System and method for bidirectional brain-computer interface

Abstract

Disclosed herein are apparatuses and methods for bidirectional brain-computer interfaces that encompass both neural sensing and neuromodulation that can improve the performance of brain-based device control or enhance neuromodulation to assist an individual's brain-controlled action, rehabilitate an individual's functions, and enhance neuromodulation through closed-loop brain-device interfacing. The bidirectional BCI technology disclosed herein involves brain sensing / decoding and brain stimulation to provide closed-loop BCI control that facilitates the neural interfacing and improves the performance of BCIs.

Description

Attorney Docket: 8350.2025-115WOSYSTEM AND METHOD FOR BIDIRECTIONAL BRAIN-COMPUTER INTERFACERelated Applications

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 718,200, filed November 8, 2024, the contents of which are hereby incorporated herein in its entirety.Government Interest

[0002] This invention was made with support of the United States Government under contracts AT009263, EB029354, NS124564, and NS131069, awarded by the National Institutes of Health (NIH). The U.S. government has certain rights in the invention.Field of the Invention

[0003] The disclosed invention relates to systems and methods for bidirectional brain computer interfaces, which perform sensing of brain data, decoding of said data to decipher the brain's "intention", and delivering certain physical energy onto certain brain regions to modulate the neural activities thereof. More specifically, the invention relates to an apparatus, methods and a system of using both sensing / decoding of brain data of an individual and transmitting physical energy such as ultrasound energy to modulate neural circuits for better brain-device control.Attorney Docket: 8350.2025-115WOBackground

[0004] A brain computer interface (BCI) is a neurotechnology that enables direct brain-based communication between an individual and the rest of the world. BCIs are designed to sense neural activity as electrical signals and to translate these signals into actions. A BCI recognizes the intent of the user through electrophysiological or other signals of the brain. In real time, neural recordings are decoded and translated into output commands that accomplish a desired task of the user. BCIs have been shown to allow individuals with compromised motor function to interact with the world, both physically and socially, making it a promising route for therapeutic intervention.

[0005] BCI is an emerging technology that can help restore, replace or rehabilitate useful function to people severely disabled by a wide variety of devastating neuromuscular and / or neurological disorders. BCI aims to produce powerful new assistive interactive and communication technologies by providing a non-physiological channel to control devices through intent and will offer not only the possibility of restoring independence to individuals suffering from paralysis, but of augmenting healthy human behavior, synergizing human and machine for increased productivity and capacity.

[0006] BCI has also been used in neurorehabilitation to assist patients from recovery and enhancement of functions. While BCIs could have broad therapeuticAttorney Docket: 8350.2025-115WO applications in various patient cohorts, as well as recreational applications in able-bodied subjects, they have traditionally targeted users suffering from neuromuscular impairments such as amyotrophic lateral sclerosis (ALS), spinal cord injury (SCI), or stroke, but whose cognitive function remains intact. Indeed, clinical applications of BCIs often complement other therapeutic strategies. This technology has also recently gained recreational popularity as neural recordings become increasingly accessible due to low- cost systems and open-source toolboxes. As such, the concept of human enhancement and restoration through brain-controlled channels now allows individuals to connect with others on a level not previously achievable.

[0007] In its most basic form, a BCI is a system that decodes the user's mental state or intention and maps such information to the action of a device that interacts with the surrounding environment. These interactions can convey information that range from answering simple yes / no questions, to constructing word-based communication, to the navigation and control of devices such as robotic arms, wheelchairs, and more. The widespread research space of BCI has also facilitated new and creative applications for this technology.

[0008] While these examples represent exciting and innovative avenues for the field of BCI, the direction of BCI subfields is often guided by how easily and robustly mental intention can be decoded from brain recordings. As such, there are various methods available to acquire neural signals, including bothAttorney Docket: 8350.2025-115WO noninvasive and invasive approaches that exhibit different signal-to-noise ratios, spatial coverage, spatiotemporal resolution, etc.

[0009] This spectrum of signal acquisition techniques is accompanied by a broad collection of mental states and tasks that can be incorporated into the BCI system. For invasive BCIs, high-resolution recordings provide information regarding detailed brain states and dynamics (often related to motor function), however, these systems are often constrained to a limited number of clinical participants or to basic research in rodents and non-human primates.

[0010] Noninvasive BCIs, on the other hand, minimally impact the user without safety concerns, are easy to use in everyday life, and facilitate long-term performance tracking in a large number of participants. Nevertheless, current noninvasive BCIs exhibit limited performance due to the limited signal-to- noise ratio and information transfer rate that occurs when neural signals transmit from brain tissues, through the skull, and to the scalp. The easy access to many suitable participants, however, offers opportunities that enable noninvasive BCI researchers to push the limits of robust paradigms and decoding techniques to optimize task accuracy, information transfer, and system reliability for human applications.

[0011] While much effort in noninvasive BCI research is confined to selection like interfaces involving pre-set tasks or keyboards, utilizing evoked potentials and visual synchronous behavior for control, recent advancements haveAttorney Docket: 8350.2025-115WO shown promise in achieving control in line with human behavior. Several methods for measuring brain signals for BCIs exist, and one promising approach records sensorimotor rhythms (SMRs) with scalp electrodes (e.g., electroencephalography, or EEG). The signals used in the technique are predictable changes mostly in the alpha (8-13 Hz) and beta (14-30 Hz) frequency bands known as event related desynchronization / synchronization (ERD / ERS), which are produced during both real and imagined movement. In the SMR-based approach, subjects imagine producing and sensing movements of various types, and the resulting changes in synchronization of the EEG signal is classified and mapped to BCI control. Fine control of multidimensional BCIs is achievable with SMR-based EEG. An advantage of the SMR approach is that it allows continuous readout of brain signals that can be used for continuous closed-loop control of a BCI.

[0012] Noninvasive BCI research and development have been conducted on both healthy subjects and individuals with impairments. Studies have demonstrated that brain-decoding functionality is preserved across impaired victims. However, even recent innovations in noninvasive BCIs have limited online accuracy rates, which leave room for improvement before seeking widespread clinical use.

[0013] Noninvasive neuromodulation has been introduced to modulate brain activity treating neurological and mental disorders that do not have risks associated with brain implants and high costs. Traditionally, noninvasiveAttorney Docket: 8350.2025-115WO modalities have included transcranial magnetic stimulation (TMS), transcranial direct current stimulation (tDCS), and transcranial alternating current stimulation (tACS). TMS uses pulsed magnetic stimulation to induce electric fields within the brain. tDCS and tACS apply electrodes on the scalp and apply direct or alternating current, respectively, to induce sub-threshold modulation of neural tissues. Transcranial electric stimulation (tES) is an umbrella term that encompasses both types of current.

[0014] Another technique, known as low intensity transcranial focused ultrasound (tFUS) is a safe, noninvasive neuromodulation technique with high spatial resolution and focality and having the capability of accessing the deep brain. During tFUS neuromodulation, pulsed ultrasound energy is transmitted though the skull with high spatial selectivity, which can be steered and utilized to elicit activation or inhibition through parameter tuning. Multipleelement tFUS arrays have the ability to electrically target and steer the focal zone in real time and the ability to simultaneously target multiple brain regions.Summary of the Invention

[0015] A potential way to improve BCI is by pairing them with neuromodulation devices. Disclosed herein are apparatuses and methods for bidirectional brain computer interfaces that encompass both neural sensing andAttorney Docket: 8350.2025-115WO neuromodulation that can improve the performance of brain-based device control or enhance neuromodulation to assist an individual's brain-controlled action, rehabilitate an individual's functions, and enhance neuromodulation through closed-loop brain-device interfacing. The bidirectional BCI technology disclosed herein involves brain sensing / decoding and brain stimulation to provide closed-loop BCI control that facilitates the neural interfacing and improves the performance of BCIs.

