System and method for hierarchical cerebrovascular modal analysis and adaptive neurovascular modulation

A wearable system with integrated stimulation and sensing technologies provides real-time adaptive modulation of cerebrovascular reactivity and collateral perfusion, enhancing stroke recovery and long-term neurological outcomes.

WO2026239418A1PCT designated stage Publication Date: 2026-11-19DUTTA ANIRBAN
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/IN2026/050839
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-15
Filing Date
2026-05-15
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Existing systems fail to monitor or modify the dynamic haemodynamic responsiveness of leptomeningeal collateral pathways in real-time, lack individualized stimulation based on subject-specific cerebral haemodynamic feedback, and do not provide continuous monitoring and adaptive stimulation for collateral perfusion during acute and chronic phases of stroke recovery.

Method used

A wearable, closed-loop neurovascular modulation system integrating transcranial electrical stimulation, neuromuscular electrical stimulation, and multimodal sensing to assess and adaptively modulate cerebrovascular reactivity, collateral perfusion, and neurovascular coupling using hierarchical state-space modeling and real-time feedback.

Benefits of technology

Enables real-time monitoring and adaptive stimulation to enhance collateral perfusion, improve neurological recovery, and maintain cerebral haemodynamics, addressing individual variability and long-term cognitive impairment post-stroke.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000045_0000
    Figure 00000045_0000
  • Figure 00000046_0000
    Figure 00000046_0000
  • Figure 00000046_0001
    Figure 00000046_0001
Patent Text Reader

Abstract

A closed-loop neurovascular modulation system includes a cerebral sensing module, a peripheral or systemic sensing module, at least one non-invasive or minimally invasive stimulation or haemodynamic augmentation effector, and a controller configured to synchronize multimodal physiological signals, identify haemodynamic modes or neurovascular state variables, and output stimulation-control instructions. The controller may implement a two-level state-space model separating global macrovascular autoregulation from local microvascular neurovascular coupling through a cross-level coupling matrix that serves as a real-time biomarker of autoregulatory integrity. Dual-loop phase-coupled control coordinates central transcranial and peripheral neuromuscular stimulation. Applications include cerebrovascular monitoring, collateral-perfusion assessment, glymphatic-flow assessment or modulation, neurovascular rehabilitation support, and adaptive stimulation-control of cerebrovascular or systemic-to-cerebral haemodynamic coupling.
Need to check novelty before this filing date? Find Prior Art

Description

PT / 2026 / 12765SYSTEM AND METHOD FOR HIERARCHICAL CEREBROVASCULAR MODAL ANALYSIS AND ADAPTIVE NEUROVASCULAR MODULATIONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is related to U.S. provisional patent application serial No.63 / 556,268, titled “SYSTEMS AND METHODS FOR NON-INVASIVE BRAIN STIMULATION- INDUCED NEUROVASCULAR UNIT DYNAMICS AND PERIVASCULAR FLOW MODULATION”, fded on, 02 / 21 / 2024 which is herein incorporated by reference in its entirety.

[0002] This application is related to U.S. patent application serial No. 62 / 204,232, titled “SYSTEMS AND METHODS FOR DETERMINING NEUROVASCULAR REACTIVITY TO BRAIN STIMULATION”, filed on August 12, 2015, which is herein incorporated by reference in its entirety.INCORPORATION BY REFERENCE

[0003] All publications and patent applications mentioned in this specification are herein incorporated by reference in their entirety, as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference in its entirety.TECHNICAL FIELD

[0004] The present disclosure relates generally to non-invasive neuromodulation, neurovascular monitoring, cerebrovascular reactivity assessment, multimodal physiological sensing, and closed-loop control of physiological systems. More particularly, the disclosure relates to wearable, portable, bedside, or ambulatory systems and methods that combine non-invasive stimulation effectors, cerebral and systemic sensing, synchronized multimodal signal acquisition, haemodynamic mode identification, hierarchical state-space modelling, and adaptive control to assess or modulate cerebrovascular reactivity, neurovascular coupling, collateral perfusion, perivascular flow, systemic-to-cerebral haemodynamic coupling, and cerebral haemodynamics.PT / 2026 / 12765

[0005] In certain embodiments, the disclosure relates to matched-territory co-registration of transcranial Doppler ultrasound (TCD), multi-wavelength optical sensing comprising functional near-infrared spectroscopy (fNIRS), also referred to herein as near-infrared spectroscopy (NIRS) where haemodynamic optical measurements are acquired from cerebral or muscle tissue, and near-infrared transillumination or backscattering, speckle contrast optical spectroscopy (SCOS), electroencephalography (EEG), systemic physiological monitoring, and stimulation logs. The multi-wavelength optical sensing may estimate one or more of oxyhaemoglobin, deoxyhaemoglobin, total haemoglobin, cytochrome-c-oxidase-related optical changes, water-related optical changes, cerebrospinal-fluid-associated oscillations, or blood-flow-related optical signals. The synchronised data may be processed to distinguish global macrovascular dynamic cerebral autoregulation, local microvascular neurovascular coupling, cerebrovascular reactivity, and perivascular or paravascular flow dynamics, and to generate stimulation-control instructions for modifying one or more haemodynamic or neurovascular state variables.BACKGROUND

[0006] In acute ischaemic stroke, leptomeningeal collaterals (LMCs) — pre-existing arterial anastomoses between distal branches of the anterior, middle, and posterior cerebral arteries — represent a known compensatory perfusion pathway distal to a primary arterial occlusion. Under normal haemodynamic conditions, such collateral vessels carry minimal flow; however, when perfusion pressure falls, they recruit to redistribute blood from less-affected vascular territories toward the ischaemic zone. The presence, structural integrity, and resting calibre of LMCs are well-characterised in the neurological and stroke literature. What is not adequately addressed in the prior art is the capacity to monitor or modify the dynamic haemodynamic responsiveness of these pathways at the bedside in real time. There remains a need for systems configured to assess leptomeningeal collateral function using non-invasive physiological signals and to generate stimulation-control instructions based on that assessment.

[0007] Reperfusion therapies including intravenous thrombolysis and mechanical thrombectomy are established treatments for acute ischaemic stroke, and successful angiographic recanalization — restoration of blood flow past the occlusion site — is a recognised procedural endpoint. However, successful angiographic recanalization does notPT / 2026 / 12765consistently produce clinical improvement. Futile recanalization, defined herein as successful angiographic revascularisation without corresponding clinical improvement, may occur when ischaemic tissue has undergone irreversible injury during the occlusion period because collateral or microvascular perfusion was insufficient to sustain penumbral viability. Prior art does not disclose systems that monitor collateral haemodynamic state continuously during the pre-recanalization window and adapt stimulation based on the monitored state. There remains a need for systems configured to assess collateral adequacy using bedside physiological markers and to modify collateral perfusion using non-invasive or minimally invasive stimulation before, during, or after recanalization.

[0008] Robust collateral circulation is associated with delayed infarct growth and preservation of salvageable penumbral tissue, whereas poor collateral flow is associated with early core expansion, microvascular collapse, and reduced clinical benefit after recanalization. It is known that venous outflow resistance and intracranial pressure are determinants of effective cerebral perfusion pressure, particularly when dynamic cerebral autoregulation is impaired. It is further known that interventions that increase lower-extremity venous return — such as calf muscle pump activation, external pneumatic compression, or positional adjustment — can augment cardiac preload and thereby influence cerebral blood flow. However, prior art peripheral haemodynamic augmentation systems apply stimulation using fixed or pre-selected parameters and are not configured to adjust stimulation timing or amplitude based on concurrent real-time cerebral haemodynamic monitoring. There remains a need for systems that integrate peripheral venous-return augmentation with cerebral haemodynamic feedback to individualise stimulation delivery based on subject- specific cerebrovascular state.

[0009] Inter-individual variability in LMC network anatomy and haemodynamic responsiveness is well-established; robust collateral networks are associated with improved tissue rescue and reperfusion responsiveness, while poorly developed networks are associated with accelerated penumbral injury. Strategies targeting venous haemodynamics, such as lower-limb neuromuscular electrical stimulation or intermittent pneumatic compression, are known to modify systemic haemodynamic oscillations and autonomic state. However, no prior art system is configured to account for individual differences in collateral haemodynamic capacity when selecting stimulation parameters, nor to assess collateral responsiveness dynamically and adjust stimulation accordingly. There remains aPT / 2026 / 12765need for a stimulation control system that uses real-time haemodynamic signals to characterise subject-specific collateral state and to adapt peripheral and central stimulation based on that characterisation.

[0010] Post-stroke cognitive impairment and dementia represent a significant long-term consequence of acute ischaemic stroke, intracerebral haemorrhage, and subarachnoid haemorrhage. It is recognised that persistent reductions in cerebral blood flow, impaired neurovascular coupling, endothelial dysfunction, and insufficient collateral circulation may contribute to chronic cerebral hypoperfusion, neuronal dysfunction, white matter injury, and progressive cognitive decline following cerebrovascular injury. However, existing stroke therapy frameworks are directed primarily toward acute reperfusion and recanalization, and do not provide systems for sustained monitoring and modification of cerebral haemodynamic coupling, neurovascular unit function, or collateral reserve during the post-acute or chronic recovery phase. There remains a need for systems configured for longitudinal neurovascular monitoring and adaptive stimulation control to modify cerebrovascular reactivity, systemic-to-cerebral haemodynamic coupling, and neurovascular function following cerebrovascular injury.

[0011] Variability in collateral perfusion — including leptomeningeal, pial, and microvascular anastomotic pathways — is associated with differences in infarct progression, tissue viability, neurological recovery, and long-term functional outcome following ischaemic stroke. It is established that collateral recruitment can preserve perfusion to vulnerable neural tissue during the acute phase; however, the extent to which collateral pathways remain available or modifiable in the post-acute and chronic phases is not adequately characterised in the prior art. Furthermore, no prior art system provides a wearable, ambulatory platform that monitors collateral haemodynamic state in the chronic phase and generates stimulation-control instructions to modify collateral-dependent perfusion during neurovascular rehabilitation. There remains a need for wearable systems configured for repeated-session or continuous assessment of collateral haemodynamic state and for adaptive modification of cerebral blood flow, cerebrovascular conductance, and neurovascular coupling in subjects following stroke.

[0012] Peripheral haemodynamic interventions including neuromuscular electrical stimulation (NMES), intermittent pneumatic compression (IPC), and remote ischaemicPT / 2026 / 12765conditioning (RIC) are known to modify venous return, peripheral vascular resistance, autonomic state, endothelial signaling, and systemic haemodynamic oscillations. NMES applied to the calf or lower-limb musculature can activate the calf muscle pump and increase venous outflow, while IPC and RIC can generate cyclic peripheral haemodynamic perturbations and endothelial conditioning. However, conventional peripheral stimulation or compression systems are not configured to acquire muscle haemodynamics, cerebral haemodynamics, or systemic-to-cerebral haemodynamic coupling simultaneously with stimulation delivery, and therefore cannot individualize stimulation timing or dose based on subject- specific neurovascular state. There remains a need for peripheral haemodynamic augmentation systems that incorporate simultaneous cerebral and peripheral haemodynamic sensing and that adapt stimulation parameters based on real-time closed-loop feedback.

[0013] Peripheral haemodynamic stimulation is known to influence central and cerebral circulation through changes in venous return, cardiac preload, arterial pressure, autonomic regulation, respiratory state, and carbon dioxide. These systemic effects interact with dynamic cerebral autoregulation and collateral perfusion, particularly when autoregulatory capacity is impaired. However, prior art peripheral stimulation systems are not configured to model the interaction between peripheral haemodynamic effects and cerebral autoregulatory state, and do not incorporate a unified controller that accounts for this interaction when selecting stimulation parameters. There remains a need for a system that jointly models systemic haemodynamic driving forces and cerebral autoregulatory response, and that uses this joint model to coordinate peripheral and central stimulation toward defined cerebrovascular targets.

