise relationship between functional recovery and the neuronal substrate remains elusive, great confidence is expressed in the belief that recovery from impairment is mainly complete at three months [3,5].Yet, this belief in an associated critical period ending at month six poststroke creates a treatment gap because the main impact of stroke on society occurs in the chronic phase when treatment is commonly not provided, also based on the belief of a critical period.Indeed, APSE emphasizes that by 2030, chronic patients must receive support in the athome phase, raising the crucial question of how this can be achieved.

Assessment and rehabilitation
There is a common understanding that deficits poststroke should be met with rehabilitation, and various treatment approaches have emerged without a clear consensus on their effectivity or mechanisms of action (for a historical review, see Ref. [5]).According to clinical needs, standard therapy comprises aspects of physical therapy (PT) to maintain the physical functionality and fitness of the body, occupational therapy (OT) to maximize independence in daily living by improving necessary skills, constraint-induced movement therapy (CIMT), which aims to restore function through forced use of a paretic limb, and cognitive and speech therapy (ST) with regional variability between these recommendations [3].The scientific and clinical basis of these interventions is also controversial on methodological grounds [6e8].Hence, rehabilitation is recommended, but it is unclear which interventions should be used.
Recovery rates of leg and arm function reported in the 1980s are about 50% in the acute phase [9], with a progressive and initial score-dependent asymptotic improvement that leveled off by about six months poststroke [10].This effect was later identified as proportional recovery, with 70% of patients regaining some functional use and about 30% recovering 70% of their former function [11].Subsequently, proportional recovery was declared a principle of recovery, that is patients are expected to recover to about 70% ( AE 15%) of their lost function within 3e6 months poststroke [12].A more recent analysis, however, has shown that the role of proportional recovery was too enthusiastically embraced and was an artifact of the scales and statistical methods [13].Despite the polemic that has ensued on the validity of the proportional recovery rule, spontaneous recovery is a relevant and only partially understood phenomenon that potentially confounds the interpretation of the impact of interventions in the acute and subacute phases.As a benchmark, interventions must show an effect beyond what can be expected from proportional recovery alone.In addition, recovery dynamics are highly variable [14], thus raising important questions on patient stratification and the effectiveness and rationale of the clinical scales deployed.
The core parameter describing proportional recovery, that is 70%, has remained constant over the last 40 years.This leads to the sobering conclusion that the impact of rehabilitation interventions has remained constant during this period.That improvement in rehabilitation approaches is urgently needed, as illustrated by the observation that there is no correlation between time spent in rehabilitation and activities of daily living [15].In addition, a longitudinal study of over 22,000 patients in Sweden reports that, five years after a stroke, 66% of ischemic and 75% of hemorrhagic stroke survivors are dead or dependent [16].Moreover, four years after hospital discharge, the average remaining functionality of poststroke patients is about 20% of their functionality at discharge, or "rehabilitation in vain" [17].However, with the growing understanding of spontaneous recovery and the limited time dedicated to rehabilitation, even these initial putative gains might have been independent of the treatment.
Deterioration in the chronic phase results from both primary and secondary causes.For instance, prolonged nonuse of the affected limb may cause the loss of function, initiating a vicious cycle in which nonuse (i.e., "learned nonuse") and poor performance reinforce each other, leading to a "use it or lose it" dynamic and the progressive degradation of function [18].This effect suggests that, for patients with deficits that respond to therapy, recovery is bistable [19]: they either enter a vicious cycle of decline because they do not use affected functions, or they build a virtuous cycle of recovery by incorporating these functions in daily activities, facilitating massed practice, driving use-dependent plastic changes where use and recovery reinforce each other.Rehabilitation must be focused on instilling the latter.Overall the evidence base behind the standard model of rehabilitation is weak, and breakthroughs are urgently needed in linking preclinical and clinical work in this area [20].In addition, these poor results reveal fundamental limitations in our current understanding of brain organization, motor learning, the mechanisms underlying recovery poststroke, and the principles on which neurorehabilitation should be based and are an invitation to look beyond the standard model.