[0016] The present invention provides a bidirectional BCI encompassing both neural sensing and decoding, as well as neuromodulation based neural encoding, to realize a closed-loop BCI. Brain signals are sensed using electrophysiological, hemodynamic, optical, and / or acoustic means, and are used to decode brain intention and status. Neuromodulation using acoustic, electrical, magnetic and optical means are used to encode information into the nervous system, modulate neural circuits activity and connectivity. Collectively, the bidirectional BCI will provide closed-loop brain-device interfacing which enables the controlling of external devices based on brain intentions decoded from recorded neural signals, enhancing neural signals associated with brain intentions and status using neuromodulation, enhancing brain-based device control as facilitated by neuromodulation, and enhancing neuromodulation through closed-loop brain-device interfacing.

[0017] The present invention further provides noninvasive bidirectional BCI encompassing noninvasive neural sensing and decoding, as well asAttorney Docket: 8350.2025-115WO noninvasive neuromodulation based neural encoding, to realize closed-loop noninvasive BCI.

[0018] The present invention further includes methods of noninvasive neural sensing using electroencephalography (EEG), decoding of brain intentions and status using various machine learning and signal processing, using noninvasive neuromodulation tools (e.g., tFUS used to modulate the central nervous system with certain spatial specificity on specific brain circuits, for effective BCI control of devices).

[0019] The present invention further includes methods to boost neural signals for brain decoding by means of neuromodulation including electric, magnetic and acoustic stimulation, at a single or multiple brain targets for BCI control.

[0020] The present invention further includes methods to induce state-specific BCI feedback by means of inducing perceptions with neuromodulation including focused ultrasound stimulation to one or more brain regions.

[0021] The present invention further includes methods to determine the specific ultrasonic dosage and delivery of ultrasonic energy onto the nerve targets for boosting performance of brain decoding using neural signals. The precise targeting and dosage control are enabled with, for example, tFUS neuromodulation, which also minimizes undesired stimulation effects to enhance the safety and reliability of neural signal-based systems. Specifically, the present invention provides a method for enhancing spatial targeting of tFUS in bidirectional brain-computer interface systems. By applying a phase-Attorney Docket: 8350.2025-115WO aberration correction technique using subject-specific head models and k- space simulations, the system improves the precision of ultrasound delivery to wanted brain targets. This approach increases focal accuracy and energy deposition, enabling more effective neuromodulation and improving the performance of tFUS-modulated BCI decoding.

[0022] The present invention further includes methods for determining and optimizing the specific electromagnetic dosage and targeted delivery of electromagnetic energy to selected nerve targets. This approach enhances the effectiveness and precision of brain decoding by amplifying neural signals and / or boosting their interpretability. By refining the electromagnetic energy application based on the physiological characteristics of the nerve targets, the invention improves brain decoding performance in various applications, including neuroprosthetic control, cognitive augmentation, and therapeutic neuromodulation.

[0023] Lastly, the present invention further includes closed-loop BCI devices integrating neural sensing, brain decoding, device control, and neuromodulation for neuroscience research, and clinical applications to provide assistive neurorobotics or treatment helping patients with impairments and neurological disorders, and non-disabled individuals to enhance their performance.Attorney Docket: 8350.2025-115WOBrief Description of the Drawings

[0024] By way of example, specific exemplary embodiments of the disclosed system and method will now be described, with reference to the accompanying drawings, in which:

[0025] FIG. 1 is a schematic diagram showing a bidirectional brain-computer interface system.

[0026] FIG. 2 is a schematic diagram showing two methods used by a BCI to interpret raw EEG signals.

[0027] FIG. 3 depicts processes for excitatory neuromodulation and inhibitory neural modulation.

[0028] FIG. 4 is a schematic diagram depicting three techniques for noninvasive neuromodulation.

[0029] FIG. 5 is a schematic diagram illustrating one embodiment of a bidirectional BCI, using neural modulation to brain area V5 to enhance motion-onset visual evoked potentials (mVEPs) in an N200 BCI speller.

[0030] FIG. 6 is an illustration showing the functional activity evoked by the mVEP BCI speller is located in the geometric center of V5.

[0031] FIG. 7 is an illustration showing an example of bidirectional BCI integrating mVEP BCI speller with transcranial focused ultrasound (tFUS) to V5.

[0032] FIG. 8 is an illustration showing that on-target tFUS (tFUS-GC) N200 modulation is conveyed through the dorsal pathway in theta and alpha frequencies in the mVEP speller bidirectional BCI.Attorney Docket: 8350.2025-115WO

[0033] FIG. 9 is a schematic view of an embodiment of sensorimotor rhythm bidirectional BCI.

[0034] FIG. 10 shows example results of human study in a group of 15 healthy human subjects without prior BCI training, where tFUS stimulation targeted at the primary motor cortex.

[0035] FIG. 11 is a schematic view of using neural modulation to target to the parietal lobe to enhance P300 cognitive-based BCI speller signals.

[0036] FIG. 12 is a schematic view of using neural modulation to target brain area VI to enhance steady-state evoked potential responses in the associated BCI speller.

[0037] FIG. 13 is a schematic demonstrating an embodiment wherein ultrasound- induced haptic-feedback can be used in conjunction with a noninvasive BCI. The user initiates BCI movement to grasp a glass of water by imagining their hand closing.

[0038] FIG. 14 is a schematic demonstrating an embodiment wherein a bidirectional visual BCI can provide assistance to the blind by having the user think about an object they want to find, such as a glass of water.

[0039] FIG. 15 is a schematic demonstrating how tFUS targeting can be further improved by considering phase-aberration correction to enhance the bidirectional BCI.Attorney Docket: 8350.2025-115WODetailed Description

[0040] The present invention can be applied to neural interfacing using various modes of neural sensing and neuromodulation. Without intending to limit the invention thereto, the invention will be explained primarily in the context of its application to noninvasive bidirectional BCIs where both neural sensing and modulation can be implemented using noninvasive techniques. Examples are provided using electroencephalography (EEG) for neural sensing, and transcranial focused ultrasound stimulation (tFUS) for neuromodulation.

[0041] FIG. 1 shows a bidirectional brain-computer interface system encompassing neural sensing, neuromodulation using physical energy, brain decoding from recorded neural signals, and device control using such decoded brain intentions. Brain-to-Machine Interface 101 includes components of a monodirectional brain-to-machine interface, where the user's brain signals are decoded and interpreted to carry out a task. One exemplary task for a BCI is to control a robotic arm 102. The most recent brain-to-machine prediction is translated into moving the robotic arm 102a. Machine-to-Brain Interface103 includes the components of the monodirectional interface through neuromodulation, where signals are encoded into devices that then are applied to the brain in the form of neurofeedback or neuromodulation. User104 has their brain signals recorded via EEG electrodes 104a and neuro modulation is administered back into their brain 104b. Combining both the brain-to-machine and machine-to-brain interfaces creates a closed-loop BCI.Attorney Docket: 8350.2025-115WO

[0042] FIG. 2 shows two methods used by a BCI to interpret brain signals. Generally, the BCI pipeline stars with signal acquisition at 202, wherein raw electrical signals (e.g., EEG signals) are collected from the brain. Features are then extracted from the raw signals 204 and are used to make a prediction 206 of the user's intent. The predicted user intent is then translated to the intended action, for example, moving a robotic arm 208.

[0043] The top view in FIG.2 depicts a first embodiment of a BCI as a sensorimotor BCI linear classifier. This embodiment involves obtaining the raw EEG signal, bandpass filtering for the mu (or alpha) rhythm, and extracting and normalizing channels C3 and C4 with a rolling z-score. For an up-down task, if the sum of the normalized C3 and C4 are greater than 0, that indicates a predicted intent of et subject to move "up". The BCI system would then control, for example, a robotic arm (or another device, such as a computer cursor) to move up.