[0014] Non-invasive or minimally invasive approaches that have been investigated for supporting cerebral blood flow or collateral perfusion include positional adjustment, carbon dioxide modulation, sphenopalatine ganglion stimulation, transcranial electrical stimulation (TES), neuromuscular electrical stimulation, intermittent pneumatic compression, and remote ischaemic conditioning. These approaches act through different physiological mechanisms, including vascular tone modulation, autonomic reflexes, endothelial signalling, venous-return augmentation, and modulation of neurovascular-unit dynamics. However, each of these interventions is typically applied in isolation and with preselected or manually adjusted parameters, without a unified controller that jointlyPT / 2026 / 12765estimates systemic haemodynamic state, cerebral autoregulatory state, neurovascular coupling, and subject-specific stimulation response. There remains a need for a multieffector stimulation system in which a unified controller selects and coordinates central and peripheral stimulation parameters based on a continuously updated estimate of subjectspecific neurovascular state.

[0015] Computed tomography perfusion and magnetic resonance perfusion imaging are established techniques for identifying the ischaemic core, salvageable penumbra, collateral pathway, and perfusion-diffusion mismatch in acute ischaemic stroke. These imaging techniques provide anatomical and perfusion targets for treatment planning. However, perfusion imaging is not configured for continuous bedside monitoring, and prior art does not disclose systems that deliver adaptive stimulation based on real-time haemodynamic feedback. The prior art does not disclose non-invasive brain stimulation systems controlled using continuously acquired, synchronised multimodal haemodynamic feedback targeting pial collateral flow, penetrating arteriole dynamics, neurovascular coupling, perivascular flow, paravascular flow, glymphatic-flow-related dynamics, or systemic-to-cerebral haemodynamic coupling.. There remains a need for systems that bridge imaging -derived perfusion targets with continuous bedside haemodynamic monitoring and real-time adaptive stimulation control.

[0016] Cerebrovascular regulation is known to operate across at least two hierarchically coupled physiological levels. At the macrovascular level, dynamic cerebral autoregulation — quantified by transfer function analysis of blood pressure to cerebral blood flow velocity — maintains relatively stable cerebral blood flow across varying systemic pressures through vascular smooth muscle and myogenic mechanisms responsive to blood pressure, carbon dioxide, cardiac output, peripheral vascular resistance, and autonomic state. At the microvascular level, neurovascular coupling links local neural activity to local haemodynamic responses through neurons, astrocytes, endothelial cells, pericytes, arteriolar smooth muscle cells, and related neurovascular-unit components. These two regulatory levels overlap in frequency, phase, and spatial distribution of their haemodynamic signatures, making their causal separation difficult using single-modality recordings or conventional transfer function methods. There remains a need for a hierarchical state-space framework that explicitly models both regulatory levels, quantifies the coupling between them, and separates their respective contributions to observedPT / 2026 / 12765multimodal haemodynamic signals in real time.

[0017] Functional near-infrared spectroscopy (fNIRS) and near-infrared spectroscopy (NIRS), including continuous-wave, frequency-domain, time-domain, and multiwavelength optical sensing configurations, speckle contrast optical spectroscopy (SCOS), transcranial Doppler ultrasound (TCD), electroencephalography (EEG), arterial pressure monitoring, carbon dioxide monitoring, electrocardiography (ECG), pulse-wave analysis, electromyography (EMG), and muscle haemodynamic monitoring individually provide complementary information regarding cerebrovascular, peripheral, and systemic physiology. In multi-wavelength optical sensing systems, shorter wavelengths, including green wavelengths, may exhibit reduced tissue penetration depth and may preferentially detect superficial vascular pulsatility, including peripheral photoplethysmographic (PPG) signals, whereas longer near-infrared wavelengths may penetrate deeper biological tissue and provide information associated with haemodynamic variation in deeper tissue regions, including cerebral tissue regions. These modalities are well-characterised individually; however, the prior art does not disclose an integrated wearable or bedside platform configured to co-register these modalities to a common anatomically defined cerebral vascular territory, synchronise the corresponding data streams to a common time base with verified timing error, identify physiologically meaningful haemodynamic modes from the co-registered signals, distinguish brain-related modes from scalp or systemic components using feature-based classification, and adapt stimulation parameters based on subjectspecific modal or state-space estimates derived from the synchronised multimodal dataset. There remains a need for an integrated matched-territory sensing platform configured to co-register multimodal cerebrovascular signals to a common vascular territory, synchronise data streams with verified timing error, identify and classify haemodynamic modes, and adapt stimulation parameters based on subject-specific modal estimates, within a wearable, portable, bedside, or ambulatory device.

[0018] Systems capable of resolving depth-dependent cerebral haemodynamic contributions — distinguishing signals from pial vasculature, penetrating arterioles, capillary networks, perivascular spaces, paravascular spaces, and cerebrospinal -fluid-associated oscillations from superficial scalp, extracerebral, systemic, or artefactual components — are not disclosed in the prior art as integrated wearable platforms. Modal analysis, state-space modelling, graph-based spatial analysis, feature -based brain-scalp classification, structured perturbation, and multimodal sensing have individually beenPT / 2026 / 12765applied in research settings to characterize haemodynamic dynamics, but no prior art system integrates these methods within a unified framework configured to attribute observed haemodynamic modes to global dynamic cerebral autoregulation, local neurovascular coupling, pial collateral flow, perivascular or paravascular dynamics, systemic oscillation, or a mixed state, in real time and at the bedside. There remains a need for an integrated depth-sensitive modal analysis framework that performs this attribution and uses it to select stimulation targets.

[0019] The prior art described in the preceding paragraphs addresses individual aspects of cerebrovascular monitoring or stimulation: transcranial electrical stimulation (TES) systems are known; peripheral neuromuscular stimulation (NMES) systems are known; modal analysis is known; two-level state-space models of vascular regulation are known as theoretical constructs; and digital-twin haemodynamic models are known. However, no prior art system integrates central stimulation, peripheral haemodynamic augmentation, matched-territory multimodal sensing, hierarchical modal and state-space analysis, featurebased brain-scalp mode discrimination, Accepted Reproducible Mode validation with gatebased safety criteria, and adaptive stimulation-control generation within a single closed-loop platform. In particular, no prior art discloses the use of a cross-level coupling matrix as a real-time biomarker of autoregulatory integrity serving simultaneously as a control variable and as a stimulation target- selection criterion. The disclosed systems and methods address all of these deficiencies.

[0020] Transcranial electrical stimulation (TES), high-definition transcranial electrical stimulation (HD-TES), galvanic vestibular stimulation (GVS), temporal interference stimulation (TIS), transcranial magnetic stimulation (TMS), focused ultrasound stimulation (FUS), sphenopalatine ganglion (SPG) stimulation, superior cervical ganglion (SCG) stimulation, mastoid-carotid autonomic stimulation, neuromuscular electrical stimulation (NMES), intermittent pneumatic compression (IPC), and remote ischaemic conditioning (RIC) are each individually known in the prior art. These effectors act through distinct physiological pathways — cortical, autonomic, venous-return, endothelial, and perivascular — that interact with cerebrovascular regulation in complementary ways. However, no prior art system is configured to coordinate two or more such effectors under a single controller that uses synchronised cerebral, peripheral, and systemic signals to select stimulation timing, phase, amplitude, frequency, duty cycle, montage, and stimulationPT / 2026 / 12765sequence in real time based on identified haemodynamic modes. There remains a need for a unified multi-effector controller that coordinates central and peripheral stimulation based on a shared real-time neurovascular state estimate.

[0021] Machine-learned state- space models including neural ordinary differential equations, Koopman operator models, sparse nonlinear dynamics models, Gaussian-process state-space models, recurrent neural networks, and physics-informed neural networks are known computational techniques for modelling dynamical systems. Reinforcement learning and sparse-identification reinforcement-learning methods — combining sparse nonlinear dynamics identification with reinforcement learning to produce interpretable, sample-efficient control policies — are known in the control systems literature. Digital twin models of cardiovascular and cerebrovascular physiology are known as simulation tools. However, no prior art system is configured to apply these methods within a closed-loop wearable neurovascular stimulation platform subject to verified safety constraints, Accepted Reproducible Mode validation criteria, observability requirements for the two-level state-space model, and interpretability limits on the number of active policy terms required for execution on an embedded or wearable processor. There remains a need for an integrated machine-learned and interpretable stimulation-control architecture that satisfies these physiological, safety, and computational constraints within a wearable device.

[0022] Accordingly, there remains a need for such systems and methods configured to assess or modify cerebrovascular reactivity, collateral perfusion, neurovascular coupling, perivascular flow, paravascular flow, glymphatic-flow-related dynamics, and systemic-to-cerebral haemodynamic coupling.BRIEF DESCRIPTION OF THE DRAWINGS

[0023] FIG. 1 illustrates a schematic representation of a digital-twin human model of a wearable, closed-loop neurostimulation and neuroimaging system for modulating cerebrovascular reactivity and modifying collateral blood flow through leptomeningeal or pial vessels, in accordance with an embodiment of the present disclosure.

[0024] FIG. 2A illustrates placement of closed-loop sensor- stimulator units over cerebralPT / 2026 / 12765watershed territories. The units may be positioned over anterior cerebral artery (ACA)-middle cerebral artery (MCA) and MCA-posterior cerebral artery (PCA) border-zone regions identified from vascular territory mapping or perfusion imaging. Each unit includes stimulation electrodes and co-registered optical or electrophysiological sensing elements for acquiring local haemodynamic or neural signals from the same watershed territory. The arrangement enables modal-analysis-guided TES actuation to assess and modulate collateral perfusion in vulnerable watershed regions. Stimulation TES electrodes 60, 74, 76, and 78 are arranged around a central sensing region. Optical or electrophysiological sensing elements 70 (short distance LED / VCSEL), 71 (long distance LED / VCSEL), and 72 (4x1 ring EEG) may include light sources, photodetectors, or EEG electrodes positioned to acquire local haemodynamic or neural signals. A central element 68 (photodetector / CMOS) may represent sensor, reference electrode (4x1 Laplacian EEG, high density TES), or registration marker. The arrangement enables local sensing, modal analysis, and adaptive TES delivery within the same wearable unit, specifically optimized for resolving haemodynamic contributions from superficial leptomeningeal (pial) collaterals versus deeper parenchymal vascular structures. FIG. 2A also shows pial vessel connectivity highlighting the expected temporal coherence of blood flow in surface-level pial collateral vessels and FIG. 2B illustrates a Modal Analysis 6 workflow for classifying physiological haemodynamic modes. A stabilization diagram is used to identify stable modal components corresponding to cardiac oscillations 12, respiratory oscillations 13, and neurovascular oscillations 14, including cross-frequency coupling between oscillatory bands. The identified modes are subjected to multi-stage clustering using haemodynamic, spatial, spectral, and physiological features associated with vascular and neurovascular-unit dynamics. The resulting physiological clusters distinguish respiratory-related modes 13, arteriolar or myogenic low-frequency modes 15, and very-low-frequency microvascular or neurovascular modes 16. The classified modes may be used by the controller to select or adjust stimulation parameters for closed-loop neurovascular modulation, in accordance with an embodiment of the present disclosure.

