Method and system for controlling the operation of a device to assist in initiating patient movement.

Abstract

[0003] Systems and methods for promoting motor function are described herein. In one aspect, a computer-implemented method for assisting a patient in initiating movement can include receiving a set of neural signals from a set of neural sensors, extracting a set of features from the set of neural signals, inputting the set of features into a classification model, determining an attempted activity of a user from the classification model, and transmitting a set of stimulation signals to one or more output effectors according to the attempted activity and the set of neural signals. TIFF2023537835000002.tif177140

Description

Technical Field

[0001] Cross - reference to Related Applications This application claims priority to U.S. Provisional Patent Application No. 63 / 049,754, filed on July 9, 2020, which is hereby incorporated by reference in its entirety.

Background Art

[0002] Background of the Invention Conventional physical therapy techniques require extensive face - to - face interaction between the patient and the therapist. For example, the therapist instructs the patient on how to perform a particular movement, the patient attempts to perform the movement, and the therapist responds with an evaluation (e.g., how to perform the movement better). This interaction is repeated, and as the patient's physical condition changes, the therapist's instructions are modified to adapt to the patient's condition.

[0003] Although automated assistive rehabilitation systems and technologies are in an early stage, these systems and technologies are suffering from similar problems. For example, while patients can use the rehabilitation system off - site, they must return on - site periodically to provide data to the therapist or system manager. Then, the therapist or system manager can calibrate the system according to the patient's needs.

[0004] Furthermore, conventional rehabilitation systems and technologies cannot identify some patterns of a patient's movement. Some patients may have lost the ability to initiate movement traditionally. Although patients may still retain a secondary pattern of movement that can be initiated by the patient during an attempt at movement, this secondary pattern is insufficient in itself to actually perform the movement.

Summary of the Invention

[0005] Summary Systems and methods for promoting motor function are described herein. In one aspect, a computer implementation method for assisting a patient initiating movement may include the steps of receiving a set of nerve signals from a set of nerve sensors, extracting a set of features from the set of nerve signals, inputting the set of features into a classification model, determining the user's attempted activity from the classification model, and transmitting a set of stimulation signals to one or more output effectors according to the attempted activity and the set of nerve signals.

[0006] This aspect can take on various forms. In one embodiment, the computer implementation method may further include a stage of training a classification model, the training stage including receiving a set of training neural signals from a set of neural sensors, receiving input indicating actions performed by a trainer, extracting a set of training features from the set of training neural signals, and mapping the set of training features to directive actions.

[0007] In some cases, the training phase of the classification model further includes determining feature thresholds from the mapping, and the user's attempted activity is further determined from features among the set of features that reach the feature threshold. In some cases, the trainer may include the user, a physical therapist, an occupational therapist, or a combination thereof.

[0008] In another embodiment, the computer implementation method may further include the steps of identifying a set of proportional values ​​between a set of neural signals and the user's attempted activity, and generating a set of stimulation signals according to the set of proportional values.

[0009] In some cases, the step of identifying a set of proportional values ​​is achieved through multilayer perceptron networks, convolutional neural networks, genetic algorithms, binary particle swarm optimization processes, generative adversarial networks, polynomial and radial basis kernel support vector machines, Kalman filters, generalized linear mixed models, particle filters, random forest algorithms, rotation forest algorithms, or combinations thereof.

[0010] In another embodiment, the computer implementation method may further include a step of identifying an activation pattern from the received neural signal, and a step of determining the attempted activity according to the identified activation pattern.

[0011] In another embodiment, the computer implementation method may further include a step of modifying the classification model according to a set of features.

[0012] In another embodiment, the set of neural signals may include scalp EEG, subgalea aponeurotica EEG, intraosseous EEG, epidural EEG, subdural EEG, intracortical LFP, deep EEG, single-unit recording, heart rate, heart rate variability, respiratory rate, electrocutaneous conductance, blood glucose level, pupil diameter, extraoculogram, electromyography, positioning of user body parts, user kinematic and dynamic signals, voice signals, keyboard entries, mouse clicks, joystick use, or a combination thereof.

[0013] In another embodiment, the output effector may include a set of electrical contacts, an electro-assistive device, a brain-computer interface, or a combination thereof.