[0044] The bottom view of FIG. 2 shows a more computationally complex classifier, using a deep learning model, which still follow the same general approach as the BCI depicted in the top view. The raw EEG signal is captured and fed into the deep learning model. The model likely consists of some combination of temporal convolutions, spatial convolutions, fully connected layers, and / or recurrent neural network layers as it condenses the data down to the final output layer. This output layer is converted to a probability through applyingAttorney Docket: 8350.2025-115WO a normalization (e.g., Softmax), and the classifier chooses the class with the highest probability. This classification is then translated into a device output.

[0045] The present invention is directed improve the accuracy of brain computer interfaces, especially noninvasive BCIs, through bidirectional interfacing. Complex brain activity, and therefore the control of any bodily function, is the result of coordinated neuron firing. Neurons themselves do not pass complex signals amongst themselves. Instead, when they are depolarized past a threshold, they fire an effectively binary signal known as an action potential, shown in FIG. 3 at 305, 308, 310). This action potential has nearly identical characteristics across all human neurons (approximately 100 mV in amplitude and 1 ms in temporal resolution). As a result, the complexity of information is carried through the rate of the neuron firing and the circuit in of which the neuron is a part 304. Neural circuits for visual processing, motor imagination, motor planning, and motor execution have canonical processing locations spatially distributed across the human brain. As a result, electrical signals from these brain areas carry relevant information to their respective tasks and can be decoded to determine a person's intentions. BCIs do just that.

[0046] Noninvasive BCIs use scalp EEG electrodes to collect the electrical signals. Instead of being able to record individual neuron action potentials, scalp EEG electrodes capture the spatial sum of the post synaptic potentials from several neurons. Because neurons are spatially distributed on the brainAttorney Docket: 8350.2025-115WO according to their function, these widespread electrical activities provide high level information about the user's intent. However, due to their placement on the opposite side of multiple physiological barriers (primarily the skull) from the neural sources, the signal captured by the EEG is quite noisy, and at times the control signal can be hard to differentiate from that noise. The primary purpose of the present invention is to boost the relevant control signal.

[0047] Because the EEG signal is the spatial sum of post synaptic currents from a plurality of neurons, if more neurons fire more often, then the signal amplitude will increase. Excitatory neuromodulation can be used to promote more frequent neural firing. Excitatory neuromodulation may do this by slightly depolarizing the neurons, such that they are closer to their action potential threshold and will be more susceptive to firing when further excited by another neuron, as depicted at 306 in FIG. 3. Inhibitory neuromodulation works the opposite way; it hyperpolarizes the neurons such that they are further away from their action potential threshold, and even receiving signals that would normally trigger action potentials may not be enough, as depicted at 309 in FIG. 3.

[0048] In addition to being applied concurrently with the task signal, neuromodulation may be applied before or after a task. Whereas concurrent neuromodulation may alter the impact of the signal as it arrives at neuron, these other applications may alter how sensitive a neuron is to such signals.Attorney Docket: 8350.2025-115WOEach neuron receives inputs from thousands of other neurons, and the strength between the connections is constantly in flux. Active connections are strengthened in a positive feedback loop known as long term potentiation / depotentiation. Inactive connections are weakened through long term depression. This process of strengthening active channels and weakening inactive channels, is the basis of learning. Neuromodulation may be used to excite and further activate relevant neural circuits to strengthen them, or to inhibit and weaken extraneous ones. In both cases, strengthening the relevant signal or weakening the irrelevant signal, the desired BCI control signal may become more emphasized and capturable by the EEG.

[0049] Beyond using neuromodulation for increasing targeted neuron excitability, tFUS can also be used to activate neurons and create relevant perceptions to targeted brain areas. In other words, tFUS targeted to the hand knob of the somatosensory brain region can lead to tactile sensations on the contralateral hand and to the visual cortex can lead to visual perceptions in the person's field of view. This type of neuromodulation can be combined with BCI to directly encode relevant information back into the brain.

[0050] FIG. 4 shows three techniques for noninvasive neuromodulation. Transcranial focused ultrasound (tFUS) is shown at 401, transcranial magnetic stimulation (TMS) is shown at 406, and transcranial electric stimulation (tES) is shown at 409.Attorney Docket: 8350.2025-115WO

[0051] TMS coils 407 generate a magnetic field that passes through the skin and bone unattenuated and creates an electric field within the brain tissue 408. tES applies direct or alternating currents 411 between scalp electrodes 410, some of which flow through brain. tFUS transducers 402 sonicate pressure waves 405 through the skin and bone 403 into the brain 404.

[0052] Using a magnetic resonance imaging (MRI) scan of a subject's brain, the subject-specific brain structural map can be discerned. Used in conjunction with common optical-based technology, the subject's 3-dimensional space can be mapped to their brain locations and used to align neuromodulation devices to target specific areas.

[0053] Because different neural circuits are spatially distributed in different cortical areas, the target for neuromodulation should be different for each task. EEG communication restoring BCI typically use the speller paradigm, where visual stimuli are presented across a virtual keyboard. Based on where, and when, a user looks at the screen, their brain responds to the different stimuli. Their brain response can then be used to determine where on the screen they were looking, and to type the intended letter. In the case of a visual motiononset based speller, the cortical area responsible for generating the control signal is area V5. As such, that area should be targeted with neuromodulation.

[0054] Example of bidirectional BCI - tFUS modulated mVEP BCI SpellerAttorney Docket: 8350.2025-115WO

[0055] FIG. 5 illustrates one embodiment of a bidirectional BCI, using neural modulation to brain area V5 to enhance motion-onset visual evoked potentials (mVEPs) in an N200 BCI speller. An example of a visual-motion stimulus 501 is a line moving across the letter the user wishes to type. Neuromodulation 502 is targeted to brain area V5, 503, which is responsible for visual-motion processing. Due to brain circuit connectivity, the modulation may be propagated along the relevant visual processing pathway(s) 503a. The modulated brain activities result in a stronger EEG signal 504. This improved signal is fed into a machine learning classifier, which then predicts that the user was intending to type 's'.

[0056] An individual's head size can be measured and fit with an EEG cap consisting of multiple electrodes. Incisions can be made to the EEG cap to expose the V5 area, which allows for direct interfacing between the scalp and the ultrasound transducer 502. Electrode placement relative to subject landmarks are captured for source imaging analysis. Raw EEG is acquired using an EEG data acquisition system. The electrode impedances are kept below 10 kQ using conductive electrolyte gel and applied using cotton swabs.

[0057] Subjects can be presented on screen with a six-by-six virtual keyboard. They are instructed to stare at the specific key they wish to type as lines flashed to the right across each row and column of the keyboard 501. Online EEG data are processed and fed to the classifier once per scan epoch. For each epoch, the classifier predicts the maximally probable row and column index of theAttorney Docket: 8350.2025-115WO subject's gaze 505. If there are multiple epoch scans per letter, the classifier would average the probabilities over the epochs.

[0058] This exemplary embodiment is designed as a cross-over study, with the same subjects tested in four conditions in a predominantly randomized order: "non-modulated" (inactive tFUS), "decoupled-sham" tFUS (decoupled, but active, tFUS to account for audio-induced confounds), "tFUS-GP" (active tFUS targeted to a control brain location located by more than 1-cm away), and V5-targeted "tFUS-GC" (estimated peak-to-peak pressure: 0.2 MPa, approximate focal beam diameter: 2 mm, approximate focal beam length: 2 mm, pulse repetition frequency: 3 kHz, pulse duration: 200 ps, sonication duration: 500 ms). EEG source imaging of the non-modulated condition confirmed alignment tFUS-GC's with the reconstructed source at the center of V5, as shown in FIG. 6.