[0025] FIG. 3 illustrates a depth-resolved multi-distance near-infrared spectroscopy system with progressively separated detector channels resolving superficial pial, mid-depth penetrating arteriole, and deep parenchymal microvascular signals. A light-emitting diode or VCSEL 501 directs near-infrared light into the scalp, and a linear photodiode array or CMOS detector receives optical signals at multiple source-detector separations. ThePT / 2026 / 12765separations sample different depths, including a pial artery region 502, a penetrating arteriole region 503, and a capillary or parenchymal microvascular region 504. The spectral plot and stabilization diagram show that the system can identify depth-dependent haemodynamic modes, enabling separation of pial, penetrating arteriole, and capillary dynamics for closed-loop neurovascular monitoring or stimulation control.

[0026] FIG. 4 illustrates patterned electrical stimulation for modulation of pial and penetrating arteriole dynamics. A transcutaneous or transcranial electrical stimulation input 2011 is applied to modulate perivascular autonomic pathways, including superior cervical ganglion- associated pathways, thereby inducing pial vessel oscillations 2012 at a vasomotor rhythm. The resulting oscillatory vascular response is propagated from pial arterial vasculature toward penetrating arterioles and downstream capillary beds. The lower panels illustrate pressure and velocity simulations showing stimulation-evoked changes in flow distribution, supporting closed-loop adjustment of stimulation parameters to modify perfusion through penetrating arterioles and high-resistance capillary regions.

[0027] FIG. 5 illustrates synchronization of peripheral neuromodulation with cardiac and vasomotor rhythms. The left panels show muscle NIRS and electromyography responses during stimulation, together with ECG timing 801, venous-return flow augmentation 802, NMES burst plus intermittent pneumatic compression timing, and vasomotor oscillation 803. The NMES, IPC, or remote ischaemic conditioning (RIC) bursts are timed relative to the cardiac cycle to augment venous return and systemic haemodynamic support. The right panels show stabilization diagrams indicating that 1 Hz NMES perturbs cardiac-related haemodynamic modes 804 in the fNIRS signal, enabling modal-analysis-guided adjustment of peripheral stimulation for closed-loop cerebrovascular modulation.

[0028] FIG. 6 illustrates the infrazygomatic crest electrode configuration for either transcutaneous (through the skin) or percutaneous (via a needle / electrode) sphenopalatine ganglion (SPG) stimulation via the pterygopalatine fossa to modulate autonomic nervous system activity and cerebral blood flow.

[0029] FIG. 7 illustrates a wearable neurovascular sensing and stimulation system (100) in two complementary views: a schematic anatomical view showing sensor and electrode placement relative to the superficial temporal artery, posterior auricular artery, and externalPT / 2026 / 12765carotid artery, and a wearable-device view showing a multi-band gel-cushion headset with symbol-coded sensing and stimulation elements. The system (100) includes a forehead band gel cushion (110) incorporating SCOS sensors, fNIRS optodes, EEG electrodes, and TES electrodes, including a temporal- artery SCOS sensor (111) positioned over or adjacent to the superficial temporal artery (120). A superior head band gel cushion may similarly incorporate SCOS sensors, fNIRS optodes, EEG electrodes, and TES electrodes. An around-the-ear gel cushion (130), positioned peri-auricularly adjacent to the posterior auricular artery (121), incorporates EEG electrodes, fNIRS optodes, and TES electrodes, and may optionally incorporate SCOS sensing elements. A mastoid-region gel cushion (140), incorporating a TES electrode and an SCOS sensor, is positioned over a mastoid or post- auricular region and, together with a carotid-region TES element associated with a carotid artery pathway (150), forms a mastoid-carotid TES stimulation pair. The carotid artery pathway (150) may correspond to one or more anatomical or functional pathways associated with the external carotid, common carotid, internal carotid, carotid sinus, carotid body, or cervical autonomic pathway. The symbol legend identifies SCOS source, SCOS detector, fNIRS source, fNIRS detector, EEG electrode, TES electrode, and gel-cushion element types.

[0030] FIG. 8 illustrates differential global-autonomic perturbation modalities integrated with a scalp-mounted SCOS-fNIRS-EEG sensing array 104. A galvanic vestibular stimulation (GVS) montage applies bilateral mastoid stimulation to activate vestibular afferents and vestibulo-autonomic pathways, thereby modulating sympathetic efferents and cerebral vasoconstriction or vasodilation. A temporal interference stimulation montage applies paired high-frequency currents to generate an interference envelope at a deep autonomic target, such as the fastigial nucleus, to produce a global brainstem or pressor autonomic perturbation. The scalp-mounted sensing array records haemodynamic and electrophysiological responses, enabling classification of global dynamic -autoregulation modes and their separation from local neurovascular-coupling responses.

[0031] FIG. 9 illustrates an image-guided closed-loop neurostimulation and sensing system for targeting ischaemic brain tissue. A plurality of fNIRS-SCOS sensors 214, 215, 216, 218 are positioned over a cortical region corresponding to an ischaemic core, vascular blockage, and alternative collateral-flow supply route. A transcranial electrical stimulation actuator system 50 communicates with the sensor array to deliver adaptive stimulation based onPT / 2026 / 12765measured haemodynamic responses. The inset shows an exemplary fNIRS -SCOS sensorstimulator unit comprising sensing module 80, modal analysis module 52, TES actuation module 54, TES electrodes 60, 74, 76, 78, and LED / VCSEL / EEG sensing elements 70, 71, 72 arranged around a central CMOS detector 68. The arrangement enables co-registered sensing, modal analysis, and TES actuation for closed-loop modulation of collateral perfusion.

[0032] FIG. 10 illustrates a wearable multi-level neurovascular sensing and stimulation system. A head-mounted fNIRS imaging assembly 17 acquires cerebral haemodynamic signals and communicates with a wearable TES controller 18. A calf-mounted NMES sleeve or sock 19 includes NMES electrodes configured to activate the calf muscle pump and a muscle NIRS imaging portion 20 configured to measure local muscle haemodynamics. The calf device may further operate with intermittent pneumatic compression to augment venous return. The combined head and calf sensing- stimulation arrangement enables coordinated monitoring of cerebral perfusion and peripheral musclepump activation for closed-loop modulation of systemic-to-cerebral haemodynamics.

[0033] FIG. 11 illustrates a human-in-the-loop model predictive control framework for TES -induced cerebrovascular and perivascular modulation. Panel (A) the distribution of the electric field on the surface of the brain resulting from 4x1 high-definition transcranial electrical stimulation. Panel (B) the structure of the pial blood vessels, which includes the adventitia layer containing autonomic nerves, the intima layer lined by endothelial cells, and the media layer consisting of smooth muscle cells. These layers are influenced by the electric field distribution; and Panel (C) a reduced pial-to-penetrating artery model coupled to pulmonary and systemic circulation with near-infrared transillumination of subarachnoid-space oscillations and haemoglobin fNIRS feedback provided to the controller. The controller uses model prediction, optimisation, objectives, and constraints to generate TES control actions and regulate blood-flow velocity in a penetrating artery, in accordance with an embodiment of the present disclosure.

[0034] FIG. 12 illustrates a human-in-the-loop model predictive control framework for coordinated peripheral and cerebral stimulation with haemodynamic oscillatory modes. The figure shows NMES, IPC, TES, TMS, or perivascular nerve stimulation applied to modulate venous return, pial and penetrating artery dynamics, with near-infraredPT / 2026 / 12765transillumination of subarachnoid- space oscillations and NIRS / SCOS haemoglobin feedback provided to the controller. The controller uses model prediction, optimisation, and safety constraints to adjust stimulation parameters. The oxyhaemoglobin, deoxyhaemoglobin, and total haemoglobin power spectra illustrate cardiac, respiratory, low-frequency, and very-low-frequency oscillatory components used to guide stimulation for neurovascular, perivascular, and metabolic regulation. In certain embodiments, very-low-frequency oscillations and TES-induced pressure pulses in pial vessels are used to modulate a pre-capillary sphincter or resistance bottleneck at the penetrating arteriolecapillary interface. In autoregulatory dysfunction or no-flow conditions, NMES or IPC may generate systemic venous-return, pressure, or vasomotor oscillations, while TES, TMS, or perivascular nerve stimulation modulates pial and penetrating artery dynamics to modulate a pre-capillary sphincter or resistance bottleneck, thereby modifying downstream capillary perfusion or local flow resistance.

[0035] FIG. 13 illustrates perfusion-imaging-guided montage design for targeting salvageable penumbra. MR perfusion, Perfusion-Weighted Imaging (PWI), Cerebral Blood Volume (CBV), Diffusion- Weighted Imaging (DWI), or PWI-DWI mismatch imaging identifies an ischaemic core, penumbra, and benign oligemia. The imaging-derived target map is provided to a human-in-the-loop model predictive controller, which applies objectives and constraints to design an fNIRS -SCOS -EEG-TES montage over the target penumbra. The montage positions optical, electrophysiological, and stimulation elements to monitor haemodynamic responses and deliver TES configured to support collateral recruitment from an arterial source toward the penumbra.

[0036] FIG. 14 illustrates a two-level state-space model architecture with cross-level coupling matrix L(t). Systemic inputs, including arterial blood pressure, carbon dioxide (CO2), cardiac output, and peripheral vascular resistance, drive Level 1 global macrovascular autoregulation. The cross-level coupling matrix L(t) quantifies propagation of Level 1 global forcing into Level 2 local microvascular neurovascular-coupling dynamics. Level 2 incorporates local inputs, including EEG-derived neural activity and TES stimulation waveforms, and outputs fNIRS oxyhaemoglobin (HbO), deoxyhaemoglobin (HbR), and total haemoglobin (HbT), and SCOS blood-flow signals. A joint observation equation combines TCD, fNIRS, and SCOS measurements, with SCOS blood-flow index (BFi) improving observability and reducing uncertainty in estimating thePT / 2026 / 12765coupling matrix L(t).

[0037] FIG. 15 illustrates feature-based characterisation and classification of haemodynamic modes. For each identified mode, the controller computes global correlation metrics, HbT global correlation, correlation-based signal improvement (CBSI) metrics, graph-Laplacian spatial smoothness, and band-specific physiological priors. These features are combined into a composite brain-scalp score to classify each mode as brainlike, scalp / systemic-like, or mixed / penetrating arteriole-like. The plot shows how the classification margin affects the fraction of modes selected for analysis, enabling threshold selection that balances confident brain-mode identification against exclusion of uncertain modes.

[0038] FIG. 16 illustrates a matched-territory multimodal sensing platform with Lab Streaming Layer (LSL) synchronization and Extensible Data Format (XDF) file recording. TCD, fNTRS, SCOS, EEG, blood pressure, CO2, ECG, and stimulation logs are coregistered to a common vascular territory, such as an MCA territory, while a reference territory may be used for comparison. The platform synchronizes all device streams through Lab Streaming Layer outlets, LabRecorder, Extensible Data Format (XDF) storage, and hardware trigger verification. The synchronized multimodal data enable territory- specific estimation of macrovascular flow, microvascular haemodynamics, neural activity, systemic inputs, and stimulation timing for two-level cerebrovascular model identification and closed-loop control.