From the modules to networks
It would appear intuitive to find an isomorphism between lesion location and functional deficits, as already demonstrated in the 19th century by, for instance, Paul Broca and the study of aphasia, or the relation between stimulation of the motor cortex and movement elucidated by Fritsch and Hitzig [21].The structurefunction isomorphy intuition goes back to at least Galen and has driven 19th-century phrenology and the attempts of 20th-century behaviorism to map laws of learning to their neuronal substrate.Early functional imaging experiments confirmed this notion of a local and hierarchical organization of motor control [22], further supported by the assumed critical role of the perilesional area, or penumbra, in recovery [23].This bias towards functional localization has also informed rehabilitation approaches emphasizing "true recovery" and modalityspecific restitution.Yet the actual poststroke structure-function mapping of the brain does not follow this traditional idea (see for a review [24]).For instance, most patients show multiple deficits (6%: 1e2 deficits; 31.1%:3e5; 50.6%: 6e10; 10.6: > ten at three months poststroke) [25] that tend to be highly correlated and low-dimensional [26].In addition, the symptoms of first-time stroke patients can be explained through 3 underlying dimensions supposedly reflecting shared neuronal operations: 1) language, memory, calculation, apraxia, and allocentric neglect; 2) left-side motor, visual, and left-egocentric neglect and overall performance deficits; and 3) right-side motor deficits.Despite their apparent heterogeneity, these dimensions map onto specific lesion clusters: left cortico-subcortical, right cortico-subcortical, and left subcortical regions, respectively.Hence, the functional modularity assumed in both traditional assessment and rehabilitation and the associated isomorphy between lesion location and symptoms appear at odds with the functional and structural organization of the brain.
Relating lesion location and volume to symptom etiology is complex due to structural and functional factors and, more specifically, the occurrence of diaschisis.Diaschisis is a phenomenon where local lesions induce changes in neural activity and connectivity beyond lesioned areas (see for a review [27]), or the brain's equivalent of action at a distance.First identified by von Monakow in 1914, diaschisis initially referred to acute excitability changes in regions distant from a lesion but now includes global connectivity changes like functional connectivity deficits and structure-function decorrelation.The most substantial disruption, correlating with deficits in higher-order cognitive function, can be seen as a loss of initial modularity.This metric quantifies the extent to which functional networks lose their intranode connectivity while increasing connectivity with initially unconnected nodes [28].Beyond connectivity, stroke lesions also significantly affect network dynamics by, for instance, inducing thalamocortical dysrhythmia [29] and impairment-related beta-band desynchronization [30].
Multiple experiments and clinical studies have shown that functional recovery depends on the reorganization of the cortico-spinal tract, brainstem pathways, and interhemispheric connections [31].For instance, recovered stroke patients show an increase in the fibers connecting the primary motor cortex to the red nucleus of the brainstem [32].Further, a detailed functional magnetic resonance imaging (fMRI) analysis of the recovery of upper limb functionality showed that a broad network of areas is relevant for functional recovery including the ipsilesional premotor cortex, contralesional supplementary motor area, insula, and cerebellum [33], with a pronounced relationship between the integrity of the frontoparietal network and motor recovery [34].This latter relationship is particularly intriguing because it directly links stroke recovery to the broader network underlying executive control [35].The comparator model of voluntary control provides a possible mechanistic understanding of these effects [36].Following this model, any action is defined by an intention or goal, mapped through an inverse model to a motor command.In contrast, a forward model predicts the sensory consequences of this command through its efference copy.A comparison between the predicted and actual outcomes of the action is subsequently used to adjust motor commands, bias attentional processing, and attribute agency [37].A mechanistic interpretation of this dual-process model proposes that the inferior parietal lobule (IPL) is the driver of activation in the frontal lobe (FL), which facilitates the supplementary motor area (SMA) to release the motor cortex (MC) from basal ganglia (BG)derived inhibition, thus initiating the motor command [38].FL subsequently returns an efferent copy to the parietal angular gyrus (AG), allowing for comparison by IPL, driving error-based correction, and defining the sense of ownership and agency.Indeed, stimulating the inferior parietal cortex (IPC) induces an urge to act [39], while the tempero-parietal junction (TPJ) increases its activity proportionally to the error between the predicted and actual effects of action [40].Further, lesions to the IPL lead to deficits of agency and ownership, where patients describe their actions as purposeless [41].