[0014] In another aspect, the system includes a neural signal processor configured to receive a set of brain signals from a user, digitize the set of brain signals, and store the digitized brain signals in a buffer; a neural signal analyzer configured to retrieve the digitized brain signals from the buffer, identify a set of spike counts, a local field potential (LFP), or a combination thereof from the digitized brain signals, extract a set of features from the set of spike counts and the LFP, input the set of features into a classification model, identify the movement of the user's attempted movement from the classification model, generate a movement control command according to the movement of the attempted movement, and transmit the movement control command; and a rehabilitation prosthesis configured to receive the movement control command and generate a corresponding movement according to the movement control command. [Invention 1001] The stage of receiving a set of nerve signals from a set of nerve sensors, The stage of extracting a set of features from a set of neural signals, <000006O> At the stage of inputting a set of features into the classification model, <000006Z> The stage of determining the user's attempted activities from the classification model, and The step of transmitting a set of stimulus signals to one or more output effectors according to the set of activity and neural signals attempted. A computer implementation method for assisting initiating patient movement, including the above. [Invention 1002] This further includes the stage of training the classification model, The training stage is, Receiving a set of training neural signals from a set of neural sensors, Receiving input indicating actions performed by the trainer, <0000O79> Mapping a set of training features to directive actions and including, A computer implementation method for the present invention 1001. [Invention 1003] A computer implementation method of the present invention 1002, wherein the stage of training a classification model further includes determining a feature threshold from the mapping, and the user's attempted activity is further determined from features among the set of features that have reached the feature threshold. [Invention 1004] A computer implementation method of the present invention 1002, wherein the trainer includes a user, a physical therapist, an occupational therapist, or a combination thereof. [Invention 1005] The step of identifying a set of proportional values ​​between a set of neural signals and the user's attempted activity, and The stage of generating a set of stimulus signals according to a set of proportional values. A computer implementation method of the present invention 1001, further comprising: [Invention 1006] A computer implementation method of the present invention 1005, wherein the step of identifying a set of proportional values ​​is brought about via a multilayer perceptron network, a convolutional neural network, a genetic algorithm, a binary particle swarm optimization process, a generative adversarial network, a polynomial and radial basis kernel support vector machine, a Kalman filter, a generalized linear mixed model, a particle filter, a random forest algorithm, a rotation forest algorithm, or a combination thereof. [Invention 1007] A computer implementation method of the present invention 1001, further comprising the step of identifying an activation pattern from received neural signals, wherein the step of determining the attempted activity follows the identified activation pattern. [Invention 1008] A computer implementation method of the present invention 1001, further comprising the step of modifying a classification model according to a set of features. [Invention 1009] A computer implementation method of the present invention 1001, wherein the set of nerve signals includes scalp EEG, subgalea aponeurotica EEG, intraosseous EEG, epidural EEG, subdural EEG, intracortical LFP, deep EEG, single-unit recording, heart rate, heart rate variability, respiratory rate, electrocutaneous conductance, blood glucose level, pupil diameter, extraoculogram, electromyography, positioning of user body parts, user kinematic and dynamic signals, voice signals, keyboard entry, mouse click, joystick use, or a combination thereof. [Invention 1010] A computer implementation method according to the present invention 1001, wherein the output effector includes a set of electrical contacts, an electrical prosthesis, a brain-computer interface, or a combination thereof. [Invention 1011] The system receives a set of brain signals from the user. Digitizing the set of brain signals, Digitized brain signals are stored in a buffer. A neural signal processor configured in such a way; Search the buffer for digitized brain signals, From digitized brain signals, we identify sets of spike counts, local field potentials (LFPs), or combinations thereof. Extract a set of features from the spike count set and LFP, Input the set of features into the classification model, The movement patterns attempted by the user are identified using a classification model. It generates motor control commands according to the movement of the attempted motion, and Transmits motion control commands. A neural signal analyzer configured in such a way; It receives a motor control command, and It generates the corresponding movement according to the motor control command. Rehabilitation assistive devices and A system that includes this. [Brief explanation of the drawing]

[0015] To better understand the nature and desired purpose of the present invention, the following detailed description, considered together with the accompanying drawings, is provided for reference, and the same reference numerals indicate corresponding parts across several figures.

[0016] [Figure 1] This disclosure describes a system for promoting motor function according to an aspect of this disclosure. [Figure 2]This disclosure provides a workflow process for promoting motor function according to an aspect of this disclosure. [Figure 3] This disclosure provides a workflow process for promoting motor function according to an aspect of this disclosure. [Figure 4] This disclosure provides a workflow process for promoting motor function according to an aspect of this disclosure. [Figure 5] This document shows a graphical user interface for a motor function enhancement system according to aspects of this disclosure. [Figure 6] This document shows a graphical user interface for a motor function enhancement system according to aspects of this disclosure. [Figure 7] The results of a Fast Fourier Transform (FFT) from a received electronic input signal of a motor function enhancement system according to an aspect of this disclosure are shown. [Figure 8] This disclosure provides a graphical user interface for a virtual arm / real arm operator according to an aspect of this disclosure. [Figure 9] This disclosure provides a graphical user interface for a virtual arm / real arm operator according to an aspect of this disclosure. [Figure 10] This document presents a high-level overview of a motor function enhancement system in accordance with the aspects of this disclosure. [Figure 11] This document shows a system architecture for a motor function enhancement system according to aspects of this disclosure. [Figure 12] This document shows a system architecture for a motor function enhancement system according to aspects of this disclosure. [Figure 13] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 14] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 15] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 16] This document describes a software architecture for a wireless controller module according to an aspect of this disclosure. [Figure 17]This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 18] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 19] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 20] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 21] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 22] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 23] This disclosure provides a software architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 24] This disclosure shows a physical layer architecture for a motor function enhancement system according to an aspect of this disclosure. [Figure 25] This disclosure shows a motor function enhancement system according to an aspect of this disclosure. [Figure 26] This disclosure shows a motor function enhancement system according to an aspect of this disclosure. [Figure 27] This document shows a clinical trial timeline for implementing the motor function enhancement system according to the aspects of this disclosure. [Figure 28] This disclosure shows a motor function enhancement system according to an aspect of this disclosure. [Figure 29]The neuroimaging results for patients participating in the clinical trial are shown. The results illustrate the diffusion sequence at the time of acute stroke. Marked diffusion restriction is observed in the right lenticular nucleus and adjacent white matter (Panel (a)). T2-weighted MRI at 2 years shows areas of cerebral softening and relative ventricular enlargement (Panel (b)). Brain functional imaging revealed an activation hotspot at the depth of the central sulcus along the hand knob area of ​​the precentral gyrus (indicated by circles) (Panel (c)). Three-dimensional reconstruction of the participant's cortical surface derived from MRI imaging the movement centroid of left hand activity (indicated by circles) (Panel (d)). Shading indicates areas responding to sensory stimulation of the left hand. Squares indicate microelectrode arrays. [Figure 30] The waveforms of action potentials recorded from participating patients, in accordance with the aspects of this disclosure, are shown. [Figure 31] This shows neuronal activity associated with movements performed in the paralyzed limb. Participants were asked to perform a series of left limb movements (indicated by horizontal coordinates) over 110 seconds. Verbal movement commands are indicated by hash marks. Rasters show the time of each action potential. Under each raster, the normalized integral firing rate derived by the "leakage integrator" formula appears; normalization was achieved by dividing by the maximum integral firing rate from the spike sequence of each unit over the displayed period. The upper unit (channel 61) was more active for hand compression than for wrist extension compared to the lower, similarly recorded unit (channel 62). Participants performed all movements. Such motions required exertion, and participants were unable to maintain a consistent level of activity for each cue, exhibiting variable reaction times. Participants were prone to fatigue and required rest and repositioning. [Figure 32]The graphs show cumulative integrated spike activity across the channel (upper graph) and residual left forearm electromyography (lower graph), which vary depending on joint position. The total spike activity across the channel, passing through the leakage integrator, appeared to vary due to specific residual activity in the left upper limb. Proximal residual activity produced a pattern that appeared normal. Biceps and triceps activity alternate, as seen at 290–310 seconds in the lower panel. However, in the distal part of the upper limb, wrist flexor and extensor activity tend to occur together in an abnormal synergistic manner. Furthermore, wrist flexor activity is abnormally synergistic with biceps activity (abnormal flexor synergy). The total integrated spike activity across the channel appears to covariate with wrist flexor activity. [Modes for carrying out the invention]