[0039] FIG. 17 illustrates an exemplary structured perturbation matrix for identification and separation of a two-level cerebrovascular model comprising global autoregulatory Level 1 modes and local neurovascular-coupling Level 2 modes. Global perturbations may include spontaneous blood-pressure fluctuations, paced CO2 modulation, galvanic vestibular stimulation, intermittent pneumatic compression, squat-stand manoeuvres, and fastigial-nucleus temporal-interference stimulation, while local perturbations may include finger tapping, photic stimulation, and high-definition TES configured to evoke territoryspecific neurovascular responses. Multimodal observables including TCD cerebral bloodflow velocity, blood pressure, end-tidal CO2, fNIRS oxyhaemoglobin, SCOS blood-flow index, and EEG coherence may be used to classify identified modes as autoregulatory -dominant, neurovascular-coupling-dominant, CO2-reactive, or mixed.PT / 2026 / 12765

[0040] FIG. 18 illustrates an Accepted Reproducible Mode identification and gate -based validation framework. Synchronized matched-territory data are first acquired from TCD, fNIRS, SCOS, EEG, blood pressure, CO2, ECG, and stimulation logs. Modal extraction is then performed using established methods such as numerical algorithms for subspace statespace system identification (N4SID), dynamic mode decomposition with control (DMDc), sparse identification of nonlinear dynamics (SINDy), stochastic subspace identification (SSI), empirical mode decomposition (EMD), stabilization-diagram analysis. Candidate modes are evaluated for reproducibility, algorithm agreement, and physiological interpretability. Modes satisfying the criteria are further tested for L(t)-matrix identifiability and mechanistic separation between dCA-dominant and NVC-dominant dynamics. Accepted modes are then used as targets for adaptive TES, with fallback procedures applied when any validation gate fails.DETAILED DESCRIPTION

[0041] The following description is provided to enable a person skilled in the art to make and use the disclosed subject matter. The description is not intended to limit the scope of the claims. Features described with respect to one embodiment may be combined with features described with respect to any other embodiment unless expressly indicated otherwise. The terms “comprising,” “including,” and “having” are open-ended and do not exclude additional elements, components, functions, or steps. The present disclosure relates to wearable, portable, bedside, or ambulatory systems and methods for hierarchical cerebrovascular modal analysis and adaptive neurovascular modulation, including modulation of neurovascular-unit dynamics, collateral haemodynamics, perivascular flow, paravascular flow, glymphatic-flow-related dynamics, and systemic-to-cerebral haemodynamic coupling.System Overview

[0042] As illustrated generally in FIG. 1, a wearable or portable closed-loop neurovascular modulation platform is provided. The platform may include one or more central stimulation effectors, one or more peripheral haemodynamic augmentation effectors, one or more multimodal sensing subsystems, and a controller. The controller is configured to acquire, synchronize, and process physiological data; identify haemodynamic modes orPT / 2026 / 12765neurovascular state variables; compute or select stimulation-control parameters; and update one or more stimulation effectors based on measured physiological responses.

[0043] The system may operate in an open-loop, semi-closed-loop, clinician-supervised closed-loop, human-in-the-loop, or fully closed-loop mode. In a human-in-the-loop mode, proposed stimulation-control updates may be displayed to a clinician or operator for approval, modification, or rejection before implementation. In an autonomous or semi-autonomous mode, stimulation-control updates may be implemented only when predefined signal-quality, safety, synchronization, and mode-confidence criteria are satisfied.

[0044] The system may be used in a hospital, ambulance, rehabilitation clinic, home, laboratory, or ambulatory setting. The system may be configured for single-session assessment, repeated therapeutic sessions, longitudinal neurovascular monitoring, research investigation, clinical decision support, or adaptive stimulation of cerebrovascular, perivascular, paravascular, glymphatic, or systemic-to-cerebral haemodynamic dynamics.Central and Peripheral Stimulation Effectors

[0045] The central stimulation effector may include, without limitation, transcranial electrical stimulation (TES), high-definition transcranial electrical stimulation, transcranial direct-current stimulation (TDCS), transcranial alternating-current stimulation, transcranial oscillating-current stimulation, temporal interference stimulation, galvanic vestibular stimulation, transcranial magnetic stimulation, focused ultrasound stimulation, optical stimulation, transcutaneous or percutaneous stimulation of craniofacial autonomic targets, stimulation of a sphenopalatine ganglion or related parasympathetic pathway, stimulation of a superior cervical ganglion, stimulation of a carotid-region pathway, stimulation of a mastoid-region pathway, stimulation of an associated cervical sympathetic pathway, or combinations thereof.

[0046] The peripheral haemodynamic augmentation effector may include, without limitation, neuromuscular electrical stimulation, intermittent pneumatic compression, remote ischaemic conditioning, limb stimulation, cyclic cuff inflation and deflation, pneumatic garment actuation, or combinations thereof. Peripheral stimulation may be delivered to the common peroneal nerve, gastrocnemius, soleus, calf muscle pump, thighPT / 2026 / 12765muscles, upper limb muscles, or another peripheral target selected to modify venous return, cardiac preload, peripheral vascular resistance, endothelial signaling, systemic pressure, autonomic state, or systemic-to-cerebral haemodynamic coupling.

[0047] As illustrated in FIG. 6, an infrazygomatic electrode or electrode pair may be positioned relative to the zygomatic arch, infrazygomatic crest, alveolar process, nasal ala, tragus, or other craniofacial anatomical landmark to direct current toward the sphenopalatine ganglion, otic ganglion, or related parasympathetic pathway. In certain embodiments, stimulation parameters are adjusted based on a measured cerebral, temporalartery, carotid, or collateral-territory haemodynamic response.

[0048] As illustrated in FIG. 7, a mastoid-carotid stimulation and optical sensing montage may include a mastoid-region stimulation electrode, a carotid-region stimulation electrode, a temporal- artery optical flow sensor, and one or more carotid-region, mastoid-region, or peri- auricular optical flow sensors. In certain embodiments, the temporal-artery optical flow sensor comprises a speckle contrast optical spectroscopy (SCOS) sensor positioned over or adjacent to the superficial temporal artery, and the mastoid -region optical flow sensor is positioned in a mastoid or post-auricular region adjacent to the posterior auricular artery. The mastoid-region and carotid-region electrodes may form a stimulation pair configured to modulate cervical autonomic, carotid-body, carotid-sinus, baroreflex -related, superior-cervical-ganglion-associated, peri- auricular vascular, or carotid-region pathways. The optical flow sensors may acquire blood-flow-index, vascular flow, pulsatility, vasomotor, or stimulation-evoked haemodynamic information during stimulation. In certain embodiments, SCOS signals acquired from the temporal- artery sensor, mastoid-region sensor, carotid-region sensing location, or combinations thereof are received by a controller, and the controller adjusts TES amplitude, frequency, phase, duty cycle, electrode weighting, or timing based on the measured vascular response, thereby enabling closed-loop assessment or modulation of peri- auricular, carotid-region, cervical-autonomic, or systemic-to-cerebral haemodynamic dynamics.

[0049] As illustrated in FIG. 4, patterned transcranial or transcutaneous electrical stimulation may be applied to modulate perivascular autonomic pathways and induce oscillatory responses in pial vessels. In some embodiments, stimulation-induced pressure pulses in pial vessels and very-low-frequency haemodynamic oscillations are used toPT / 2026 / 12765modify flow resistance at a pre-capillary, penetrating-arteriole, capillary, perivascular, or paravascular compartment. The controller may adjust stimulation frequency, amplitude, phase, duty cycle, electrode weighting, or waveform based on measured haemodynamic responses, simulated pressure-flow relationships, or modal features extracted from optical or ultrasound signals.

[0050] As illustrated in FIG. 5, peripheral stimulation may be synchronized with systemic physiological rhythms. In certain embodiments, neuromuscular electrical stimulation bursts, intermittent pneumatic compression cycles, or remote ischaemic conditioning cycles are timed relative to ECG-derived cardiac phase, venous-return augmentation, respiratory phase, vasomotor phase, or cerebral haemodynamic response. The controller may coordinate peripheral NMES or IPC-induced global oscillations with central TES, TMS, focused ultrasound, optical stimulation, or perivascular stimulation-induced local vascular responses.

[0051] As illustrated in FIG. 8, global-autonomic perturbation effectors may include galvanic vestibular stimulation (GVS) delivered through mastoid electrodes and temporal interference stimulation directed toward a deep autonomic target such as a fastigial nucleus or brainstem-related autonomic pathway. Such perturbations may be used to probe or modulate global dynamic cerebral autoregulation (dCA) while scalp-mounted SCOS, fNIRS, and EEG sensors acquire haemodynamic and neurophysiological responses.

[0052] In certain embodiments, near-infrared spectroscopy, speckle contrast optical spectroscopy, near-infrared transillumination or backscattering, transcranial Doppler ultrasound, or imaging-derived information obtained prior to or independently of wearable device operation may be used to monitor vascular or perivascular responses to stimulation. Such measurements may be used to estimate changes in haemoglobin concentration, bloodflow index, flow velocity, tissue compliance, cerebrospinal-fluid-associated oscillations, perivascular-space dynamics, or intracranial-compliance-related variation. Glymphatic-flow-related dynamics may be estimated indirectly from one or more physiological signals associated with perivascular, paravascular, cerebrospinal-fluid, tissue-compliance, or low-frequency haemodynamic oscillatory activity, through known physiological relationships between perivascular- space fluid movement, arterial pulsatility, respiratory driving, and low-frequency vasomotion. The controller may derive a glymphatic-flow-related parameter — comprising a low-frequency oscillatory component, a phase relationship betweenPT / 2026 / 12765cardiac, respiratory, and very-low-frequency haemodynamic modes, a cerebrospinal-fluid-associated oscillation, a perivascular-space-associated optical variation, or a tissue-compliance-related signal — from near-infrared transillumination, near-infrared backscattering, speckle contrast optical spectroscopy, functional near-infrared spectroscopy, transcranial Doppler ultrasound, respiratory signals, cardiac pulsatility, or multimodal state-space outputs. The controller may use the glymphatic-flow-related parameter as an input to target-mode selection, safety gating, model predictive control, or stimulation-control instruction generation.

[0053] In certain embodiments, a computational model of a neurovascular unit, pial vessel, penetrating arteriole, capillary bed, perivascular compartment, paravascular compartment, or systemic circulation may be used to simulate a haemodynamic response to stimulation. The controller may compare simulated responses with measured responses and update stimulation parameters accordingly.Sensor-Stimulator Units and Watershed Placement

[0054] As illustrated in FIG. 9, an image-guided closed-loop sensor- stimulator unit may include a sensing module, a modal analysis module, and a TES actuation module. The unit may further include stimulation electrodes, light-emitting diodes, vertical-cavity surfaceemitting lasers, photodetectors, CMOS detectors, EEG electrodes, local electronics, and wireless communication components. In operation, the sensing module acquires local haemodynamic or neurophysiological signals, the modal analysis module identifies haemodynamic modes or state variables, and the TES actuation module adjusts stimulation parameters according to the identified mode or state variable.

[0055] As further illustrated in FIG. 9, distributed scalp-mounted fNIRS-SCOS sensorstimulator units may be positioned over a cortical region corresponding to an ischaemic core, a surrounding penumbra, a vascular blockage, or an alternative collateral-flow route. Each unit may comprise one or more stimulation electrodes, optical sources, optical detectors, EEG electrodes, a local controller, and wireless communication hardware. A base station or controller may receive physiological signals from the units, perform modal analysis, reconstruct haemodynamic or perfusion maps, and adapt TES actuation to modulate collateral perfusion toward a target region.PT / 2026 / 12765

[0056] As illustrated in FIG. 2A, one or more sensor- stimulator units may be positioned over cerebral watershed territories, including ACA-MCA and MCA-PCA border-zone regions. Placement may be determined using anatomical landmarks, atlas-derived vascular territory maps, perfusion imaging, optical topography, transcranial Doppler measurements, or combinations thereof. The stimulation electrodes may be arranged around the target watershed territory, while optical or electrophysiological sensing elements acquire local signals from substantially the same vascular territory. This co-registration enables modal-analysis-guided stimulation of collateral perfusion in vulnerable watershed regions.Multimodal Sensing and Matched- Territory Configuration

[0057] The cerebral sensing subsystem may include functional near-infrared spectroscopy, speckle contrast optical spectroscopy, near-infrared transillumination or backscattering, electroencephalography, transcranial Doppler ultrasound, or combinations thereof. The peripheral or systemic sensing subsystem may include muscle near-infrared spectroscopy, electromyography, electrocardiography, blood pressure monitoring, pulse-wave analysis, plethysmography, respiratory monitoring, carbon dioxide monitoring, limb perfusion sensing, or combinations thereof.