The above observations challenge the standard model with its modular view of motor control, providing an alternative and network-based perspective on the brain in health and disease.It suggests that stroke must be considered a network pathology affecting the brain at multiple spatiotemporal scales including the genomic, connectomic, vascular, neuronal, symptomatic, and environmental levels.The structural and functional impact of stroke then reflects a pathological state of a multi-scale network of networks (NoN) underlying the control architecture of voluntary goal-directed action rather than deficits to a localized functional module.This reorientation is immediately relevant to understanding the neuronal principles underlying motor control and the rehabilitation of poststroke deficits [24].The question is whether there is empirical support for this NoN perspective that places voluntary executive control at the heart of action orchestration and so rehabilitation post-stroke.

Stroke as a network pathology of brain architecture
In the 19th century, Hughlings Jackson advanced the notion that we must consider the brain a multi-layered control system.About 150 years later, we understand the various systems that form this architecture much better, from the spinal cord and brainstem motor nuclei to the cerebellum, basal ganglia, and neocortex serving decision-making, choice, commands, and execution of goal-oriented action [42].The effects of stroke on the NoN can be best understood in the context of this architecture.In this reorientation, the emphasis moves from the local and isolated effects of lesions and their peripheral manifestations to the perturbation of overall brain architecture they cause and the subsequent emergence of symptom clusters, comorbidities, and their interactions.To realize this mapping, a theoretical framework is needed to crystalize the core multi-scale organizational principles guiding precision digital brain health interventions and its specific applications in the real world.Such a candidate framework is provided by the Distributed Adaptive Control theory of mind and brain (DAC, Figure 1) [43].DAC is particularly relevant in this discussion because it articulates specific hypotheses on neural organization into one comprehensive architecture, which has given rise to advanced biomimetic robot control systems and a novel and clinically effective approach to poststroke rehabilitation.DAC aims to capture invariant principles of vertebrate brain organization [42].Returning to Claude Bernard and Ivan Pavlov, DAC assumes that the brain evolved to generate action, facing performance constraints such as efficiency, robustness, and flexibility [44].Given the potential conflicts in optimizing these objectives, brain architectures evolved to find solutions that are good enough rather than optimal or "satisficing" [45].DAC proposes that embodied action results from satisficing the core objective functions of why (motivation/drive/ goal), what (objects), where (space), when (timing of action), and who (other agents), or H5W [43].This posits that action unfolds within a specific ontological frame and raises the critical question of how the landscape of trade-offs is regulated or what the "constraints that deconstrain" are [46].Specifically, which priors or constraints must a nervous system incorporate to be maximally deconstrained in its task space?For instance, the dopamine (DA) system deconstrains the task space by supporting reward-driven policy learning and the adaptive regulation of action sequences [47].Yet, DA becomes a constraint when compromised in addiction or depleted, as in Parkinson's disease [48].DAC views the brain as a self-organizing system bootstrapped from relatively simple priors.The multiple control layers are continuously reconfigured dependent on task demands, satisficing dynamic constraint boundaries.DAC incorporates the comparator model of voluntary control by using the efference copy of the generated action to update the internal models of behavioral policies, biasing attention, and future action.The DAC provides a framework facilitating a focus on the network of networks embedded in brain architecture, the H5W ontological frame of action, core protocols, deconstraining constraints, and failure modes of this architecture to diagnose and repair the effects of stroke.