[0017] definition The present invention is best understood by reference to the following definitions.

[0018] As used herein, the singular forms “a,” “an,” and “the” include multiple references unless the context explicitly indicates otherwise.

[0019] Unless otherwise specified or made clear from the context, the term “about” as used herein is understood to mean within the normal range of acceptance in the art, for example, within two standard deviations from the mean. “About” can be understood to mean within 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, 0.5%, 0.1%, 0.05%, or 0.01% of the stated value. Unless otherwise made clear from the context, all numerical values ​​provided herein are qualified with the term “about.”

[0020] As used herein and in the claims, the terms “comprises,” “comprising,” “contains,” “having,” and others may have the meanings they have in U.S. patent law, and may mean “includes,” “including,” and others.

[0021] Unless otherwise specified or made clear from the context, the term "or" as used herein is understood to be inclusive.

[0022] The ranges provided herein are understood to be abbreviations for all values ​​within that range. For example, the range 1–50 is understood to include any number, combination of numbers, or subrange from the group consisting of 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, or 50 (as well as their fractional parts unless the context explicitly indicates otherwise).

[0023] Detailed description of the invention Motor function promotion system A motor function enhancement system may include a motor function device having modular components, software, and calibration protocols. The motor function enhancement system can be used to restore independent voluntary movement of the hands, arms, trunk, and legs in adults and children with weakness or paralysis due to neurological disorders or injuries.

[0024] The software infrastructure and calibration approach of the motor function enhancement system can be extended to other types of treatment, including rehabilitation and daily function assistance for people with cognitive impairments resulting from neurological disorders and injuries; as an adjunct to the treatment of mental disorders by incorporating the principles of cognitive behavioral therapy, dialectical behavior therapy, mindfulness therapy, and hypnotherapy; and to the treatment of non-neurological conditions for which long-term or telemedicine-mediated follow-up is valuable, such as for orthopedic or cardiac surgery post-rehabilitation.

[0025] The motor function enhancement system can be deployed as a medical device designed to treat adults and children with neurological disorders and injuries such as paralysis, paresis, or coordination disorders caused by spinal cord injury, stroke, ALS, TBI, MS, muscular dystrophy, neuropathy, transverse myelitis, brachial plexus injury, amputation, tumor resection, or cognitive impairment caused by autism spectrum disorder, Alzheimer's disease, TBI, stroke, Parkinson's disease, DLB, FTD, MSA, PSP, CBD, PPA, MS, CP, chromosomal abnormalities, and others. Furthermore, the motor function enhancement system can also be used as part of rehabilitation, functional recovery, or both for children and adults with weakness, apathy, fatigability, or other symptoms caused by non-neurological conditions such as chemotherapy for cancer or autoimmune conditions, recovery from cardiovascular conditions or surgery, and orthopedic interventions.

[0026] The motor function enhancement system can be deployed as a consumer product for healthy individuals to provide a novel input device, to enhance performance in military or industrial settings, to refine athletic skills, or for entertainment purposes. In addition, the system's software and calibration protocols can be used not only to provide control of telephones, computers, and other devices, but also for text entry, communication, and productivity (e.g., via mechanical sensors, electromagnetic sensors, and other wearable sensors) for physically healthy children and adults.