[0058] As illustrated in FIG. 3, a depth -resolved optical sensing subsystem may include a light-emitting diode or vertical-cavity surface-emitting laser and a linear photodiode array, CMOS detector, or other detector array. Progressively increasing source-detector separations may sample different depths, including superficial pial artery regions, penetrating arteriole regions, and deeper capillary or parenchymal microvascular regions. Frequency-domain and modal-analysis outputs may be used to distinguish pial, penetrating arteriole, and capillary haemodynamic modes.

[0059] As illustrated in FIG. 16, a matched-territory multimodal sensing platform may coregister at least two sensing modalities to a common vascular territory, such as an anterior cerebral artery territory, middle cerebral artery territory, posterior cerebral artery territory, or watershed territory. In one embodiment, TCD insonation provides macrovascular cerebral blood-flow velocity from a named artery; fNIRS optodes acquire haemoglobin oxygenation signals from cortical tissue supplied by the same artery; SCOS source-detector pairs provide an oxygenation-independent blood-flow index; and EEG electrodes providePT / 2026 / 12765neural reference signals for neurovascular-coupling discrimination.

[0060] SCOS blood-flow index and fNIRS haemoglobin signals provide complementary observables. In embodiments using a two-level state-space model, co-located SCOS and fNIRS measurements may increase the rank of an observability matrix relative to either modality alone and may reduce uncertainty in estimating cross-level coupling parameters.

[0061] All sensing streams and stimulation logs may be synchronized to a common time base. In one embodiment, device streams are acquired using Lab Streaming Layer outlets and stored as XDF files. Hardware trigger pulses may be delivered to multiple devices for timing verification. A session may be accepted for modal analysis or adaptive stimulation only when inter-device timing error satisfies a predetermined threshold, for example not more than about 5 milliseconds.Vascular- Territory and Perfusion-Imaging Guidance

[0062] As illustrated in FIG. 2A, a vascular-territory map may relate sensor and stimulator locations to ACA, MCA, PCA, and watershed territories. The map may represent source arteries, collateral pathways, pial networks, penetrating arterioles, capillary networks, vascular obstruction, and alternative supply routes. The map may be atlas -derived, image-derived, subject-specific, or updated using real-time physiological measurements.

[0063] As illustrated in FIG. 13, perfusion imaging may identify an ischaemic core, salvageable penumbra, benign oligemia, vascular occlusion, collateral pathway, or perfusion-diffusion mismatch. The imaging information may be obtained from MR perfusion, CT perfusion, arterial spin labelling, diffusion-weighted imaging, perfusion-weighted imaging, angiography, or combinations thereof. The imaging-derived target map may be transformed into scalp coordinates or a wearable-device coordinate system to guide placement of fNIRS, SCOS, EEG, and TES components.

[0064] In certain embodiments, imaging-derived information is provided to a human-in-the-loop model predictive controller. The controller may use the imaging -derived target map, objectives, and constraints to select an optical sensing montage, electrophysiological sensing montage, and stimulation montage over the target penumbra or collateral territory. The montage may be configured to monitor haemodynamic responses and deliverPT / 2026 / 12765stimulation to support collateral recruitment from an arterial source toward the penumbra.Modal Analysis and Physiological Mode Classification

[0065] As illustrated in FIG. 2B, the controller may implement a modal-analysis workflow configured to classify physiological haemodynamic modes. A stabilization diagram may identify stable modal components corresponding to cardiac oscillations, respiratory oscillations, and neurovascular oscillations, including cross-frequency coupling between oscillatory bands. Identified modes may be subjected to multi-stage clustering based on haemodynamic, spatial, spectral, and physiological features.

[0066] The resulting physiological clusters may distinguish respiratory-related modes, arteriolar or myogenic low-frequency modes, and very-low-frequency microvascular, metabolic, or neurovascular modes. The classified modes may be used by the controller to select, modify, or suppress stimulation parameters for closed-loop neurovascular modulation.

[0067] As illustrated in FIG. 15, each identified haemodynamic mode may be characterized using spatial and temporal features to distinguish brain-related modes from scalp, systemic, superficial, or artef actual components. A candidate mode may be represented by a mode shape across measurement channels and a corresponding modal time series. The controller may compute a global correlation metric, total-haemoglobin global correlation, correlationbased signal improvement, graph-Laplacian spatial energy, and a band- specific physiological prior.

[0068] The controller may combine the features into a composite brain-scalp score. A mode may be classified as brain-like, scalp / systemic-like, or mixed / penetrating-artery-like based on the score and a selected classification margin. Modes classified as scalp / systemic-like may be excluded from adaptive stimulation targeting, while brain-like or sufficiently confident mixed modes may be eligible for Accepted Reproducible Mode evaluation.Two-Level State-Space Model and Cross-Level Coupling Matrix

[0069] As illustrated in FIG. 14, the controller may implement a two-level hierarchical state- space model of cerebrovascular regulation. A first level represents globalPT / 2026 / 12765macrovascular autoregulatory dynamics, and a second level represents local microvascular neurovascular-coupling dynamics. Interaction between the levels is represented by a crosslevel coupling matrix L(t).

[0070] In one embodiment, Level 1 dynamics are represented by xi(t+l)=Ai(t)xi(t)+Bi(t)u_global(t)+wi(t), where xi includes latent macrovascular state variables such as cerebral perfusion pressure, large-vessel compliance, or macrovascular smooth-muscle tone; u_global includes systemic inputs such as arterial blood pressure, end-tidal or transcutaneous CO2, cardiac output, or peripheral vascular resistance; Ai(t) is a time-varying state transition matrix; Bi(t) is a time-varying input matrix; and wi is process noise.

[0071] Level 2 dynamics may be represented by X2(t+l)=A2(t)x2(t)+L(t)xi(t)+B2(t)u_local(t)+w2(t), where X2 includes latent microvascular state variables such as arteriolar smooth-muscle tone, pericyte contractility, endothelial nitric oxide, astrocytic calcium dynamics, or metabolic state; u_local includes local inputs such as neural activity, EEG bandpower, task input, or stimulation input; A^t) is a timevarying state transition matrix; B2(t) is a time- varying input matrix; and W2 is process noise.

[0072] The joint observation equation may be represented by y(t)=Ci(t)xi(t)+C2(t)x2(t)+v(t), where y(t) comprises multimodal measurements including TCD cerebral blood-flow velocity (CBFv), fNIRS HbO / HbR / HbT, SCOS BFi, EEG features, blood pressure, carbon dioxide, ECG, muscle NIRS, or stimulation logs, and v(t) represents measurement noise.

[0073] The time-varying cross-level coupling matrix L(t) quantifies propagation of global macrovascular haemodynamic forcing into local microvascular dynamics. A relatively high norm or selected entry of L(t) may indicate pressure-passive microvascular behaviour or impaired autoregulatory separation. A relatively low norm of L(t) together with preserved local neurovascular gain may indicate intact autoregulatory separation. The controller may compute a spectral norm, selected matrix entries, confidence intervals, subject- specific reference values, or sliding-window estimates as biomarkers, control variables, or targetselection criteria.

[0074] The controller may estimate the two-level model using subspace state-spacePT / 2026 / 12765identification, N4SID, DMDc, sparse identification of nonlinear dynamics, stochastic subspace identification, operational modal analysis (OMA), Kalman filtering, Bayesian inference, or combinations thereof. The controller may compute confidence intervals for L(t) over sliding windows, assess L(t) identifiability using an observability matrix condition number, and revert to output-only modal analysis when L(t) is not identifiable.

[0075] Uncertainty in estimated state variables, model parameters, modal features, or the cross-level coupling matrix L(t) may be quantified using one or more uncertainty metrics, including confidence interval width, posterior covariance, bootstrap variance, Kalman-filter covariance, observability Gramian condition number, Fisher information, residual prediction error, cross-validation error, or an identifiability score. The controller may use the uncertainty metric to select a sensing configuration, reject a candidate model, switch to output-only modal analysis, select a different Accepted Reproducible Mode, or inhibit adaptive stimulation.

[0076] In certain embodiments, the controller implements a machine-learned or hybrid state-space model in which one or more state-transition, observation, stimulation-response, uncertainty-estimation, or cross-level coupling functions are parameterised using a model trained on synchronized multimodal physiological data, population datasets, or digital-twin simulations. The model may include a neural ordinary differential equation, Koopman operator model, sparse nonlinear dynamics model, Gaussian-process state-space model, recurrent neural network, transformer-based temporal model, or physics-informed neural network. The model output may include a latent macrovascular state, latent microvascular state, stimulation-response estimate, cross-level coupling estimate, or uncertainty estimate used for stimulation-control selection or safety gating.

[0077] In certain embodiments, the controller implements a sparse-identification reinforcement-learning module for interpretable and data-efficient model-based control. The module may combine sparse identification of nonlinear dynamics with reinforcement learning to learn at least one of a sparse dynamics model, sparse reward model, sparse value function, or sparse stimulation-control policy from synchronized multimodal physiological data, digital-twin rollouts, structured perturbation data, or previously observed stimulationresponse data.

[0078] The sparse-identification reinforcement-learning module may construct aPT / 2026 / 12765dictionary of candidate nonlinear terms from haemodynamic state variables, modal features, stimulation parameters, physiological constraints, phase-coupling metrics, crosslevel coupling estimates, or derived biomarkers, and may select a sparse subset of terms representing the dynamics model, reward function, or control policy. The resulting sparse representation may be used to generate an interpretable stimulation-control policy executable on an embedded or wearable controller.

[0079] In certain embodiments, the reinforcement- learning reward is constrained by physiological safety and interpretability criteria. The reward may include one or more terms corresponding to target haemodynamic mode amplitude, damping, phase, cross-level coupling matrix norm, collateral-flow index, neurovascular-coupling gain, systemic-to-cerebral coupling index, stimulation energy, charge density, signal quality, or adverse physiological response. The controller may reject or modify a learned policy when the sparse representation violates safety constraints, observability criteria, Accepted Reproducible Mode criteria, or clinician-defined limits.

[0080] In certain embodiments, the sparse-identification reinforcement-learning module uses a digital twin or reduced-order physiological model as a surrogate environment to reduce subject interactions during policy training. The policy may be trained, validated, or stress-tested in simulation before being presented to a clinician or applied to the subject. The sparse policy may be updated online or between sessions using subject- specific data while maintaining a bounded number of active terms to preserve interpretability and computational efficiency.Structured Orthogonal Perturbation and Mode Attribution

[0081] As illustrated in FIG. 17, the system may apply or process a structured orthogonal perturbation set to separate Level 1 global macrovascular dynamics from Level 2 local neurovascular-coupling dynamics. The perturbation set may include spontaneous physiological perturbations, task-evoked perturbations, and externally delivered stimulation perturbations.

[0082] Level 1 perturbations may include spontaneous blood-pressure fluctuations, paced breathing or controlled carbon dioxide modulation, galvanic vestibular stimulation, intermittent pneumatic compression, squat- stand manoeuvres, remote ischaemicPT / 2026 / 12765conditioning, mastoid-carotid transcutaneous stimulation, superior cervical ganglion stimulation, sphenopalatine ganglion stimulation, or temporal interference stimulation targeting a deep autonomic brain structure. Level 2 perturbations may include finger tapping, photic stimulation, cognitive task input, sensory stimulation, task-evoked neurovascular activation, high-definition transcranial electrical stimulation, focal transcranial alternating-current stimulation, transcranial magnetic stimulation, focused ultrasound stimulation, optical stimulation, or task-locked stimulation.