[0059] Performance of the BCI is evaluated for each condition's online testing session based on the Euclidean distance error (EE) from the subject's intended letter to the BCI classifier's output. The output for each letter (i.e., C, A, etc.) is considered as one separate trial. For each subject, outlier trials are identified using the interquartile range test and discarded. Afterwards, all the trials for all the subjects are pooled together. The Euclidean distance is converted to a percent error by normalizing it against the maximum possible distance (diagonal distance of the grid: 6V2). To account for repeated measures across subjects and the potential effects of learning and / or fatigueAttorney Docket: 8350.2025-115WO resulting from the number of repeated BCI scans per epoch and the order in which the condition is tested, the Euclidean Errors are fitted with a linear mixed effect model (See Eq. (1-2)). The model's fit for Euclidean errors are compared across conditions using a type III Analysis of Variance (ANOVA) test and a one-tailed z-test. Bonferroni p-value adjustment is applied to reduce likelihood of false positive errors. Euclidean errors are considered significantly different if padjusted< a = 0.05.EE~ Condition + Scans + Order + Order-. Scans + l\Subject)(1)

[0060] Mathematically, this is equivalent to:

[0061] Where i corresponds to the itfltrial of subject k, and j denotes the experimental condition of trial i. ? is a (1 x J) vector of the weights for each condition, and Condition is a (J x 1) one-hot vector. P0, / 32, p3, (34are scalar quantities. Scansi and Orderj are discrete integer values corresponding to the number of repeat scans of trial i and the order in which condition j was tested, respectively. bk iis the random-effect of the subject k for trial i, defined as a Gaussian distribution with a subject-specific mean and variance. ei kis the error term.Attorney Docket: 8350.2025-115WO

[0062] Source imaged time series are transformed into time-frequency power responses using Morlet waveforms. The N200 theta, alpha, beta, and low gamma powers are extracted by taking the [4, 8), [8, 12), [12, 30), or [30, 40) Hz frequency power response, respectively, in the 100 to 250 ms poststimulus window. Outlier trials are rejected by the interquartile range test. The data are exported to R and fit with a linear mixed effect model to account for repeated measures across subjects (Equation 3). The effects of conditions are compared with a one-tailed z-test with false discovery rate correction for multiple comparisons. Effects are considered significant if the adjusted p-value was below 0.05.N200 Power —Condition + l\Subject)(3)

[0063] This process is repeated with the Desikan-Killiany (aparc) atlas' parcellations for the left hemisphere superior parietal lobe (superiorparietal-lh) and the left hemisphere inferior temporal lobe (inferiortemporal-lh) to examine possible downstream effects in visual processing pathways.

[0064] tFUS to the geometric center of V5 significantly reduces mVEP BCI speller errors. The errors were analyzed using a linear mixed effect model (Eq. (1)), and a Type III Analysis of Variance (ANOVA) indicated significant differences in means across conditions (p < 0.001). FIG. 7 shows experimental testing in a group of 25 healthy human subjects. Further condition-to-condition comparisons using z-tests and Bonferroni multiple comparisons p-valueAttorney Docket: 8350.2025-115WO adjustment (FIG. 7a) revealed that the Euclidean error for the tFUS-GC condition (N = 25 subjects / 356 trials; mean error = 13.3 ± 18.4%) was significantly lower than those of the non-modulated (N = 25 subjects I 351 trials; mean error = 15.5 ± 18.7%; Padjusted < 0.01), decoupled-sham (N = 19 subjects I 268 trials; mean error = 16.9 ± 20.8%; padjusted < 0.05), and tFUS-GP conditions (N = 16 subjects I 214 trials; mean error = 17.0 ± 18.2%; Padjusted < 0.001). No significant difference was detected between the decoupled-sham, non-modulated condition, and tFUS-GP conditions (Padjusted > 0.05). The effect size can be quantified with Cohen's d value. The effect of tFUS-GC compared to all other controls is greater than 0.5 (FIG. 7b), indicating a moderate effect. The effect of experimental condition was further quantified using Bayes Factor (BF) analysis (FIG. 7c). For the recommended default hyperparameter settings, the analysis found experimental condition to have a strong effect (median BF: 14.0; 95% confidence interval: 13.8 to 14.4) on Euclidean error. Rather than just present one BF corresponding to the recommended default hyperparameter, we present calculations of BF through a range of hyperparameter scaling up to "ultrawide" (1.0). These results provide a robust view that experimental condition has at least a moderate effect (BF > 3), and often strong effect (BF > 10) throughout nearly the entire range of hyperparameters scaling up to "ultrawide", thus providing a higher degree of confidence in the results than one singular value.Attorney Docket: 8350.2025-115WO

[0065] tFUS to the center of V5 significantly amplifies the relevant N200 powers. To enhance the analysis, EEG data were further processed through electrophysiological source imaging (ESI). The data were z-scored with respect to the 200 ms pre-stimulus baseline. N200 powers were computed by applying time-frequency Morlet waveform power transformation to the 100 to 250 ms post-stimulus window for theta (4 - 8 Hz), alpha (8 - 12 Hz), beta (12 - 30 Hz), and low gamma (30 - 40 Hz) frequencies, as shown in FIG. 8. Linear mixed effect models were fitted to N200 powers as a function of experimental condition (Eq. (3)). Comparisons between conditions revealed tFUS-GC (z-scores: theta power = 7.29 ± 5.97, alpha power = 5.59 ± 4.10, beta power = 3.82 ± 2.43, gamma power = 3.48 ± 2.20) to V5 during the mVEP BCI significantly amplified the N200 theta and alpha responses in the V5 region compared to the non-modulated (z-scores: theta power = 5.74 ± 4.36, alpha power = 4.75 ± 3.23, beta power = 3.51 ± 2.15, gamma power = 3.20 ± 1.92;Padjusted < 0.0001 for theta band, Padjusted < 0.0001 for alpha band), decoupled-sham (z-scores: theta power = 5.11 ± 3.74, alpha power = 4.38 ± 2.91, beta power = 3.32 ± 1.96, gamma power = 3.06 ± 1.79; Padjusted < 0.0001 for theta band, Padjusted < 0.0001 for alpha band), and tFUS-GP (z- scores: theta power = 6.83 ± 5.58, alpha power = 5.25 ± 3.79, beta power = 3.70 ± 2.38, gamma power = 3.35 ± 2.13; Padjusted < 0 05 for theta band, Padjusted < 0.05 for alpha band) conditions. Additionally, tFUS-GC beta and gamma N200 were significantly amplified over the non-modulated (padjustedAttorney Docket: 8350.2025-115WO< 0.001 for beta band, padjusted < 0.001 for gamma band) and decoupled- sham (padjusted< 0.0001 for beta band, Padjusted < 0.0001 for gamma band) conditions, but not over tFUS-GP.

[0066] tFUS-GP was significantly amplified in comparison to the non-modulated (Padjusted < 0.0001 for theta band, Padjusted < 0.001 for alpha band, Padjusted < 0.05 for beta band, Padjusted < 0-05forgamma band) and decoupled-sham (padjusted < 0.0001 for theta band, Padjusted < 0.0001 for alpha band, Padjusted < 0.001 for beta band, Padjusted < 0.01 for gamma band) conditions across all four frequency ranges. Additionally, the decoupled-sham condition was found to be significantly damped (Padjusted < 0.001 for theta band, padjusted< 0.001 for alpha band, Padjusted < 0.05 for beta band, Padjusted < 0.05 for gamma band) compared to the nonmodulated condition for all four frequency ranges.

[0067] Visual evoked potentials are dominated by low frequency (alpha and below) components, providing evidence that on-target tFUS significantly amplifies the control signal in a bidirectional BCL

[0068] In certain embodiments, the bidirectional BCI system described herein is designed to interface with distinct cortical pathways involved in visual processing, namely, the dorsal and ventral visual streams. These anatomical and functional pathways enable the system to achieve targetedAttorney Docket: 8350.2025-115WO neuromodulation and improved decoding of brain activity in support of enhanced BCI performance.