[0027] The motor function enhancement system includes the following: 1) To meet the user's needs; 2) To achieve the user's functional objectives; 3) Incorporate user preferences regarding device use / embedding and technology (e.g., whether the person wants to wear the component all the time or only in specific situations); 4) Incorporate incremental adjustments to complexity, informed by outcomes achieved through use over time. This could include an optimal network of devices and tools for that purpose.

[0028] A key technological innovation in motor function enhancement systems is their ability to function as a service rather than a static device. In this sense, motor function enhancement systems can be "life services" using compatible modular components that constantly respond to user needs and incorporate changes in the user's abilities and goals. Motor function enhancement systems can be scalable and flexible platforms that allow for the integration of multiple neural control devices, sensors, and software tools.

[0029] The software for the motor function enhancement system can manage, control, and operate a network of commercially available devices and sensors with the goal of achieving optimal rehabilitation / support outcomes. Furthermore, the motor function enhancement system can be a modular, flexible, scalable, and expandable closed-loop system capable of sensing patient-specific physiological signals and using these to generate optimal control and activation commands for a network of patient-specific neuromodulatory tools. Figure 3 shows the workflow for the feature optimization and classification optimization processes implemented by the motor function enhancement system. The motor function enhancement system can be highly customizable based on medical and patient needs. The motor function enhancement system can be used in multiple settings, including remote health support at home. To facilitate home use, the motor function enhancement system software can feature a patient-specific user interface (UI). This UI can provide necessary daily and task-specific adjustments (parameter settings and stimulation type and dosage) without the need to change the settings of individual neuromodulatory devices.

[0030] component A motor function enhancement system may include several auxiliary actuators, electrical stimulators, and / or electrical signal sensors. For example, a motor function enhancement system may include or incorporate wearable power devices such as the following examples: OmniHi5®, Myomo®, and MyoPro® motorized orthoses; upper limb rehabilitation / feedback systems such as MyndMove®; lower limb function systems such as WalkeAide®; whole-body exoskeletons such as SuiteX® and Ekso®; lower limb rehabilitation and functional mobility systems such as Myolyn® and MyoCycle®; Empi® portable neuromuscular electrical stimulator; Cyclone Xcite portable multi-channel FES therapy system; Zynex Medical Neuromove; Bioness, L300, L300 Plus, and H200 systems; Saebo MyoTrac Infiniti biofeedback electrical stimulator, and others.

[0031] Software environment and graphical user interface The motor function enhancement system can interconnect with a user performance tracking system and link with software application programs (apps) such as calendars, alarms, emails, reminders, medication management, and others. The motor function enhancement system's software can provide entries for information about a specific user, such as medical history, educational history, and previous test results, including motor tests, neuropsychological tests, or other tests. The software can access a normative database for specific tasks and the user's previous baselines to generate standard deviations and other scoring of performance on a per-use or per-calibration basis. The software can display visual information (e.g., images, videos, etc.) on the display screen, for example, either on the left side or across the entire screen. In some cases, corresponding text can also be displayed (e.g., on the right side of the screen), and audio (e.g., voice commands) can be played through the computer or mobile device's speaker.

[0032] Figure 1 shows a motor function enhancement system according to an embodiment of the claimed invention. The system may include a manager suite 105. Manager suites 1-5 can receive sensed electrical data, input setting parameters, and other operational characteristics such as operational characteristics of a Bluetooth wireless controller 130 having input / output components. In some cases, the wireless controller may be implemented by a wireless telecommunication protocol such as Zigbee or others.

[0033] The system may also include various electrical data sensors (e.g., those transmitting neural activity 110, physiological signals 115, electrical stimulation 125, hand orthoses 135, instrumentation gloves 145, arm rehabilitation units 140, and others). In some cases, the sensors may include wearable sensors (e.g., long-term or intermittent wearables) for the user, along with a microcontroller and signal acquisition. A battery can power the sensors, controller, wireless components, and motors.

[0034] Some units within the system can also function as output effectors. For example, the arm rehabilitation unit 140, instrumentation glove 145, hand orthosis 135, and others can also transmit electrical signals to the user (e.g., via the manager suite 105).

[0035] Input source Figure 2 shows an exemplary motor function enhancement system (e.g., the FREEDOM system) according to the claimed aspects of the present invention. This system may include wearable or implantable sensors (e.g., a sensor network 205). Wearable or implantable sensors may include electroneural sensors (e.g., low-impedance EEG, high-impedance microelectrodes as individual wires or arrays, bioconstructs; sensors in the scalp, subgalea aponeurotica, intraosseous, epidural, subdural, intracortical or deep locations, etc.), electrocardiovascular sensors (ECG for heart rate or heart rate variability), neuromuscular sensors (EMG), mechanical switches, respiratory rate sensors, skin electroreactivity sensors, temperature sensors, accelerometers / gyroscopes (external wearable or implantable), cameras, microphones, keyboards, mice, touchpads, touchscreens, joysticks, off-the-shelf user input devices, motion capture sensors, quotient limb, head or eye jitter, eye movement (visual or infrared camera or EOG) sensors, pupil diameter, decoded facial state sensors, voice analysis sensors, and others. User control (e.g., user control 210) may include a mouse, actual keyboard / keypad, virtual keyboard / keypad, hand gestures, actual / virtual button clicks / dwells, cursor trajectory, cursor position, respiratory input (sip / puff), EMG controller, voice activation, and others. Physiological / electrical signals captured by these sensors may be referred to herein as “neural signals.”