[0083] A mode that preferentially responds to global perturbations and exhibits a response to local neurovascular perturbations below a predefined response threshold may be classified as dynamic cerebral autoregulation-dominant. A mode that preferentially responds to local neurovascular perturbations with territory- specific EEG coherence and exhibits a response to global autonomic perturbations below a predefined response threshold may be classified as neurovascular-coupling-dominant. Modes satisfying neither classification may be classified as mixed, CCE-reactive, systemic, or uncertain.

[0084] As further illustrated in FIG. 17, high-definition TES applied over a precentralgyrus or primary-motor-cortex territory may be used as a local Level 2 perturbation, galvanic vestibular stimulation may be used as a Level 1 vestibulo-autonomic perturbation, and fastigial-nucleus temporal interference stimulation may be used as a deep Level 1 autonomic perturbation. Comparative haemodynamic response signatures from TCD, SCOS, fNIRS, EEG, blood pressure, and systemic inputs may be used to attribute identified modes to NVC-dominant, dCA-dominant, CO2-reactive, or mixed mechanisms.Accepted Reproducible Modes and Gate-Based Validation

[0085] As illustrated in FIG. 18, the controller may identify an Accepted Reproducible Mode by applying reproducibility, algorithm- agreement, and physiological-interpretability criteria. Reproducibility may be assessed using intraclass correlation coefficient, test-retest frequency stability, damping stability, spatial loading similarity, or a comparable reproducibility metric. The intraclass correlation coefficient threshold may be selected from population reference data, subject-specific baseline data, digital-twin simulation data, session-to-session variability, or a clinician- or protocol-defined acceptance criterion. Algorithm agreement may be assessed across N4SID, DMDc, SSI, SINDy, OMA,PT / 2026 / 12765experimental modal analysis (EMA), or another modal identification method. Physiological interpretability may be assessed using frequency band, perturbation response, spatial loading, TCD contribution, fNIRS contribution, SCOS contribution, EEG coherence, or consistency with the structured perturbation matrix.

[0086] In certain embodiments, a gate -based validation procedure is implemented. A first gate confirms platform synchronization and data quality. A second gate confirms reproducible mode extraction. A third gate confirms mechanistic separation between regulatory mode families or confirms identifiability of the cross-level coupling matrix. A fourth gate confirms that adaptive stimulation produces a target modal change exceeding session-to-session variability or another predefined control criterion.

[0087] If a confidence score, gate criterion, signal-quality metric, safety metric, or identifiability metric is not satisfied, the controller may execute a fallback procedure, including observation-only mode, output-only modal analysis, reduced sensor-set analysis, clinician-supervised mode, stimulation suspension, or selection of a different target mode.Phase- Coupled Global and Local Oscillation Control

[0088] In certain embodiments, the controller identifies a first haemodynamic mode classified as a global or dynamic cerebral autoregulation-dominant mode and a second haemodynamic mode classified as a local neurovascular-coupling-dominant mode. The controller estimates a phase relationship between the first and second modes and generates coordinated control instructions for a first effector configured to modulate the global mode and a second effector configured to modulate the local mode.

[0089] The first effector may include NMES, intermittent pneumatic compression, remote ischaemic conditioning, galvanic vestibular stimulation, mastoid-carotid stimulation, superior cervical ganglion stimulation, sphenopalatine ganglion stimulation, or temporal interference stimulation targeting a deep autonomic structure. The second effector may include primary motor cortex high-definition TES, focal transcranial alternating-current stimulation (tACS), transcranial oscillating-current stimulation (tOCS), TMS, focused ultrasound stimulation, optical stimulation, or task-locked stimulation configured to modulate a local neurovascular-coupling response.PT / 2026 / 12765

[0090] The controller may maintain, advance, delay, entrain, suppress, amplify, or adjust a selected phase offset or coupling relationship between the global and local modes. The phase relationship may include phase-locking value, phase lag, coherence, cross-frequency coupling, phase-amplitude coupling, transfer-function phase, Hilbert-transform phase, wavelet phase, modal phase, or state-space phase estimate.Adaptive Stimulation Control and Safety

[0091] For a neurovascular-coupling-dominant target mode, a spatially targeted Level 2 effector such as cortical high-definition TES may be selected. The controller may adjust current amplitude, stimulation frequency, waveform, phase, pulse width, duty cycle, electrode weighting, stimulation duration, or stimulation montage to amplify, suppress, phase-align, or shift the selected mode toward a desired criterion.

[0092] For a dynamic-cerebral-autoregulation-dominant target mode, a Level 1 global or autonomic effector such as GVS, mastoid-carotid stimulation, SCG stimulation, SPG stimulation, temporal interference stimulation targeting a deep autonomic structure, IPC, NMES, or RIC may be selected. The controller may adjust the effector based on Level 1 state trajectory, L(t)-matrix estimates, systemic haemodynamic inputs, or modal response.

[0093] As illustrated in FIGS. 11 and 12, a human-in-the-loop model predictive controller may use fNIRS haemoglobin spectra, near-infrared transillumination, subarachnoid- space oscillation signals, systemic haemodynamic signals, pial and penetrating artery models, and safety constraints to select stimulation parameters. Spectral features may include cardiac, respiratory, low-frequency, and very-low-frequency oscillatory components.

[0094] As further illustrated in FIG. 12, the human-in-the-loop model predictive control framework may coordinate one or more peripheral haemodynamic augmentation effectors with one or more cerebral or perivascular stimulation effectors based on identified haemodynamic oscillatory modes. The peripheral effectors may include neuromuscular electrical stimulation, intermittent pneumatic compression, or remote ischaemic conditioning configured to generate venous-return, cardiac -preload, pressure, or vasomotor oscillations. The cerebral or perivascular effectors may include transcranial electrical stimulation (TES), transcranial magnetic stimulation (TMS), focused ultrasoundPT / 2026 / 12765stimulation, optical stimulation, or perivascular nerve stimulation configured to modulate pial vessel dynamics, penetrating arteriole dynamics, neurovascular-unit signaling, or local metabolic vasodilatory signaling.

[0095] In certain embodiments, transcranial electrical stimulation (TES) or transcranial magnetic stimulation (TMS) is delivered to a cortical or perivascular target to evoke neural, glial, endothelial, or metabolic signaling associated with local vasodilation. The evoked signaling may propagate in a direction from a capillary, pre-capillary, or penetratingarteriole region toward an upstream penetrating arteriole or pial vessel segment. The controller may time such vasodilatory signaling with a pial or penetrating-arteriole pressure wave generated by perivascular nerve stimulation, transcranial electrical stimulation (TES), transcranial magnetic stimulation (TMS), or another cerebral stimulation effector, and may further coordinate the pressure wave with IPC- or NMES-mediated venous-return augmentation.

[0096] Near-infrared transillumination, near-infrared spectroscopy, speckle contrast optical spectroscopy, transcranial Doppler ultrasound, electroencephalography, bloodpressure, electrocardiography, respiratory, or carbon-dioxide signals may be provided to the controller as feedback. The controller may use model prediction, optimisation, modal analysis, and safety constraints to adjust stimulation timing, phase, amplitude, duty cycle, frequency, or sequence. The fNIRS HbO power spectra may include cardiac, respiratory, low-frequency, and very-low-frequency oscillatory components used to guide stimulation for neurovascular, perivascular, metabolic, or systemic-to-cerebral haemodynamic regulation.

[0097] In certain embodiments, very-low-frequency haemodynamic oscillations, stimulation-induced pial pressure waves, and vasodilatory signaling are coordinated to modify flow resistance at a pre-capillary sphincter, penetrating arteriole, capillary entrance, perivascular compartment, or local resistance bottleneck. In no-reflow conditions, autoregulatory dysfunction, collateral-flow insufficiency, or spatially restricted low-perfusion states, the controller may generate coordinated stimulation-control instructions configured to modify local vascular compartment resistance or flow distribution within predefined safety constraints. In certain embodiments, the controller detects or estimates spatially restricted low-perfusion regions, hypoxia-associated haemodynamic modes, or spatially restricted low-oxygenation events from near-infrared spectroscopy, specklePT / 2026 / 12765contrast optical spectroscopy, near-infrared transillumination, transcranial Doppler ultrasound, or multimodal state-space outputs, and adjusts the timing of IPC, NMES, TES, TMS, or perivascular stimulation to modify downstream vascular resistance while remaining within predefined safety constraints.

[0098] In certain embodiments, the controller estimates an oedema-associated physiological parameter from one or more multimodal signals, including near- infrared transillumination, near-infrared backscattering, functional near-infrared spectroscopy, speckle contrast optical spectroscopy, transcranial Doppler ultrasound, blood pressure, carbon dioxide, or multimodal state-space outputs. The oedema-associated physiological parameter may comprise a tissue-water-related optical change, cerebral blood-volume change, intracranial-compliance-related variation, cerebrospinal-fluid-associated oscillation, low-frequency haemodynamic mode, or spatially restricted low-oxygenation pattern. The controller may use the oedema- associated physiological parameter as an input to a model predictive control cost function, safety layer, target-mode selection procedure, or stimulation-control instruction.

[0099] In certain embodiments, the controller monitors the oedema-associated physiological parameter over a plurality of time windows and generates stimulation-control instructions configured to bias the parameter toward a target range associated with reduced tissue-water signal, improved intracranial-compliance estimate, reduced oedema-associated haemodynamic abnormality, or improved local flow-resistance estimate, without requiring a determination of clinical oedema resolution.

[0100] The controller may implement a model predictive control cost function including a target mode objective, a stimulation-energy penalty, a physiological safety penalty, and one or more constraints. Constraints may include maximum current, maximum charge density, maximum pulse width, maximum duty cycle, stimulation duration, skin impedance, temperature, blood pressure change, heart rate change, signal quality, artefact level, and Accepted Reproducible Mode confidence score.

[0101] If a safety constraint is violated or a confidence metric falls below a threshold, the controller may inhibit, limit, modify, or suspend stimulation and revert to observation-only mode. In certain embodiments, no adaptive stimulation update is delivered without a valid concurrent mode estimate satisfying a confidence threshold.PT / 2026 / 12765Peripheral Calf Muscle Pump and Muscle NIRS Integration

[0102] As illustrated in FIG. 10, a limb-worn assembly may comprise a calf sleeve, sock, band, or garment integrating neuromuscular electrical stimulation electrodes, electromyography electrodes, muscle near-infrared spectroscopy optical source-detector windows, and optional speckle contrast optical spectroscopy source-detector pairs.

[0103] The controller may use muscle NIRS, EMG, ECG, blood pressure, pulsewave analysis, and cerebral optical signals to adjust NMES parameters, including amplitude, frequency, pulse width, burst duration, duty cycle, and timing. The controller may synchronize peripheral stimulation with central stimulation to modify systemic-to-cerebral haemodynamic coupling.

[0104] In certain embodiments, the peripheral assembly supports intermittent pneumatic compression or remote ischaemic conditioning in addition to, or instead of, NMES. The controller may coordinate cuff inflation, cuff deflation, NMES bursts, cardiac phase, respiratory phase, vasomotor phase, and cerebral haemodynamic response.Digital Twin and Implementation

[0105] As illustrated in FIG. 1, FIG. 11, and FIG. 14, a digital twin or reduced-order model may represent systemic, cerebral, pulmonary, cardiac, peripheral, pial, penetrating arteriole, capillary, perivascular, or paravascular haemodynamic compartments. The digital twin may generate synthetic physiological datasets, simulate stimulation effects, benchmark modal analysis algorithms, and support subject-specific calibration.