[0069] The dorsal visual pathway, extending from the primary visual cortex to the posterior parietal cortex and including intermediate areas such as the middle temporal visual area (V5 or MT), is primarily responsible for processing motion, spatial relationships, and visually guided motor actions. In embodiments where the system utilizes motion-onset visual evoked potentials as the primary input signal, the dorsal stream is of particular significance. The tFUS module may be configured to deliver neuromodulatory stimulation to V5, thereby modulating the neural substrates of motion perception and enhancing the signal-to-noise characteristics of the mVEP response. The improved signal fidelity supports more accurate and rapid classification of user intent in the BCI decoding pipeline.

[0070] The ventral visual pathway, in contrast, projects from the primary visual cortex toward the inferior temporal cortex and is associated with object recognition and semantic processing. In certain configurations, the system may incorporate stimuli or tasks that engage the ventral stream, such as symbol or image recognition tasks, in which case neural responses originating from ventral stream regions may also be monitored and decoded. While the primary neuromodulatory focus may reside within the dorsal pathway, outputs from ventral regions may complement BCI operation by encodingAttorney Docket: 8350.2025-115WO higher-order perceptual content or cognitive features relevant to user communication.

[0071] Thus, the BCI system may be configured to exploit the functional specialization of the dorsal and ventral streams to enable enhanced bidirectional communication. The tFUS stimulation may be applied selectively to dorsal pathway structures (e.g., V5) to influence motion-related visual processing, while EEG or other neural recordings may be used to decode task-relevant activity from either or both visual streams depending on the application context.

[0072] tFUS targeting can be further optimized through phase-aberration correction for bidirectional BCI.

[0073] In certain embodiments, the disclosed bidirectional BCI system leverages tFUS stimulation module to deliver ultrasonic energy to a target cortical region, such as the middle temporal visual area (V5L). To enhance the spatial precision and energy delivery efficiency of tFUS stimulation, the system may incorporate a subject-specific aberration correction framework. This framework is configured to compensate for skull-induced phase distortions that arise from the heterogeneous acoustic properties and irregular geometry of the human skull.

[0074] The aberration correction method may utilize individualized head models derived from Tl-weighted magnetic resonance imaging (MRI) and pseudocomputed tomography (pCT) data to define both skull morphology and theAttorney Docket: 8350.2025-115WO targeted cortical region. In some implementations, a 128-element random phased-array ultrasound transducer is employed, and ultrasound field simulations are performed using a k-space pseudospectral method. Each transducer element is independently activated in both free-field and skullpresent conditions, and phase distortions are determined by comparing the received waveforms at the focal point. A set of corrected continuous waveforms is then computed by incorporating phase delays estimated through time-reversal of peak displacement data, such that the reconstructed acoustic focus is realigned with the desired target.

[0075] Empirical validation of the correction framework demonstrates substantial enhancement in spatial targeting specificity. Specifically, phase correction leads to an approximately twofold increase in the spatial overlap volume between the reconstructed ultrasound focus and the anatomical V5L target, with the overlap volume increasing substantially relative to the uncorrected condition. In addition, axial focal positioning errors are reduced substantially and the total ultrasound energy delivered to the target region is improved. These performance gains are achieved without materially increasing the lateral focal spread or degrading the axial spatial resolution of the acoustic beam.

[0076] FIG. 15 illustrates the ultrasound transducer 1501 and a piece of skull over the left occipital lobe (to increase the spatial resolution at the focal area) that covers the V5L region. The skull-induced aberrations can be effectivelyAttorney Docket: 8350.2025-115WO corrected, enhancing the precision and efficacy of the tFUS focusing procedure, which can further enhance the bidirectional BCI.

[0077] By improving the precision and efficiency of energy deposition in the target area, this aberration correction method enables more effective excitation or inhibition of neural populations associated with sensory processing and perceptual integration. In the specific context of a bidirectional BCI architecture utilizing tFUS-modulated mVEPs for brain-computer communication, spatial targeting of the V5L region is functionally significant. The enhanced alignment of acoustic energy with the neuroanatomical substrate of interest facilitates consistent and robust modulation of neural signals, thereby improving closed-loop performance, signal fidelity, and user control accuracy within the BCI system. This integration of optimized spatial targeting further enables the system to adaptively modulate cortical excitability in a subject-specific and functionally meaningful manner, contributing to the overall efficacy of the bidirectional BCI framework.

[0078] Example of bidirectional sensorimotor rhythm BCI with tFUS modulation of motor cortex.

[0079] Another embodiment of the invention is to enhance the sensorimotor rhythm BCI by modulating motor related rhythms using tFUS stimulation. Movement restoring BCIs may utilize changes in the motor, somatosensory, or premotor cortices to decode a subject's intended movement and control a computer cursor, a wheelchair, or a robotic arm. Neuromodulation may beAttorney Docket: 8350.2025-115WO targeted to one location, as shown in FIG. 9, or a plurality of locations to enhance the user's motor execution, motor imagination, and / or motor planning signals.

[0080] FIG. 13 illustrates one embodiment of neuromodulation-integrated sensorimotor rhythm BCI for controlling a virtual or physical device. The sensorimotor rhythm BCI is used to modulate neurophysiological rhythmic activities recorded over the sensorimotor cortex by actual movement, motor intention, or motor imagery. The decoded sensorimotor rhythm brain control signals can be used to move a computer cursor, operate a virtual object, or control a robotic device, where neuromodulation is employed to enhance the brain control signals, increase subject attention, or improve signal-to-noise ratio.

[0081] In this paradigm, users imagine 901 themselves moving, which creates a change in signal in the sensorimotor rhythm in the sensorimotor cortex. This change can be detected to control a robotic arm 902 accordingly 902a, in a manner similar to imagined movement 901. This BCI can be modified into a closed loop bidirectional BCI by applying neuromodulation 903 to the sensorimotor cortex 904. This may increase the strength of the control signal 905, resulting in a more stable and accurate control of the device.

[0082] The modulation manifests as decreases in the alpha (8-13 Hz, also known as mu rhythm) and beta (14-26 Hz) frequency bands accompanied by increase in the gamma frequency band (>30 Hz), which is repeatedly observed in EEG,Attorney Docket: 8350.2025-115WO electrocorticography (ECoG), local field potentials (LFPs) as well as electromagnetic recordings (magnetoencephalography (MEG). Delta (1-4 Hz) frequency band signals have also observed to well accompany limb movement. Such rhythmic brain activities measured by EEG or MEG over the sensorimotor cortex are collectively referred to as the sensorimotor rhythms (SMR). Motor intention or motor imagery, or actual movement, can be decoded from the sensorimotor rhythms, which forms the basis of neural control in SMR-based BCIs. Studies have demonstrated that people can learn to increase and decrease the amplitude of sensorimotor rhythm using motor execution or mental strategy of motor imagery, and thereby control physical or virtual devices.

[0083] SMR-based BCIs have been widely investigated in healthy human subjects, as well as in people with amyotrophic lateral sclerosis (ALS) and in those with severe central nervous system damage from spinal cord injuries and stroke resulting in substantial deficits in communication and motor function. By modulating their SMR signals, users are able to acquire 2D or 3D movement control with degrees of freedom and performance comparable to studies using intracortical single unit recordings.

[0084] EEG signals exhibit endogenous oscillation that is widespread across the entire brain and have been found to be related to important aspects of motor function, sensory perception, or cognition. Task-related modulation in sensorimotor rhythms is usually manifested as amplitude (or power)Attorney Docket: 8350.2025-115WO decrease in the low-frequency components (alpha / beta band) (also known as event-related desynchronization (ERD). In contrast, an amplitude increase in a frequency band is known as event-related synchronization (ERS).

[0085] Planning and execution of movement has been found to lead to predictable decreases in the alpha and beta frequency bands. Also, studies have demonstrated that motor imagery can cause ERD (and often ERS) in primary sensorimotor areas. Discriminant information can be extracted from the spatial patterns of sensorimotor rhythmic modulations. More importantly, the modulation of alpha- and beta-band SMR have been found to be organized in a somatotopic manner. Source imaging studies of SMR have revealed that movement or motor imagery of different body parts were associated with decrease in SMR from regions along the primary sensorimotor cortex corresponding to different body parts. Such characteristic changes in EEG sensorimotor rhythms can be used to classify brain states relating to the planning / imagining of different types of limb movement, which forms the basis of neural control in SMR-based BCIs.