[0036] Because input sources can be dynamically removed or added, motor function enhancement systems can be scalable. For example, a user might find it more difficult to use certain surface EMG input signals, and as a result, they may be removed as possible inputs, while new external or embedded sensors can provide new input sources that can be added.

[0037] Output effector The motor function enhancement system may also include output effectors (e.g., an output device network 215). After a signal is recorded and decoded (e.g., by a manager suite 105), the signal can be expanded to trigger several actions. This expansion may include display text, images, or videos shown on a screen (desktop, laptop, phone), or audio (language, music), and even olfactory cues. Vibratory tactile feedback outputs may include wearable, implantable, or device-type tactile devices (chair, phone) or electrical stimulation (tactile "phosphene").

[0038] Two of the most frequently used output effectors to restore movement are motor initiation and electrical stimulation. A motor, positioned on a rigid brace with a movable rod component, can induce movement in a hinge joint. Electrical stimulation can be used to contract the underlying muscle or muscle group (functional electrical stimulation). These two approaches can be used independently to achieve movement across the same joint, or simultaneously in parallel on one or different joints (e.g., the elbow and wrist).

[0039] When a given output is triggered, it can be a binary (e.g., a motor moving continuously from one position to another), continuous (proportional), or more complex, pre-programmed sequence. The motor function enhancement system allows for the pre-programming of different movements and stimuli targeting various muscles. The protocol may involve input from a healthcare provider (e.g., a physical or occupational therapist) to ensure correct setup. System calibration can overcome limitations imposed by other commercially available pre-programmed functional electrical stimulation systems. For example, the motor function enhancement system can modify pre-programmed motions with respect to the quality of motion and the timing of individual muscle groups. Muscle activation sequences can be tailored to the individual patient's needs without the consistent manual intervention from a therapist required by other systems. The motor function enhancement system can automate the output for specific users in a customizable manner without manipulating centralized manual setup and control. Furthermore, the motor function enhancement system can incorporate the use of multiple sensors, some for control and some for feedback about the system itself (e.g., strain gauges, electrometers, push buttons, capacitive buttons, and accelerometers for joint angles), ensuring the system is closed-loop and can be operated in the absence of a medical therapist. Additionally, the motor function enhancement system allows for the dynamic removal or addition of output effectors as needed.

[0040] One or more orthotic or assistive device components can be installed as a system (for example, for starting or stimulation). These components can be purchased off-the-shelf or custom-made using 3D printing technology for users, including the same user over time (such as growing from childhood to adulthood).

[0041] Decoder and mapping The closed-loop nature of the motor function enhancement system allows for dynamic and efficient stimulation methods and the real-time integration of different technologies. As the patient continues to use the system, artificial intelligence (AI) can update and modify parameters and output controls. The system can continuously explore optimal settings based on the patient's performance data. Furthermore, an alarm system can be implemented to ensure the efficacy of the treatment and the safety of the patient. In a rehabilitation setting, the system provides automated feedback and task guidance, thereby allowing patients more time to train effectively. This allows healthcare providers to spend time with patients focusing on interventions and treatments, rather than having to observe repetitions of the same task. The closed-loop nature of the system can also provide healthcare providers and insurance payers with quantitative outcome evaluation metrics that can be used to 1) provide information on intervention protocols and 2) keep track of change and progress.

[0042] The motor function enhancement system can also trigger electromagnetic stimulation using a variety of device components at various anatomical sites and with various settings. Components may include metal contacts, disk electrodes, soft conductive pads, bio-constructs, and others. Furthermore, stimulation may include transcranial stimulation of the scalp, subgalea aponeurotica stimulation, subdural stimulation, deep tissue stimulation, spinal cord stimulation, skin stimulation including peripheral sites (e.g., median nerve, posterior tibial nerve), vagus nerve stimulation (implanted or from the cervical or external auricular branch VNS), and others. Current can be delivered directly through the stimulating component as alternating current, variable current, random noise bias current, or frequency sweep. The system can also implement multi-site stimulation, which may include crossed short pulses or temporal interference.

[0043] In addition to simply triggering the start or end of stimulation at a given site, the motor function enhancement system can analyze the decoded signal and adjust stimulation parameters such as pulse width, pulse waveform, pulse polarity, pulse amplitude, frequency, duration, and grouping (single pulse, train, frequency sweep), with its contacts acting as source and sink, among others. In some cases, the motor function enhancement system can also be used to trigger the control of a magnetic stimulation device.

[0044] A motor function enhancement system can be configured to apply a given type of stimulus (electric, mechanical, thermal, etc.) based on the stage of a functional task, for example, during a specific phase of rehabilitation exercise, when the user has completed setting a timer, when a decoded signature of fatigue or involvement is received via a specific language, visual or tactile cue or command, or when a sensor becomes aware of the task situation (for example, when a hand is next to a water bottle with an RFID chip attached and the device is alerted to prepare to grasp it, or when a hand is non-contactually slid to a wall switch to turn it on), or other factors.