The earlier analysis including the presented empirical results, shows that the view of the standard model of single hidden factors causing modular pathologies, that is the notion of a "broken brain" or "broken movement," is empirically inadequate and insufficient as both a scientific model and clinical tool.Alternatively, stroke can be conceptualized as a pathological state of the NoN framed within the brain architecture underlying goaldirected action.This NoN is maintained in equilibrium against continuous internal and external perturbations.Pathology results when the multiscale NoN loses its resilience to perturbation and its dynamics become entrained by specific pathological states between and within its subnetworks and control layers.This multiscale system-oriented view gives an alternative perspective on the high variability and comorbidity observed in stroke and mental and brain disorders in general [49] and has given rise to the emerging notion of network medicine [50].I will use four examples to illustrate how empirical and theoretical systems neuroscience can be mobilized to address the practical challenges of stroke rehabilitation, focusing on goal-driven coordination of action, counterfactual error, homeostatic regulation of excitatory-inhibitory balance, and behavioral feedback.

Translating brain theory into interventions
The DAC ontological frame of action emphasizes the goals of embodied agents advancing cognitive control, task situatedness, multimodal stimulation, predictionbased learning, and interaction as core principles of motor learning.Yet, to translate these concepts into rehabilitation protocols, compatible technology must be found to control these factors relative to the user's actions and the required rehabilitation protocols.The most appropriate technology to satisfy these requirements is virtual reality (VR), and the rehabilitation gaming system (RGS) was created to map DAC towards poststroke motor, cognitive, and language rehabilitation, that is the three low-dimensional components of deficits poststroke [2].
Starting from the DAC principles, a broad number of protocols have been defined and clinically validated.One of these is spheroids which targets upper limb function and asks participants to use both hands to intercept virtual spheres under various task conditions, for example distractors, rule-based sorting, rewards, etc.
[2].An analysis of the performance of 251 stroke patients showed a significant and clinically relevant effect on both the Fugl-Meyer Assessment (FMA) and Chedoke Arm and Hand Activity Inventory (CAHAI) scores, irrespective of their chronicity [51].This result is also remarkable due to the low intensity of the intervention, lasting a total of 3 hours over three weeks with three weekly sessions of 20 minutes.A study with chronic stroke patients showed RGS-induced changes in movement kinematics consistent with motor learning [52].In contrast, an intervention that focused on repetitive movement using high-intensity, VR-delivered, and robot-supported arm training over a total of 30 hours with acute stroke patients, that is within the putative recovery window, showed no impact on FMA (2 daily sessions of 1 h each, five days a week over three weeks) [53].Surprisingly, this study also reported adverse side effects of the treatment, which is rare in the clinical use of VR.These results again illustrate that the standard model of stroke recovery, focusing on repetition and volume, has limitations that are overcome by a systemoriented perspective.

RGS drives the network of goal-oriented action
An fMRI analysis of healthy participants performing the RGS-spheroids protocol showed task segment-specific activation of brain networks involved in grasping, action observation, and motor imagery (Figure 2) [54].
The activated areas overlap with known resting-state brain networks, including the default mode network, the executive control network (ECN) or frontoparietal network, and activate the supplementary motor area, temporal and parahippocampal gyri, the inferior parietal lobule, and the inferior frontal gyrus related to the mirror neuron system.The overlap with the ECN occurs across all three stages of the protocol which is particularly relevant to the current discussion because it coincides with the ECN underlying volition and agency which is captured in the earlier mentioned comparator model of voluntary control and a component of the DAC theory.
The broad brain activation by goal-oriented and taskbased protocols starkly contrasts with the well-documented limited brain activation by repetitive movement alone, which is restricted to the sensorimotor resting state network [55].
A further guided transcranial magnetic stimulation (TMS) experiment showed that chronic stroke patients who used the spheroids protocol at home also displayed significant clinical improvements combined with enhanced activation of the distal thumb muscle (abductor pollicis brevis) and a marked shift of the centroid of MC activation that showed a significant correlation with clinical improvements [56].Together, these results indicate that protocols based on a systemlevel NoN perspective and focused on goal-oriented action drive global brain networks, emphasizing the ECN underlying voluntary action.The clinical outcomes confirm that this global activation creates conditions for motor learning and functional recovery including the reorganization of MC and the reinnervation of distal effector systems.This raises the question of what specific mechanisms of plasticity and learning underlie these observed functional changes.