[0086] In addition to modulation in the alpha / beta frequency bands, SMR signals in the low frequency band (~1 or <1 Hz) have also been explored for studying the kinematic information. Researchers have been able to decode 2D and 3D velocity of hand movement from the very low frequency sensorimotor rhythms. Online BCI systems based on such slow SMR have been shown toAttorney Docket: 8350.2025-115WO allow users to acquire 2D movement control with a relatively short training time.

[0087] Neuromodulation at sensorimotor cortex through acoustic, electrical or magnetic energy would enhance / reduce sensorimotor rhythm oscillations, leading to enhanced performance of BCI tasks. FIG. 10 shows example data from 15 human subjects to control a computer cursor for left and right movement using the motor imagination paradigm with tFUS neuromodulation targeted at motor cortex. tFUS (2-3 kHz pulse repetition frequency (PRF) was applied to both the left and right hand-region of a subject's primary motor cortex every 1.5 seconds over the course of each trial. Experimental conditions were performed in a random order. The trial timed out after 6 seconds of a subject failing to move the cursor to a target, either in the correct direction or incorrect direction. The accuracy was quantified by the percent valid correct (PVC) metric, which considers the total correct cursor hits over the total of non-aborted trials. Due to the binary nature of the outcome, a 50% PVC corresponds to random chance. In this pilot study with healthy, primarily naive human subjects, their baseline (no active tFUS) performance was not significantly different from random chance. tFUS conditions, however, resulted in significant improvement over random chance, as well as significant improvement over the baseline condition (two-tailed 2-sample t-tests; *p < 0.05) - indicating using tFUSAttorney Docket: 8350.2025-115WO neuromodulation of pulse repetition frequencies of 3000 - 2000 Hz can have a clear improvement in performance.

[0088] FIG. 11 shows another embodiment of the invention, where neuromodulation is applied onto parietal cortex to enhance the P300 BCI speller. The P300 is an endogenous event-related potential (ERP) within the EEG and is detected in electrodes covering the parietal cortex. P300 signals occur in the context of an oddball paradigm and rely on a user's implicit ability to distinguish a rarely presented target stimulus from other more common non-target stimuli. This structure makes such an event-related response useful for spelling applications where a specific letter must be chosen from a larger set of irrelevant letters. P300-based BCIs usually use the visual row / column paradigm, in which a matrix (e.g., 6 x 6 cells or variable) containing the alphabet, numbers, and other items is presented to the user for selection. A P300 response is then elicited when the rows and columns of the matrix, flashing in random order, converge on the desired item being attended to. In this embodiment, P300 ERPs are recorded using an array of electrodes, and brain sources localized from the scalp recorded ERPs. The tFUS stimulation is targeted at the parietal cortex and other brain targets based on source imaging results, to facilitate information processing during the odd-ball paradigm.

[0089] The P300 speller occasionally flashes a light across a virtual keyboard key. If the subject is looking at the key when a light 1101a is presented, their brainAttorney Docket: 8350.2025-115WO responds cognitively 1101b to the 'oddball' stimulus. This is detectable on the EEG signal as a positive deflection approximately 300 ms post stimulus presentation in the parietal region. The signal can be fed into a classifier to determine which key the subject was looking at, and to type the prediction to the screen 1102. This speller system can be improved through combining it with neuromodulation devices 1103 targeted the parietal region 1104. The modulated brain region may produce a stronger P300 signal, which may result in more stable typing predictions.

[0090] FIG. 12 shows another embodiment of the invention to use neuromodulation to enhance BCI using steady-state visual evoked response (SSVEP). Visual evoked potentials (VEPs) depend on external stimuli, which generally consist of lights flickering with different temporal profiles. These profiles fall into five broad categories: (1) frequency modulation, (2) time modulation, (3) code modulation, (4) phase modulation, and (5) motion-onset. In all of these cases, each stimulus carries a unique temporal pattern according to the modulation scheme that can also be detected in the scalp EEG to indicate the selected stimulus. Frequency modulation and phase modulation stimuli flicker at different frequencies and / or phases. Attending to a particular target increases the band power of the corresponding oscillation in electrodes covering the occipital cortex. These particular VEPs are also known as SSVEPs. tFUS neuromodulation can be targeted at visual cortex including VI toAttorney Docket: 8350.2025-115WO facilitate visual information processing to enhance the SSVEP based BCI performance.

[0091] In this embodiment, lights 1201a flash across the keyboard at various frequencies, which are reflected in the occipital cortex's frequency response profile 1201b. These response profiles can be used to predict which key the user intended to type. Neuromodulation devices 1203 may target the occipital cortex 1204 to modulate the underlying brain activity and enhance the quality of the signal 1205.

[0092] Example of neural stimulation-based bidirectional BCI

[0093] FIG. 13 and FIG. 14 demonstrate use cases for noninvasive percept-inducing neuromodulation within the context of bidirectional BCI. For a sensorimotor BCI, a user may control a robotic arm to grab an object, and, when it does so, activated pressure sensors in the robotic hand induce tFUS firing to the somatosensory hand-knob. This causes the user to receive not just visual confirmation of grasping the object, but haptic feedback as well through induced vibration sensations in their hand (FIG. 13). This multi-modal feedback is more closely aligned with a healthy biological feedback system than a monodirectional BCI, which can lead to improved performance. In addition to generating virtual sensory feedback, in some cases, to leverage the intrinsic haptic sensory circuit of the human nervous system, single- element / multi-element-based tFUS can also modulate the somatosensory hand-knob in the brain when a vibrational actuator delivers the hapticAttorney Docket: 8350.2025-115WO feedback at the moment of the robotic arm reaching the object and activating the pressure sensors. The tFUS neuromodulation here with specific parameters can amplify the somatosensory signaling in the brain and thus boosting the sensitivity of a human subject in perceiving the events of grabbing the object. The multi-element-based tFUS with a higher spatial specificity in targeting at the somatosensory hand-knob outperforms the single-element-based tFUS.

[0094] FIG. 13 demonstrates how ultrasound-induced haptic-feedback can be used in conjunction with a noninvasive BCI. The user initiates BCI movement to grasp a glass of water by imagining 1301their hand closing. This produces a brain signal 1302detectable by the EEG, which, in turn, is converted into a command for the robotic arm to close its grip 1303. Pressure sensors on the robot relay information back to the user that they have made contact with something, and tFUS 1304 is applied to the hand region of the somatosensory cortex 1305. This causes the user to feel tactile perceptions 1306. In addition using neuromodulation to generate haptic feedback, vibrational sensations may be directly generated from actuators in cases where BCI users still have partially working sensory pathways 1307. In these instances, neuromodulation to the somatosensory cortex can be used to enhance the user's perception of these natural inputs.

[0095] A bidirectional visual prosthesis can also be achieved to restore low-level sight to the visually impaired (FIG. 14). A user, through BCI-decodedAttorney Docket: 8350.2025-115WO thoughts, can direct a camera to search for their object of interest. The camera, combined with computer vision machine learning, identifies the object's relative location to the user. tFUS activates the visual cortex neurons corresponding to that relative location, and thus the user "sees" visual phosphenes appear where the object should be. This can be extended for multiple ultrasound elements to produce a more complex stimulation pattern and encoded higher-level vision (such as the table that the water is on, or potential obstacles in the way).

[0096] In this embodiment, the user thinks 1401about they want to find, such as a glass of water. Their desire is decoded with EEG and machine learning techniques. A camera 1402 scans the area for the object of intent. Upon identifying the location of object 1403, the camera triggers a tFUS stimulation 1404 to a location in the user's visual cortex corresponding to the water's relative location. A visual phosphene 1405is created in the user's field of view corresponding to the object's location.