[0045] The motor function enhancement system can also adjust the settings of any given input signal or output source. For example, the root mean square of the average input signal recording muscle activity can be calculated. Similarly, a user, healthcare provider, or software can set a threshold at which input signals intersect to trigger a given output. The output range of a given effector can be arbitrarily constrained or expanded (e.g., the allowable current amplitude range for FES, angular displacement of motion, or net joint change from acceleration measurement or strain gauge feedback, etc.). The system software can also provide the user or service provider with the ability to apply certain scaling and transformation rules to inputs and outputs (e.g., a given input signal can be band-passed, scaled by a multiplier, offset by a coefficient, or convolved using a filter of any shape, such as the sigma transformation common in sensory physiology).

[0046] The system decoder can be configured to deploy in real time. Therefore, recorded signals can continuously trigger various output actions in real time as the user engages in daily life and diverse activities. While offline analysis can be performed by technicians or healthcare providers, the motor function enhancement system can also achieve functional movements performed in that moment.

[0047] Discrete decoding Discrete decoding refers to receiving a given input signal and mapping this signal to one or more discrete output states. For example, an input signal can be mapped to a specific motion, such as a fully extended or fully flexed elbow. More than one input signal can be combined to decode a discrete state, and a given decoded discrete state can achieve more than one effect (for example, if the state "move hand towards head" is decoded, multiple motion effectors can be activated to achieve the desired motion).

[0048] The motor function enhancement system can discretely map any arbitrary assignment between inputs and outputs. If input signals are combined, or if a given input signal is noisy or complex, a formal decoding algorithm can be deployed. In some cases, a Bayesian decoder can be implemented that combines calibration data with the prior probability of a given action. The system's software can be instructed on contextual states under various input constraints (e.g., practice state, home-based daily state, academic state, task-specific state, etc.). The motor function enhancement system can also include a graphical display for the user to view the decoder output (e.g., before the output is mapped to the effector).

[0049] Continuous decoding In addition to decoding discrete states, a motor function enhancement system can include an extensible algorithm bank to identify continuously fluctuating output states. Therefore, instead of simply creating trigger rules based on thresholds (e.g., the root mean square of the EMG signal exceeds a predefined threshold), a continuous decoder can vary the output over a proportional scale range (for example, the amplitude of the EMG root mean square can be continuously scaled to precise joint angles rather than full flexion or full extension). Decoder options can include multilayer perceptron networks, convolutional neural networks, genetic algorithms, binary particle swarm optimization, generative adversarial networks, polynomial and radial basis kernel support vector machines, Kalman filters, generalized linear mixed models, particle filters, random forests, rotational forests, and others. Principal component analysis and independent component analysis can be implemented in both discrete and continuous decoders.

[0050] Rapid state-space search In addition to mapping input signals to output effectors, various discrete and continuous decoders can also explore the state space of the input signal to constrain subsequent calibration. In some cases, the calibration procedure may involve having the user imagine, attempt, or (if physically possible) perform a specific action. In other cases, the calibration procedure may involve the user observing another individual or virtual avatar performing the action. In some cases, the calibration procedure may involve an output effector passively "dragging" the user. However, further control signals can be derived by having the user imagine or attempt to perform a particular posture and sequence. Similarly, certain signals (such as small motor units detected by independent component analysis in high-density surface EMG of the trapezius muscle) may not, in some cases, have no apparent or conscious control correlation phenomena, and by requesting a specific imagined / attempted movement, the user's activation pattern may be induced as useful for real-time decoding. Generative adversarial networks, random forests, and all the other algorithms already cited can be set up to generate appropriate instruction sets or other timestamped goals and tasks for target movements.

[0051] Feedback-based error minimization recalibration In certain cases, the user can deploy a motor function enhancement device for a functional task as soon as a preliminary mapping or collection of mappings from the input source to the output effector is made. In other cases, there may be significant advantages to repeating a certain calibration stage after the initial calibration to minimize errors. This approach can provide rapid recalibration of the system, as a result the user can practice using the effector, and the technician or provider may use the data collected during these initial practice attempts to identify a new set of discrete or continuous decoder parameters (such as linear filter coefficients). The motor function enhancement system may, in some cases, periodically query the user, technician, or provider regarding whether a calibration or recalibration session should be repeated.

[0052] calibration Calibration refers to the process of adjusting the mapping so that the system registers active input and output options and the user can control various outputs to achieve functional goals. Weighting and transformation of the mapping can occur at the sensor level or by combining all inputs together as a whole. Calibration can take into account previously programmed macros or other pre-programmed routines provided by healthcare providers, family members, users, or automatically generated by the system itself. Furthermore, the calibration process can be initiated and occur automatically by the user (or group of users), technicians, healthcare providers, friends or family, or caregivers designated by the user, or a combination of these options.

[0053] A graphical user interface can be used for calibration procedures to receive input. The system can then map the input to outputs that include meta-modulation outputs to the brain, peripheral nerves, or spine, which may bias the net motion.

[0054] The data processing performed on any or all of the above may include power spectral fast Fourier transform, wavelet convolution, Hilbert transform, Bayesian classifiers, filters (linear, particle, Kalman, or hybrid), hysteresis tracking, and others. The motor function enhancement system can track and timestamp any stage of calibration, training, and functional recovery tasks, as well as any biomarker signature or provider-tracker (presentation of new information, encoding-storage-retrieval, customized occupational cognitive math-language learning vs. testing) and task performance (individual scores, various baselines, previous sessions, and past performance against prescriptive datasets).