The role of real and counterfactual error in goal-oriented learning
The constraints that deconstrain include control signals that serve motor learning and action orchestration and percolate globally through the control architecture.This notion has been elaborated in an alternative view on motor learning [57].Standard models of optimal motor control such as the classical Kawato Feedback Error Learning model, are based on the notion that predefined feedback control is progressively replaced by acquired feed-forward control [58].In contrast, motor control can be reconceptualized as coopting the predefined feedback control systems by adapting the error signals that drive them through a predicted or counterfactual error.This model is consistent with the modern view on motor detecting and signaling states from the physically instantiated self; and c) action (red), which establishes the interface between self and the world.The names of key neuronal structures are placed at locations in the architecture consistent with their role in generating goal-oriented action following [42].The arrows show the primary flow of information mapping exosensing and endosensing into action, defining a continuous loop of interaction with the world.Soma designates the body and its sensors, organs, and actuators.It defines the needs the organism must satisfy to survive, or the self-essential functions (SEF).The reactive layer (RL) comprises dedicated core behavior systems (CBS), each implementing predefined sensorimotor mappings serving the SEFs, defining homeostatic drive reduction processes controlling reflex action.To allow for priority selection, task switching, and conflict resolution, all CBSs are, in turn, regulated via an allostatic controller that sets their internal homeostatic dynamics relative to overall system demands and opportunities.The adaptive layer (AL) acquires and shapes the state space representing the world, agent, and action constrained by the value functions derived from the RL.Learning by the AL minimizes perceptual and behavioral prediction errors, building a model-free action generation system that integrates perceptual and behavioral learning.The acquired representational primitives are compressed into integrated episodic memories serving the contextual layer (CL).CL expands the time horizon in which the agent can operate, realizing model-based action policies through sequential short and long-term memory systems (STM and LTM, respectively).STM acquires conjunctive sensorimotor representations that the AL generates as the agent acts in the world.STM sequences are retained as goal-oriented models in LTM when a positive value is encountered, as defined by the RL and AL.The contribution of these stored LTM policies to goal-oriented decision-making, s, depends on five factors: perceptual evidence (p), memory chaining (m), the value and expected cost (v) of reaching a given goal state (g), and stored action policies (a).The virtualization layer builds abstract models of the agent and its task space, binding events into ontological frames organized around the dimensions: how (action), why (motivation), what (objects), where, when (sequential order, interval), and who.These virtualized event memory episodes serve internal parallel and subconscious simulation systems that underlie volition and agency.The compression of these parallel forward models into a single scene defies the content of consciousness, which serves norm extraction and meta-learning (see [43] for a detailed description).
control [42] and explains recent results such as the presence of cortical predictive error signals driving cerebellar learning [59].The notion of counterfactual error was incorporated in a variant of spheroids by artificially reducing visual feedback error derived from a physical reaching error, that is a manipulation for which VR is required.In a trial with chronic stroke patients, an acquired nonuse reaching bias was resolved, and bimanual symmetric action was restored in two sessions of 200 trials each, with only 100 trials being reinforced through intention-compatible reinforcement [60].Similar signatures of a clinically relevant virtuous loop of recovery through goal-oriented training have been reported in aphasia rehabilitation and multimodal feedback training of the upper extremities [2].The shaping of error signals, real or counterfactual, thus provides one relevant explicit channel to define rehabilitation and motor learning protocols.The theoretical DAC framework behind RGS also provides insight into the trade-off between vicious and virtuous recovery cycles through behavioral feedback [43].As an agent develops knowledge structures, behavioral policies, and habits, it also creates an effective environment that becomes an emergent constraint on behaviordprogressively biasing future action and learning that can become pathological, leading to a vicious cycle of decline or constructively shaped through targeted interventions, instilling a virtuous cycle of recovery.The Spheroids protocol capitalizes on these principles that were initially identified through brain theory.