[0106] The systems and methods may be implemented at least in part using one or more processors executing instructions stored on a non-transitory computer-readable medium. The processors may include a microcontroller, digital signal processor, embedded processor, application- specific integrated circuit, field-programmable gate array, mobile computing device, server, or distributed computing architecture.

[0107] The non-transitory computer-readable medium may store instructions forPT / 2026 / 12765data acquisition, synchronization, preprocessing, artefact rejection, modal analysis, statespace estimation, brain-scalp classification, L(t)-matrix estimation, target-mode selection, safety checking, model predictive control, stimulation command generation, clinician display, data storage, or communication with external systems.Methods of Operation

[0108] In one embodiment, a method comprises positioning a wearable sensing and stimulation assembly on a subject; acquiring baseline cerebral and systemic physiological signals; identifying haemodynamic modes; estimating neurovascular state variables; selecting a target mode or target vascular territory; generating a stimulation -control instruction; measuring stimulation-evoked responses; and updating stimulation parameters based on real-time feedback.

[0109] In another embodiment, a method comprises acquiring cerebral haemodynamic signals using fNIRS, SCOS, near-infrared transillumination, TCD, EEG, or combinations thereof; acquiring peripheral or systemic feedback using muscle NIRS, EMG, ECG, blood pressure, respiration, carbon dioxide, pulse-wave analysis, or combinations thereof; estimating modal parameters or state variables; and dynamically adjusting stimulation to modulate cerebrovascular reactivity, collateral perfusion, neurovascular coupling, perivascular flow, paravascular flow, glymphatic flow, or systemic-to-cerebral haemodynamic coupling.

[0110] In another embodiment, a method comprises receiving imaging information identifying at least one of an ischaemic core, penumbra, collateral pathway, vascular territory, or perfusion-diffusion mismatch; selecting a sensing montage and a stimulation montage based on the imaging information; acquiring real-time physiological feedback; and generating stimulation-control signals according to a target mode or state variable.

[0111] In another embodiment, a method comprises applying or processing a structured perturbation set including at least one global-autonomic perturbation and at least one local neurovascular perturbation; extracting haemodynamic modes from resulting signals; and classifying the modes according to convergence of responses across perturbation conditions.PT / 2026 / 12765

[0112] As illustrated in FIG. 17 and consistent with the method of Claim 22, in another embodiment a method comprises identifying a global haemodynamic oscillation associated with systemic, autonomic, venous-return, pressure, carbon-dioxide, or dynamicautoregulation forcing; identifying a local haemodynamic oscillation associated with neurovascular coupling in a matched vascular territory; estimating a phase difference or coupling metric between the global haemodynamic oscillation and the local haemodynamic oscillation; and coordinating a global-mode effector and a local-mode effector to drive the phase difference or coupling metric toward a target range. In certain embodiments, the controller additionally or alternatively estimates a harmonic phase relationship between a dominant haemodynamic oscillatory component and at least one higher-order harmonic component of the same or a related haemodynamic mode, wherein both components are derived from fNIRS, SCOS, near-infrared transillumination, TCD, or multimodal statespace outputs. The controller may compute the harmonic phase relationship using Fourier phase, Hilbert-transform phase, wavelet phase, modal phase, cross-spectral phase, or statespace phase estimates. The harmonic phase relationship may comprise a phase difference, phase-locking value, harmonic phase ratio, phase-amplitude coupling metric, or crossfrequency coupling metric associated with oscillatory flow dynamics in a pial vessel, penetrating arteriole, pre-capillary compartment, perivascular compartment, paravascular compartment, or glymphatic-flow-related pathway. The controller may adjust stimulation frequency, phase, duty cycle, amplitude, electrode weighting, or timing to bias the global-local phase relationship, harmonic phase relationship, or coupling metric toward a target range, within predefined safety constraints including maximum current, charge density, and duty cycle limits.

[0113] The disclosed methods may be used for neurovascular assessment, cerebrovascular monitoring, mode identification, stimulation parameter optimisation, rehabilitation support, ambulatory monitoring, prehospital monitoring, acute stroke support, post-stroke rehabilitation, vascular cognitive impairment monitoring, cerebral small vessel disease monitoring, autonomic dysfunction assessment, perioperative brain monitoring, or research investigation.

Claims

PT / 2026 / 12765WE CLAIM:

1. A neurovascular modulation system comprising:a cerebral sensing module configured to acquire at least one cerebral haemodynamic signal from a subject;a peripheral or systemic sensing module configured to acquire at least one peripheral or systemic physiological signal from the subject;at least one non-invasive or minimally invasive stimulation or haemodynamic augmentation effector configured to deliver stimulation or haemodynamic augmentation to at least one central, autonomic, vascular, or peripheral target of the subject; anda controller operatively coupled to the cerebral sensing module, the peripheral or systemic sensing module, and the at least one non-invasive or minimally invasive stimulation or haemodynamic augmentation effector;wherein the controller is configured to:(a) synchronize the at least one cerebral haemodynamic signal with the at least one peripheral or systemic physiological signal to a common time base;(b) identify at least one haemodynamic mode or neurovascular state variable from the synchronized signals; and(c) adaptively control the at least one non-invasive or minimally invasive stimulation or haemodynamic augmentation effector based on the identified haemodynamic mode or neurovascular state variable so as to modify at least one temporal, spatial, frequency, phase, or amplitude characteristic of cerebrovascular reactivity, collateral perfusion, neurovascular coupling, perivascular flow, paravascular flow, glymphatic-flow-related dynamics, or systemic-to-cerebral haemodynamic coupling.

2. The system of claim 1, wherein the cerebral sensing module is configured for matched-territory sensing in which at least two sensing modalities are co -registered to a common cerebral vascular territory selected from an anterior cerebral artery, middle cerebral artery, posterior cerebral artery, or watershed territory; wherein the sensingPT / 2026 / 12765modalities comprise at least one of functional near-infrared spectroscopy, speckle contrast optical spectroscopy, electroencephalography, or transcranial Doppler ultrasound; and wherein functional near-infrared spectroscopy and speckle contrast optical spectroscopy co-located within the common territory provide complementary haemoglobin-oxygenation and oxygenation-independent blood-flow-index signals that together increase a rank of an observability matrix of a two-level state-space model relative to either modality alone, thereby reducing uncertainty in a cross-level coupling matrix estimate.

3. The system of claim 1, further comprising a data synchronisation subsystem comprising Lab Streaming Layer outlets, XDF file recording, and hardware trigger pulses simultaneously delivered to a plurality of devices, wherein the subsystem enforces a verified inter-device timing-error threshold before permitting modal analysis or adaptive stimulation, and wherein sessions failing the timing verification criterion are excluded from modal analysis.

4. The system of claim 1, wherein the controller implements a two-level state-space model comprising:a first level representing global macrovascular autoregulation, with state dynamics xi(t+l) = Ai(t) xi(t) + Bi(t) u_global(t) + wi(t), wherein u_global comprises at least one of arterial blood pressure, end-tidal or transcutaneous carbon dioxide, cardiac output, or peripheral vascular resistance, and wherein a carbon dioxide signal from a paced-breathing protocol or monitoring device is incorporated as an explicit first-level input to separate carbon dioxide-driven from neurovascular-coupling-driven haemodynamic variance;a second level representing local microvascular neurovascular coupling, with state dynamics X2(t+1) = A2(t) x2(t) + L(t) xi(t) + B2(t) u_local(t) + w2(t), wherein u_local comprises at least one of neural activity, electroencephalography-derived bandpower, or a local stimulation input; and a cross-level coupling matrix L(t); anda joint observation equation y(t) = Ci(t) xi(t) + C2(t) x2(t) + v(t);wherein the controller computes a spectral norm or selected entry of L(t) as a real-time biomarker of cerebrovascular autoregulatory integrity, assesses identifiability of L(t) using a condition number of an observability matrix or a confidence interval width, and reverts to output-only stochastic subspace identification when L(t) is non-identifiable.PT / 2026 / 127655. The system of claim 1, wherein the controller computes, for each candidate haemodynamic mode, a composite brain-scalp score based on one or more of global spatial correlation of the mode temporal course across sensing channels, total-haemoglobin global correlation, correlation-based signal improvement derived from oxyhaemoglobindeoxyhaemoglobin anti-correlation, graph Laplacian spatial energy of a mode spatial loading map, or band-specific physiological prior energy; classifies the mode as brain-related, scalp / systemic-related, or mixed; and excludes scalp / systemic-classified modes from adaptive stimulation targeting and Accepted Reproducible Mode identification.

6. The system of claim 1, wherein the controller processes a structured perturbation set comprising:at least one global-autonomic perturbation selected from spontaneous blood pressure fluctuation, paced-breathing carbon dioxide modulation, galvanic vestibular stimulation delivered via bilateral mastoid electrodes with reduced dependence on direct local cortical neurovascular-coupling activation, intermittent pneumatic compression, or temporal interference stimulation with a carrier frequency offset directed toward a fastigial nucleus of the cerebellum or another deep autonomic target; andat least one local neurovascular perturbation selected from task-evoked neurovascular activation, photic stimulation, sensory stimulation, or cortical high-definition transcranial electrical stimulation;wherein a haemodynamic mode that preferentially responds to global-autonomic perturbations is classified as dynamic cerebral autoregulation-dominant, and a mode that preferentially responds to local neurovascular perturbations with territory -specific electroencephalography coherence is classified as neurovascular-coupling-dominant.

7. The system of claim 1, wherein the controller is configured to identify a first haemodynamic mode classified as dynamic cerebral autoregulation-dominant and a second haemodynamic mode classified as neurovascular-coupling-dominant, estimate a phase relationship between the two modes using at least one of phase-locking value, coherence, wavelet phase, or state-space phase estimate, and generate coordinated stimulation-control instructions for a first effector configured to modulate the first mode and a second effectorPT / 2026 / 12765configured to modulate the second mode, so as to maintain, advance, delay, or entrain a selected phase offset or coupling relationship between the two modes.

8. The system of claim 1, wherein the controller is configured to identify an Accepted Reproducible Mode satisfying simultaneously:a reproducibility criterion comprising a test-retest intraclass correlation coefficient at or above a pre-specified threshold for at least one of modal frequency, damping ratio, or spatial loading, assessed across at least two independent sessions;an algorithm- agreement criterion comprising agreement between modes identified by at least two algorithms selected from N4SID, DMDc, SSI, sparse identification of nonlinear dynamics, or operational modal analysis, within pre-specified tolerance intervals; anda physiological-interpretability criterion classifying the mode as dynamic cerebral autoregulation-dominant or neurovascular-coupling-dominant based on perturbation response profile, electroencephalography coherence, and spatial loading across transcranial Doppler, functional near-infrared spectroscopy, and speckle contrast optical spectroscopy channels.

9. The system of claim 8, wherein the controller implements a gate-based validation procedure comprising a first gate confirming inter-device timing below a threshold, a second gate confirming identification of at least one Accepted Reproducible Mode, a third gate confirming cross-level coupling matrix identifiability or mechanistic mode separation, and a fourth gate confirming a measurable modal change from adaptive stimulation exceeding session-to-session variability; and wherein the controller executes a pre-specified fallback procedure and reverts to observation-only mode when any gate criterion is not met.