[0097] To enhance spatial visual processing in the brain, tFUS can also be used to enhance spatial vision perception by modulating neural activity across key visual processing regions. For examples, by targeting the primary visual cortex (VI), tFUS can improve contrast sensitivity, edge detection, and feature discrimination, leading to sharper spatial resolution. Beyond VI, higher-order visual areas (V2, V3, V4, and MT / V5) play critical roles in refining depth and motion perception. Modulating these regions with tFUS couldAttorney Docket: 8350.2025-115WO enhance motion discrimination and object tracking in dynamic environments. Additionally, the thalamus (specifically the lateral geniculate nucleus, LGN) and basal ganglia contribute to sensorimotor integration and predictive spatial tracking. By strengthening feedforward and feedback loops between these structures, tFUS could refine spatial accuracy and object localization. Furthermore, the dorsal visual stream is crucial for spatial awareness, attention, and visuomotor coordination as was previously discussed with respect to FIGS. 5-8. Applying tFUS to modulate this neural pathway can improve spatial visual processing in the brain and thus enhance the bidirectional BCI control.

[0098] A visual prosthesis and sensorimotor rhythm BCI may be combined with multi-site tFUS stimulation using several single-element transducers or multielement transducer with sufficient acoustic aperture to cover and stimulate the wanted brain targets as aforementioned. The visual prosthesis portion may allow a user to identify an object's location, and the same user may control a robotic device and / or wheelchair to retrieve the object.

[0099] Haptic feedback may also be incorporated beyond providing grasp feedback to providing information relative to proprioceptive, or muscle stretch, information. When the robotic arm moves in a particular direction, a corresponding tactile sensation can be encoded back into the user to provide a more naturalistic feeling of controlling the device. This, for example, could be as simple as inducing left / right hand vibrations when the robot movesAttorney Docket: 8350.2025-115WO left / right, or, for higher degrees of freedom, the tFUS elements can be steered and applied one or more individual finger regions.

[0100] Such implementations may also be extended to lower-body based BCI. A user may control robotic legs / exoskeleton using their brain signals, and the proprioceptive movement of the legs, or some binary force of feet being in contact with the ground, would be encoded back into the user via ultrasound to the leg / foot region of the sensorimotor cortex.

[0101] Bidirectional BCI for injury rehabilitation.

[0102] An estimated 1 in 4 adults will experience a stroke in their lifetime. The severity of the stroke varies, but it is not uncommon for the stroke to result in some form of motor impairment. The brain is plastic, and stroke recovery studies have shown that, with training, healthy brain areas can sometimes be trained to accommodate the injured area's function(s). Monodirectional BCI have been shown to be helpful in this process on their own, by providing patients a means to practice modulating their sensorimotor rhythms. Bidirectional BCI may speed up this process even more, by increasing excitability of nearby brain regions during the task, promoting plasticity and learning.

[0103] Bidirectional BCI may also be used in conjunction with an exoskeleton to augment existing motor control. Users with limited mobility, or those who may benefit from additional support, such as those working in manual labor- intensive jobs, may find some relief in the form of exoskeletons. The externalAttorney Docket: 8350.2025-115WO support provided by the exoskeleton can be controlled through noninvasive brain signal decoding. Neuronal excitability-increasing modulation can be used in conjunction with exoskeleton to improve the control signal quality. Neural stimulation and modulation may be used to provide feedback on how the exoskeleton is responding to the task (i.e., if all the weight of a heavy box is being supported by the exoskeleton, tactile sensations can be encoded back into the user to indicate good contact with the object being lifted and the assistive device).

[0104] Bidirectional BCI with MEG-sensing and neuromodulation. Similar to electrical sensing described above, neural excitation generates currents within the brain that produce electrical potential field, as well as magnetic field. Magnetoencephalography (MEG) is an established sensing modality to record brain's magnetic signals due to neural excitation. Recent wearable MEG technology such as 0PM MEG sensors enable MEG sensing similar to EEG in a wearable manner. The current invention can be practiced using MEG sensing, with decoding of brain intents from recorded MEG, and enhancement by neuromodulation targeted at relevant brain circuits to facilitate neural information processing and enhancing brain signals that are used in BCI decoding. Such implementations may include but not limited to mVEP BCI speller, SSVEP BCI speller, P300 BCI speller, and sensorimotor rhythm BCI controller.Attorney Docket: 8350.2025-115WO

[0105] Bidirectional BCI with functional near-infrared spectroscopy (fNIRS) sensing and neuromodulation. Similar to electrical sensing described above, neural excitation may be sensed using functional near-infrared spectroscopy (fNIRS) sensors over the scalp. fNIRS is an established sensing modality to record brain's hemodynamic signals. The current invention can be practiced using fNIRS sensing, with decoding of brain intents from recorded fNIRS signals, and enhancement by neuromodulation targeted at relevant brain circuits to facilitate neural information processing and enhancing brain signals that are used in BCI decoding. Such implementations may include but not limited to mVEP BCI speller, SSVEP BCI speller, P300 BCI speller, and sensorimotor rhythm BCI controller.

[0106] Bidirectional BCI with ear EEG and neuromodulation. Ear EEG using electrodes over the ears have been found providing information about brain processes. In this embodiment, scalp EEG electrodes are replaced by ear EEG electrodes, and integrated with neuromodulation devices to constitute a bidirectional BCI.

[0107] Bidirectional BCI with multimodal neural sensing and neuromodulation. Multimodal neural sensing may be used in conjunction with neuromodulation to realize a bidirectional BCI, whereas EEG, MEG or fNIRS may be used together to achieve enhanced neural sensing.

[0108] Bidirectional BCI with minimally invasive sensing and neuromodulation. While brain decoding-encoding based bidirectional BCI may greatly facilitateAttorney Docket: 8350.2025-115WO noninvasive BCI, the present invention may also facilitate minimally invasive BCI using sub-skin electrodes, ex-dural electrodes or sub-dural electrodes for long term recordings. Wearable neuromodulation devices can be integrated with such minimally invasive sensing device to improve BCI performance.

[0109] Bidirectional BCI integrating neural sensing and neuromodulation with application of Al and machine learning. Al (artificial intelligence), deep learning or other machine learning approaches may be used to facilitate and optimize the brain decoding, and controlling neuromodulation dosage and ways of stimulations based on analysis of sensed neural signals from an individual. Such machine learning approaches include analyze and decode brain signals using conventional algorithms such as common spatial filters, autoregressive models, support vector machine, etc. They also may include convolution neural networks, recurrent neural networks, or other form of artificial neural networks for signal processing, classification and decoding, as well as controlling the neuromodulation stimulation.

[0110] In addition to enabling bidirectional interfacing, which may significantly enhance the quality of relevant neural signals, neuromodulation plays a crucial role in encoding state-specific feedback, allowing the system to provide tailored sensory or cognitive responses to the user. This feedback capability is particularly valuable for applications requiring fine motor control and awareness, as it can mimic naturalistic sensory experiences that are otherwise absent in traditional brain-to-machine interfaces. Specifically, tFUSAttorney Docket: 8350.2025-115WO offers a high degree of spatial precision and has been shown to evoke tactile perceptions when applied to the primary somatosensory cortex, evoke phosphene when directed onto the visual cortex, impact decision making process when targeting at a specific prefrontal cortical area. For example, the capability of evoking the tactile sensation enables tFUS to be used in encoding sensory feedback specific to user movements and object interactions in a robotic arm-controlling BCI paradigm, thus enhancing the immersive experience by simulating force and proprioceptive (muscle stretch) information, which are naturally present in healthy human biological systems.