Excitatory-inhibitory balance as a target for rehabilitation
The RGS interventions and those based on similar principles [2] show clinical impact beyond interventions based on the standard model.This raises the question of what principles of plasticity and neuronal reorganization could be at work.The phenomenon of diaschisis and its effect on the homeostatic regulation of excitatory-inhibitory (E-I) balance provides some initial insights into this additional constraint that deconstrains.A whole-brain model of the slow process of E-I balance regulation showed that after simulated focal gray-matter lesions, the response of the cortical graph is bimodal.On the one hand, local E-I homeostasis can drive the spontaneous and widespread restoration of disrupted network properties outside of the penumbra, such as modularity and structurefunction coupling.Conversely, it can also lead to maladaptive plasticity, contributing to the etiology of lateonset stroke symptoms [61].E-I balance regulation thus provides a mechanistic explanation for aspects of the nonspecific spontaneous recovery seen poststroke and functional reorganization also reflecting the bistability of recovery dynamics.Moreover, the global activation of the broad set of brain networks involved in voluntary motor control through goal-oriented protocols (Figure 2) creates conditions for E-I balance to be restored and novel pathways of motor function to be established, capitalizing on the mechanisms of learning and memory underlying goal-oriented behavior.In addition, this perspective on the local-global relationship between E-I balance and functional brain networks can be leveraged by direct modulation of the E-I balance of selected nodes of the cortical graph through noninvasive technologies such as transcranial electrical stimulation and functional ultrasound stimulation, providing additional promising strategies to improve functional recovery and consolidating its gains.

Discussion
Stroke is a significant and growing challenge for society.The dominant standard model of rehabilitation poststroke focuses on a modality-specific localized approach where modules are defined in peripheral and functional terms, assuming that function restoration is only feasible in a restricted critical period.Despite showing only a limited impact, the standard model has served as a rationale the support to stroke survivors.To rescue the traditional model, attention has shifted towards highintensity training instead of questioning its foundational dogmas.In addition, the reliance on Delphi analyses and roundtables to set rehabilitation standards suggests that stroke rehabilitation is prescientific, and alternative paradigms must be urgently explored.
Current evidence points to a link between the etiology of poststroke symptoms and disruptions to the functional and structural networks of the embodied and situated volitional brain.This network perspective is further supported by the fact that recovery is achieved through large-scale functional reorganization of these networks, with the frontoparietal executive control network playing a central role.In addition, poststroke lesions disrupt brain architecture globally through diaschisis and thalamocortical dysrhythmia.This suggests that progress in rehabilitation could be achieved by tackling stroke deficits through the lens of brain architecture and a network of networks (NoN) perspective.In this way, interventions would respond to the brain's integrated nature, placing motor recovery in the context of embodied voluntary goal-oriented action in ecologically valid environments.The complexity of the mind, brain, and behavior nexus requires that brain theory be mobilized to inform poststroke rehabilitation such as the Distributed Adaptive Control (DAC) theory of mind and brain.Indeed, protocols derived from the principles captured in DAC and delivered via the rehabilitation gaming system (RGS) have shown an impact in treating motor, language, cognitive, and affective disorders poststroke beyond those provided by the standard model [2].These effects are achieved by activating broad brain systems underlying voluntary action and agency, creating conditions for functional reorganization that are realized through a combination of explicit motor learning mechanisms (e.g.reinforcement and counterfactual error) and implicit self-organization (e.g.homeostatic E-I balance).The theoretical models underlying these principles provide inroads to the realization of patient-specific digital twins that can improve diagnostics and prognostics [61].These theoryinformed results offer new approaches toward neurorehabilitation and question several dogmas underlying the standard rehabilitationmodel, particularly the beliefs in distinct functional modules and a restricted recovery period.Future research should focus on the specific mechanisms of plasticity and learning underlying the system-level and functional changes observed in recovery and brain reorganization after stroke at multiple spatiotemporal scales, the trade-off between the vicious and virtuous recovery cycle, restorative network perturbations through neuromodulation techniques, and the combination of such non-specific methods with specific ones such as RGS.A core challenge for basic science and brain theory is to gain traction in the clinic, while clinical practice must be shaped to deliver rehabilitation protocols effectively and economically in the community-dwelling context, especially in the currently underserved chronic phase poststroke.Indeed, there is no scientific justification for maintaining a standard model of rehabilitation, which underdelivers in the acute phase, while effectively withholding treatment from chronic stroke survivors.Yet, despite the best aspirations of evidence-based medicine, the paradigm shift required to deliver appropriate support for stroke survivors appears to need more than empirical adequacy alone.Sufficient motivation for this crucial step should be found in the observation that placing the volition and agency of patients at the center of rehabilitation leads to better outcomes.