10. The system of claim 4, wherein the controller implements a state-space model in which at least one of a state-transition function, observation function, or cross-level coupling function is parameterised by a machine-learned model trained on synchronized multimodal physiological data, population physiological data, or digital-twin simulation data; wherein the machine-learned model outputs at least one uncertainty estimate associated with a latent macrovascular state, a latent microvascular state, or the cross-levelPT / 2026 / 12765coupling matrix L(t); and wherein the machine-learned model is constrained by at least one of physiological priors, observability criteria, or safety constraints.

11. The system of claim 10, wherein the machine-learned model comprises at least one of a neural ordinary differential equation, a Koopman operator model, a sparse nonlinear dynamics model, a Gaussian-process state-space model, a recurrent neural network, a transformer-based temporal model, or a physics-informed neural network trained with constraints from known cerebrovascular physiology.

12. The system of claim 1, wherein the controller comprises a sparse-identification reinforcement-learning module configured to learn or update at least one of a sparse dynamics model, sparse reward model, or sparse stimulation-control policy from synchronized multimodal physiological data, structured perturbation data, or digital-twin simulation data; wherein the sparse stimulation-control policy is limited to a predetermined maximum number of active dictionary terms or non-zero coefficients to maintain interpretability and computational efficiency for execution on an embedded or wearable processor; and wherein the policy is subject to one or more physiological safety constraints before delivery.

13. The system of claim 12, wherein the sparse-identification reinforcement-learning module constructs a dictionary of candidate nonlinear terms comprising at least one of haemodynamic state variables, modal features, stimulation parameters, cross-level coupling matrix estimates, phase-coupling metrics, physiological indices, or safetyconstraint variables; selects a sparse subset of the candidate nonlinear terms to form an interpretable stimulation-control policy; uses a reward function including at least one term corresponding to haemodynamic mode amplitude, cross-level coupling matrix norm, neurovascular-coupling gain, stimulation energy, or adverse physiological response; and trains, validates, or stress-tests the policy using a digital twin or reduced-order physiological model before applying it to a subject or presenting it to a clinician.

14. The system of claim 1, further comprising a candidate-generating module configured to generate, evaluate, rank, or refine a plurality of candidate physiological models or candidate stimulation-control policies by conditioning a generative, probabilistic, search-based, or simulator-conditioned model on at least one of synchronized multimodal physiological data, population haemodynamic data, or digital-twin simulationPT / 2026 / 12765outputs; to select or present a candidate based on one or more of stimulation safety limits, physiological response bounds, observability criteria, or model-identifiability criteria for the cross-level coupling matrix L(t); and, in a human-in-the-loop mode, to present the ranked candidate together with a predicted haemodynamic response trajectory, uncertainty estimate, and safety constraint summary for clinician or operator approval before stimulation is delivered.

15. The system of claim 1, further comprising a limb-mounted wearable assembly comprising neuromuscular electrical stimulation electrodes, muscle near-infrared spectroscopy optical source-detector windows, optional speckle contrast optical spectroscopy source-detector pairs, and surface electromyography electrodes integrated in a sleeve, sock, or garment; wherein the controller synchronizes neuromuscular electrical stimulation bursts with at least one of cardiac cycle phase, venous return augmentation, respiratory phase, or vasomotor phase; adjusts stimulation parameters using electromyographic amplitude feedback and muscle near-infrared spectroscopy haemodynamic feedback including detection of a biphasic contraction-reperfusion response comprising an initial compression-mediated decrease in muscle total haemoglobin followed by a post-contraction vasodilatory overshoot; and applies modal analysis to cerebral functional near-infrared spectroscopy or cerebral speckle contrast optical spectroscopy signals acquired during neuromuscular electrical stimulation to identify stimulation-perturbed cardiac -related haemodynamic modes, adjusting stimulation timing or frequency based on the identified modal response.

16. The system of claim 1, wherein the at least one non-invasive or minimally invasive stimulation or haemodynamic augmentation effector further comprises at least one of: an infrazygomatic electrode for transcutaneous or percutaneous stimulation of a sphenopalatine ganglion or related parasympathetic craniofacial structure via an infrazygomatic pathway, wherein the controller adjusts stimulation parameters based on a measured cerebral, temporal-artery, or carotid haemodynamic response; a mastoid-carotid electrode pair for transcutaneous or percutaneous modulation of a superior cervical ganglion or cervical autonomic pathway, optionally co-located with a temporal artery or carotid artery optical flow sensor; or a temporal interference electrode array for deep autonomic stimulation at a fastigial nucleus or brainstem target; and wherein the controller is configured to receive perfusion imaging information identifying at least one of anPT / 2026 / 12765ischaemic core, salvageable penumbra, or collateral pathway, and to select a sensing montage and stimulation montage targeting the identified region.

17. The system of claim 2, wherein the cerebral sensing module further comprises a near-infrared transillumination or backscattering subsystem configured to detect physiological oscillations in a transilluminated or backscattered optical signal associated with at least one of subarachnoid space dynamics, cerebrospinal-fluid motion, cerebral blood volume, tissue compliance, or intracranial compliance; and wherein the controller decomposes the transilluminated or backscattered optical signal into modal components characterised by natural frequency, damping ratio, or mode shape, and generates a modelbased estimate of intracranial compliance or intracranial-pressure-related variation from changes in the modal components.

18. The system of claim 17, wherein the cerebral sensing module is further configured to estimate an oedema- associated physiological parameter comprising at least one of a tissue-water-related optical change, cerebral blood-volume change, intracranial-compliance-related variation, cerebrospinal-fluid-associated oscillation, or low-frequency haemodynamic mode, and wherein the controller is configured to use the oedema-associated physiological parameter as an input to the model predictive control cost function, safety layer, or stimulation-control instruction.

19. The system of claim 4, wherein the controller is configured to estimate, from nearinfrared spectroscopy, speckle contrast optical spectroscopy, or near-infrared transillumination signals, a phase difference between a dominant haemodynamic oscillatory wave and at least one higher-order harmonic thereof, and to adjust at least one stimulation parameter to modify the phase difference towards a target range associated with penetrating-arteriole, pre-capillary, perivascular, paravascular, or glymphatic-flow-related dynamics; and wherein the peripheral or systemic sensing module is configured to derive an arterial pulse-wave feature comprising augmentation index, pulse-wave velocity, or arterial compliance metric, and the controller incorporates the arterial pulse-wave feature into u_global or uses it to estimate arterial compliance or peripheral vascular resistance within the two-level state-space model.

20. The system of claim 1, wherein the controller implements a dual-loop control architecture comprising a first control loop regulating at least one central stimulationPT / 2026 / 12765effector based on cerebral haemodynamic feedback and a second control loop regulating at least one peripheral haemodynamic augmentation effector based on peripheral or systemic feedback, coordinated by a supervisory controller to modify systemic-to-cerebral haemodynamic coupling; further comprising a safety layer configured to inhibit, limit, or suspend stimulation based on at least one of charge density, current amplitude, signal quality, artefact level, or Accepted Reproducible Mode confidence score, with automatic reversion to observation-only mode on constraint violation; and a digital twin module comprising a computational haemodynamic model configured to generate synthetic physiological datasets with known modal parameters for pre-specified algorithm validation and stimulation-policy stress-testing.

21. The system of claim 7, wherein the controller is further configured to coordinate at least one cerebral or perivascular stimulation effector configured to generate a stimulation pattern associated with vasodilatory signaling propagating from a capillary, pre-capillary, or penetrating-arteriole region toward an upstream penetrating arteriole or pial vessel segment, with a pial or penetrating-arteriole pressure wave generated by perivascular nerve stimulation, transcranial electrical stimulation, or transcranial magnetic stimulation; and further coordinate the pressure wave with intermittent pneumatic compression- or neuromuscular electrical stimulation-mediated venous-return augmentation based on timing derived from near-infrared spectroscopy, speckle contrast optical spectroscopy, transcranial Doppler ultrasound, or multimodal state- space outputs; wherein the controller detects or estimates spatially restricted low-oxygenation events from at least one of nearinfrared spectroscopy, speckle contrast optical spectroscopy, near-infrared transillumination, transcranial Doppler ultrasound, or multimodal state-space outputs, and adjusts the timing of the coordinated stimulation to modify downstream vascular resistance within predefined safety constraints.Method Claim22. A computer-implemented method of operating a neurovascular modulation system as claimed in claim 1, the method comprising:acquiring, using one or more processors, at least one cerebral haemodynamic signal and at least one peripheral or systemic physiological signal;PT / 2026 / 12765synchronizing the signals to a common time base with a verified inter-device timing error below a predetermined threshold;extracting at least one haemodynamic mode or neurovascular state variable from the synchronized signals using at least one of operational modal analysis (OMA), numerical algorithms for subspace state-space system identification (N4SID), dynamic mode decomposition with control (DMDc), sparse identification of nonlinear dynamics, stochastic subspace identification (SSI), or multiple-input multiple-output modal analysis;computing a composite brain-scalp score for each candidate haemodynamic mode and excluding scalp / systemic-classified modes from stimulation targeting;estimating a two-level state-space model and a cross-level coupling matrix L(t) from the synchronized signals; identifying at least one Accepted Reproducible Mode satisfying reproducibility, algorithm- agreement, and physiological-interpretability criteria; andoutputting a stimulation-control instruction to cause at least one non-invasive or minimally invasive stimulation or haemodynamic augmentation effector to modify at least one temporal, spatial, frequency, phase, or amplitude characteristic of cerebrovascular reactivity, collateral perfusion, neurovascular coupling, perivascular flow, paravascular flow, glymphatic-flow-related dynamics, or systemic-to-cerebral haemodynamic coupling, subject to one or more safety constraints.

23. The method of claim 22, further comprising: learning or updating, using sparse identification of nonlinear dynamics combined with reinforcement learning, at least one sparse dynamics model, sparse reward model, or sparse stimulation-control policy from synchronized multimodal physiological signals, structured perturbation data, or digitaltwin simulation data, the policy being limited to a predetermined maximum number of active dictionary terms and subject to physiological safety constraints; estimating a phase relationship between a dynamic cerebral autoregulation-dominant mode and a neurovascular-coupling-dominant mode and generating coordinated stimulation-control instructions for a global-mode effector and a local-mode effector to drive the phase relationship toward a target range; and, in a human-in-the-loop mode, presenting proposed stimulation parameters with predicted haemodynamic response trajectory and uncertainty estimate for clinician or operator approval before delivering the stimulation-controlPT / 2026 / 12765instruction.Computer-Readable Medium Claim24. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the processors to perform operations comprising:receiving multimodal physiological signals comprising at least one cerebral haemodynamic signal and at least one peripheral or systemic physiological signal, and synchronizing the signals to a common time base;extracting haemodynamic modal features or neurovascular state variables; computing a composite brain-scalp score for each candidate mode; evaluating reproducibility, algorithm- agreement, and physiological-interpretability criteria to identify at least one Accepted Reproducible Mode; and estimating a two-level state-space model and a crosslevel coupling matrix L(t), with automatic fallback to output-only modal analysis when L(t) is non-identifiable;learning or updating at least one of: a machine-learned state- space model that outputs uncertainty estimates for a latent macrovascular state, a latent microvascular state, or L(t); or a sparse-identification reinforcement-learning policy limited to a predetermined maximum number of active dictionary terms, trained on synchronized physiological data or digital-twin outputs and subject to physiological safety constraints;generating and ranking a plurality of candidate stimulation-control policies based on one or more of predicted haemodynamic response, uncertainty estimate, safety limits, observability criteria, or model-identifiability criteria for L(t); anddetermining and outputting a stimulation-control instruction targeting the at least one Accepted Reproducible Mode, subject to safety constraints, and reverting to observation-only mode when any constraint is violated or any gate criterion in a four-gate validation procedure is not met.