[0111] In typical BCI setups, users lack the proprioceptive and force-related feedback that humans usually receive from muscle stretch and tactile sensors during movement and object manipulation. By combining sensorimotor rhythm (SMR) decoding with evoked tactile feedback using neuromodulation, such as tFUS, users may experience more naturalistic and intuitive control over robotic or prosthetic devices, effectively bridging the gap between native limb control and artificial interfaces. Additionally, these neuromodulation-induced feedback mechanisms may extend beyond tactile sensations to include visual or cognitive feedback components, aligning with the broader concept of bidirectional BCIs that support multi-modal feedback loops for improved control and user engagement.Attorney Docket: 8350.2025-115WO

[0112] Control over neuromodulation parameters, such as frequency, intensity, and pulse duration of tFUS, is essential to customize the feedback and optimize the bidirectional BCI for specific tasks or user needs. This modulation of parameters allows the BCI system to adjust in real-time to variations in the acquired neural signals or intended movements, potentially enhancing decoding accuracy and the fidelity of the feedback. These parameter adjustments may also provide safety and comfort improvements by limiting unwanted stimulation effects and aligning the feedback intensity with the sensory perception threshold of the user.

[0113] In summary, by integrating precise parameter control for neuromodulation to induce state-specific sensations, visual cues, or cognitive feedback within a bidirectional BCI framework, the invention enables a more holistic and effective interaction between the user and the machine. This advanced approach aims to replicate natural feedback pathways, ultimately contributing to smoother, more intuitive, and context-aware control in braincomputer interface applications.

[0114] As would further be realized by one of skill in the art, many variations on implementations discussed herein which fall within the scope of the invention are possible. Specifically, many variations of the parameters, the components used, and their arrangement could be used to obtain similar results. The invention is not meant to be limited to the particular exemplary model disclosed herein. Moreover, it is to be understood that the features ofAttorney Docket: 8350.2025-115WO the various embodiments described herein were not mutually exclusive and can exist in various combinations and permutations, even if such combinations or permutations were not made express herein, without departing from the spirit and scope of the invention. Accordingly, the method and apparatus disclosed herein are not to be taken as limitations on the invention but as an illustration thereof.

Claims

Attorney Docket: 8350.2025-115WOClaims:

1. A bidirectional closed-loop brain-computer interface (BCI) system comprising: a neural sensing module configured to detect brain signals in a brain of a subject; a signal processing unit configured to decode brain intention and status from the detected brain signals using machine learning and / or signal processing algorithms; and a neuromodulation module configured to apply at least one of acoustic, electrical, magnetic, or optical energy to a targeted brain region associated with the brain intention, to modulate neural activity.

2. The system of claim 1 wherein the neural sensing module uses at least one of electrophysiological, hemodynamic, optical, or acoustic sensing techniques to detect the brain signals.

3. The system of claim 1 wherein the brain signals are detected noninvasively.

4. The system of claim 1 wherein the neuromodulation module applies neuromodulation to increase the amplitude of the brain signal.

5. The system of claim 4 wherein the neuromodulation depolarizes neurons in the brain, such that the neurons are closer to an action potential threshold and thus moreAttorney Docket: 8350.2025-115WO susceptive to firing when further excited by another neurons.

6. The system of claim 5 wherein the neuromodulation excites and further activates relevant neural circuits to strengthen them.

7. The system of claim 1 wherein application of the neuromodulation changes the amplitude of the brain signal.

8. The system of claim 7 wherein the neuromodulation hyperpolarizes neurons in the brain, such that the neurons are further away from an action potential threshold, and thus less susceptible to firing.

9. The system of claim 7 wherein the i neuromodulation inhibits and weakens extraneous neural circuits.

10. The system of claim 1, wherein the neural sensing module comprises electroencephalography (EEG) electrodes for detecting sensorimotor rhythms, event- related potentials, evoked potentials, or spontaneous EEG.

11. The system of claim 1, wherein the neuromodulation module comprises a transcranial focused ultrasound (tFUS) device configured to apply ultrasonic energy to the targeted brain regions to modulate neural circuit activity.Attorney Docket: 8350.2025-115WO12. The system of claim 11, wherein the tFUS device is configured to enhance brain signal detectability by modulating the excitability of targeted neurons, thereby improving the accuracy and reliability of BCI control.

13. The system of claim 1, wherein parameters of the applied energy are dynamically adjusted based on real-time feedback from the neural sensing module.

14. The system of claim 1, wherein the neuromodulation module is configured to induce sensory perceptions in the somatosensory cortex to provide haptic feedback for bidirectional communication.

15. The system of claim 1, wherein the neuromodulation module is configured to induce visual percepts in the visual cortex to facilitate sensory substitution for visually impaired individuals.

16. A method for closed-loop bidirectional brain-computer interfacing comprising: detecting brain signals from a subject using a neural sensing module; decoding brain intention and status from the detected signals using a signal processing unit; and applying neuromodulation to targeted brain regions using a neuromodulation module to enhance neural circuit activity.Attorney Docket: 8350.2025-115WO17. The method of claim 16, wherein the neural sensing module detects sensorimotor rhythms to facilitate motor imagery or motor execution based BCI control.

18. The method of claim 16, wherein the neuromodulation module applies tFUS to the motor cortex to enhance movement-related brain activity for BCI control.

19. The method of claim 16, wherein the neuromodulation module applies tFUS to the somatosensory cortex to induce haptic perceptions for bidirectional interaction.

20. The method of claim 16, wherein the neuromodulation module applies tFUS to the visual cortex to induce visual percepts and spatial visual processing for sensory augmentation and object localization.

21. The method of claim 16, wherein the neural sensing comprises EEG recordings using a plurality of electrodes on the surface of head.

22. The method of claim 21, wherein brain activity induced by the neuromodulation is localized by electrophysiological source imaging and is used to guide refinement and optimize brain target of neuromodulation.

23. A noninvasive bidirectional brain-computer interface (BCI) device comprising:Attorney Docket: 8350.2025-115WO a plurality of EEG electrodes for neural sensing; a machine learning-based signal processing unit for decoding brain signals; a tFUS module for noninvasive neuromodulation; and a closed-loop controller for real-time BCI adaptation and optimization.

24. The device of claim 23, wherein the tFUS module is configured to apply stimulation to multiple brain targets simultaneously for enhanced BCI performance.

25. The device of claim 23, wherein the closed-loop controller adapts neuromodulation parameters in response to real-time changes in brain activity to optimize BCI control.

26. A method for neurorehabilitation using a bidirectional BCI system, comprising: detecting and decoding neural activity associated with motor intention in individuals with neurological disorders; applying neuromodulation to motor-related brain regions to enhance neuroplasticity and motor recovery; integrating real-time neural feedback to improve rehabilitation outcomes and optimize training paradigms.Attorney Docket: 8350.2025-115WO27. The method of claim 26 wherein the individuals with neurological disorders are stroke patients or individuals with spinal cord injuries and further wherein the individuals are assisted in regaining motor function through BCI-driven neurostimulation.

28. A method for treating neurological and neuropsychiatric disorders using a bidirectional BCI system, comprising: sensing and decoding brain activity to identify dysfunctional neural patterns associated with various neurological and neuropsychiatric conditions; applying neuromodulation to regulate aberrant neural activity and restore functional brain connectivity; implementing closed-loop control to dynamically adjust neuromodulation parameters based on real-time brain state assessments; wherein cognitive function and emotional regulation and enhanced through targeted brain stimulation.

29. The method of claim 28 wherein the neurological conditions include Parkinson's disease, epilepsy, depression, or chronic pain.

30. A bidirectional BCI system for neurorehabilitation and treatment of neurological disorders, comprising:Attorney Docket: 8350.2025-115WO a neural sensing module for detecting pathological neural activity; a neuromodulation module configured to deliver precision brain stimulation to restore functional neural circuits; a therapeutic feedback system that adapts treatment based on patient response; and a machine learning-based system for personalized therapy optimization.

31. The system of claim 30, wherein the neuromodulation module applies targeted stimulation to improve gait control and balance in individuals with neurodegenerative disorders.

32. The system of claim 30, wherein the neuromodulation module facilitates neural recovery in traumatic brain injury patients by promoting cortical reorga nization and connectivity restoration.

Images