Declaration of competing interest
PFMJV is the founder and shareholder of Eodyne Systems S.L., which brings scientifically validated neurorehabilitation and education technologies to society.FPS is employed by Eodyne Systems SL.A breakthrough study reveals that deficits poststroke can be represented in a low-dimensional behavioral space that maps onto complex brain networks instead of localized lesions.An argument is made that this observation requires new methods for diagnostics derived from network neuroscience and challenges the standard view of neuropathology.29.van Wijngaarden JBG, Zucca R, Finnigan S, Verschure PFMJ: The impact of cortical lesions on Thalamo-Cortical network dynamics after acute ischaemic stroke: a combined experimental and theoretical study.e1011279.This breakthrough computational study shows that a biologically constrained whole-brain model of diaschisis and homeostatic excitatory-inhibitory regulation can replicate empirically valid benchmarks of stroke, symptom etiology, and recovery and that novel network-oriented metrics can effectively describe recovery dynamics.

Figure 2
Figure 2 and PHRASE (European Innovation Council, 101058240) and the euSNN project (Erasmusþ MSCA-ITN ETN H2020dID 860563) and ReHyb (Horizon2020 871767).The authors thank Dante Avino Martinez for his support in preparing the figures.ReferencesPapers of particular interest, published within the period of review, have been highlighted as:*of special interest * * of outstanding interest 1. Feigin VL, Brainin M, Norrving B, Martins S, Sacco RL, Hacke W, Fisher M, Pandian J, Lindsay P: World stroke organization (WSO): global stroke fact sheet 2022.Int J Stroke 2022, 17: , Ballester BR, Verschure PFMJ: Principles of neurorehabilitation after stroke based on motor learning and brain plasticity mechanisms.Front Syst Neurosci 2019, 13.This meta-analysis and review presents a comprehensive and unbiased summary of principles of motor learning derived from the neuroscientific literature and clinical studies that have shown effectiveness in rehabilitation post-stroke.3 * .Kwakkel G, Stinear C, Essers B, Munoz-Novoa M, Branscheidt M, Cabanas-Valdés R, Laki cevi c S, Lampropoulou S, Luft AR, Marque P, et al.: Motor rehabilitation after stroke: European Stroke Organisation consensus-based definition and guiding framework.Eur Stroke J; 2023.A relevant summary of treatment guidelines that is representative of the current state of the art and the methods pursued to establish standards for treatment.4 * .Joy MT, Carmichael ST: Encouraging an excitable brain state: mechanisms of brain repair in stroke.Nat Rev Neurosci 2020, 22:38-53.2020 22:1.A very helpful review that provides a solid overview of the multi-scale mechanisms underlying the adult plasticity that occurs after a stroke.The review of the literature leads the authors to advance a novel hypothesis on recovery mechanisms hinging on the re-allocation of neurons to recovering circuits.* .Páscoa dos Santos F, Verschure PFMJ: Excitatory-inhibitory homeostasis and diaschisis: tying the local and global scales in the post-stroke cortex.Front Syst Neurosci 2022, 15, 806544.This review provides an updated view of diaschisis, its genesis, and underlying mechanisms.It also links this phenomenon to recent advances in connectomics and whole-brain modeling to extract potential principles of recovery.The review shows that excitatory-inhibitory homeostasis can be relevant in explaining diaschisis and recovery.28 * * .Siegel JS, Shulman GL, Corbetta M: Mapping correlated neurological deficits after stroke to distributed brain networks.Brain Struct Funct 2022, 227:3173-3187.2022 227:9.