[0055] Calibration procedures can be performed with or without the presence of a technician, and with or without the availability of a technician remotely. The screen can display data about the user's health and medical status (to healthcare providers, the user, and designated caregivers). Calibration can involve systematic identification of specific sensors by desired available inputs and their signatures. If a particular sensor requires mechanical adjustment (e.g., conductive contacts with excessive impedance), the system can generate alarms and commands for the corresponding adjustment. In addition to considering time-series and trigger data from various sensors or pre-programmed routines, as well as ongoing annotation by the user, technician, or others, the system can also track other performance features, such as positive deflection 300 milliseconds after stimulus presentation in EEG (P300), peak alpha frequency, user queries for insights, subsequent memory recall, arousal and other assessment metrics via EEG / EMG / EKG, cognitive effort, and others. Furthermore, calibration can involve 2D and 3D animations, video and virtual reality displays, audio, and haptic feedback, and may deploy distinctive content specific to a particular user, such as music, audio, personalization, photographs, and other images. Sensor signal quality, such as conductive sensor impedance and mechanical switch positioning, can be continuously tracked in the background during calibration and throughout use.

[0056] Calibration can be tailored to specific tasks, and the calibration can be "meta" for each task. Therefore, there can be regularly scheduled calibration sessions incorporating diaphragmatic breathing, guided imagery, relaxation exercises, hypnotic instructions, eye movements, and specific types of biofeedback. This type of meta-calibration can both help optimize input-output settings and mapping, and can train the user to optimally deploy the system in diverse settings. Biofeedback can include EEG-based neural feedback to optimize learning of functional tasks by gradually increasing or decreasing arousal levels.

[0057] Calibration can also provide an environment for testing output on a variety of tasks. Basic physical motor tasks (e.g., opening and closing the hand) can be linked to functional tasks (grabbing a cup) and cognitive tasks (speaking words in a new language, e.g., pronouncing the word for "cup" in a new language). A cognitive version of the motor function enhancement system can guide calibration toward teaching new material in mathematics, language, engineering, finance, and other topics.

[0058] Rapid quantitative methods for mapping visual receptive fields can be adapted to other sensor inputs. Bayesian active learning methods, including a utility function for selecting stimuli to minimize the mean posterior variance of sensor inputs, can analyze the relationship between prior parameterization, stimulus selection, and active learning performance. Alternative approaches may include stochastic gradient descent and generalized linear classification schemes. Rapid sequential visual presentations can be deployed for calibration, allowing participants and software to identify and detect optimal mappings.

[0059] The calibration system can guide or "walk" the user through a variety of daily activities (e.g., motor-based activities such as lifting a laundry basket, brushing teeth, picking up a cup and drinking; and / or cognitive-based activities such as filling out a tax return, conducting banking transactions, managing medication).

[0060] Calibration can also use cues to facilitate task completion. Examples of cues may include disappearance cues, olfactory cues, tactile cues, vibratory or electrotactile artificial contexts, vibratory or electrotactile pair associations, spatial intervals, intentionally altered task settings, multimodal multisensory feedback, metaphor education, memory enhancement, method of loci, motor learning, teachback, intentionally alternating between focused, intense practice and relaxation, yoga postures, Qigong postures, permissible mind-wandering intervals, and others.

[0061] Fast, reliable, effortless, and free input search. A motor function enhancement system includes a method for determining the user's optimal free signals. For individuals with motor impairments, non-injured signals exist somewhere in their brain or body that can be used to control devices for restoring movement. For example, if someone has a paralyzed hand, their elbow control may be non-injured, and consequently, elbow motion can be used to trigger a device for restoring hand movement. The system's software algorithms, when deployed, can detect these "FREE" inputs.

[0062] The system can measure residual control signals from the user during a set of calibration tasks. Technicians, healthcare providers, therapists, or clinical engineers can attach sensors (e.g., EMG, accelerometers, gyroscopes) to multiple parts of the human body, and other sensors such as video and audio recordings can also be registered simultaneously. A central virtual reality workspace, which is viewed directly by the user or used as a cue system for therapists and technicians, can demonstrate specific actions and commands. Commands can include actually performing or simply imagining performing a specific movement or set of movements. Using a virtual reality model, a therapist can act out a target movement to demonstrate the movement, or a participant can move their own limbs via powered robotics or stimuli, while being commanded to pretend that they are controlling their own limbs.

[0063] The system can track signals received from various sensors and assign timestamps to them in parallel with registering timestamps for various commands. After sufficient calibration information has been obtained (for example, by having a participant imagine or attempt a variety of functional movements of one or more limbs or the whole body), the technician can trigger the system (or the system can automatically trigger itself) to build a mapping model to identify which inputs or combinations of inputs correlate with the target activity. This can generate a set of preliminary inputs and mappings, and one or more of the commanded activities can be repeated with the preliminary input-motor-output mappings in place. The system can then track queries for accuracy and speed and subjective evaluation, if appropriate. The software can then update the mappings and repeat the cycle itself between the variety of commanded activities and the various mapping options. Assuming a combinatorial large-scale search space of all possible input sources, mappings, and motor outputs, a set of mappings and "Gdynuff" thresholds can be used in parallel to achieve a given functional outcome (e.g., picking up a cup and bringing it to the mouth and drinking from it).

[0064] Morse-sama Macro Morse-like macros can map FREE inputs to alternative motion mechanisms. For example, instead of using an intact shoulder EMG / mechanical switch to drive distal stimulation / powered motion, the system can map one or more FREE inputs to specific actions (e.g., shrugging the shoulder once turns on a light, and shrugging twice turns it off). Morse-like macros can read sequences of intervals as Morse code or specific assignments such as the "shave-and-a-haircut-two-bits" rhythm, allowing any given FREE input to be used rhythmically. The identity and rhythm (e.g., …_.._.) of the FREE input (e.g., EMG from the proximal trapezius of the weak side) can be combined to achieve custom effector outcomes (e.g., mapping the rhythm to opening a specific phone or computer application, or typing a specific word or phrase). This can affect the ease and reliability with which the human brain can generate intervals of specific discrete actions that allow it to significantly speed up the engineering decoding phase of a system.

[0065] Neural Graffiti The system can also implement neurograffiti. When a person imagines, or actually performs, a conversation or movement, specific brain regions are activated. The goal of neurograffiti is to make the daunting task of decoding intended conversation or movement commands from recorded brain activity much easier. Decoding neural activity into imagined or intended conversation or movement is difficult because there are so many possible conversation and language components, and the potential combinations are enormous. Neurograffiti constrains the problem by requiring the user to learn a specific list of imagined language-movement "gestures" and a set of their rules. By assembling these gestures into intended conversation or text, voice activation and keyboard typing can be replaced. Imagined gestures are analogous to imagined typing, imagined shorthand, and imagined gestural language, making them manageable, fast, and real-time, thus constraining the decoding problem (i.e., allowing intended information to be recorded and decoded at the speed of thought).

[0066] There are several ways to tackle decoding neural activity into a desired text or conversation: • Imagine a conversation. You can imagine saying "I wish I could fly" without visualizing the words. Imagine typing text. Throughout using the predicted text, imagine your fingers typing the "I" key, then the "spacebar," then the "w," then the "i" key, etc., on the keyboard to end a word and choose one of the options. Imagine holding a pen or pencil in your hand and writing the words by hand, or using your elbows, shoulders, feet, or even your whole body to draw the shape of each letter (such as carving words into the ice with ice skates). Imagine writing in shorthand or stenographic terminology, indicating each character, sound, ideogram, word, or phrase. Imagine gesturing through the text, letter by letter or word by word (like a gestural language using hand / arm gestures).

[0067] Users can imagine a series of fast neural gestures characterized by quick movements and compact notation. For example, Siri as an imagined right-hand rotation, extending the hand, and quickly beckoning with a finger. Gesture options for writing, letter by letter and word by word, can be stored in a calibrated gesture library.

[0068] Closed-loop trigger-associated stimulation The system may include "signal sources," "trigger rules," and "trigger outcomes." "Signal sources" may include neural activity recorded from the brain, such as scalp EEG, subgalea aponeurotica EEG, intraosseous EEG, epidural EEG, subdural EEG, intracortical LFP, deep EEG, and single-unit recordings; from other physiological sources, such as heart rate, heart rate variability, respiratory rate, electrocutaneous conductance, blood glucose levels, pupil diameter, extraocular view, and electromyography; and from other data about the user, such as limb, head, or body position inferred from an accelerometer, gyroscope, or visuomotor analysis; as well as user input from the user, such as voice, gestures, keyboard entries, mouse clicks, and joystick usage; as well as input from other users, such as doctors, teachers, and caregivers; as well as pre-set input from a computer; and predefined events, such as calendar items, email arrivals, pre-set timers / tasks, recognition of a specific person via a glasses-type camera or ear microphone voice, events, or arrival at a specific GPS coordinate.

[0069] A "trigger rule" can capture one or more of these data streams in real time within a continuous, overlapping or non-overlapping window of a constant or variable period, and a trigger is set (e.g., set from 0 to 1) if a certain set of features in this data matches, such as an oscillatory force feature (e.g., subgalea aponeurotic EEG recording or root mean square of EMG across a particular muscle), passing a predefined threshold (e.g., power in a specific frequency band exceeding a predefined multiple of the standard deviation within a baseline calibration period), which can cause a predefined "trigger outcome," with a specific outcome assigned to a specific trigger rule event. This outcome can include numerous parameters that can vary, such as electrical stimulation at scalp, subgalea aponeurotic, epidural, intraosseous, subdural, deep, and intracortical contacts, muscle FES, and sets of two or more contacts (location, spacing, distance, impedance), and stimulation characteristics (duration, pulse width, pulse polarity, pulse waveform, isolated pulses or pulse trains, train interval relative to train frequency, amplitude, frequency, phase angle). The “trigger outcome” can also be the presentation of a specific stimulus (words or images on a screen, vibrational tactile patterns, sounds emanating from a speaker worn in the ear, for example). For example, the “signal source” could be a continuously recorded subgalea EEG. The “trigger rule” could be an online classifier that recognizes a person’s unique signatures for apathy, poor memory, confusion, and drowsiness, if previously tagged using an initial calibration algorithm (manual, automatic, or hybrid). The “trigger outcome” could be a subgalea stimulus at a specific location, with a specific power and frequency, as well as a specific duration. There may or may not be a record of the command spoken into the person’s earphone speaker, and a log to the cloud or a data buffer on the worn smartphone device or the device itself.

[0070] Priming Labeling Method This three-stage process is designed to enhance the encoding, storage, and retrieval of episodic memories. "Prime" uses transcranial electrical stimulation (from scalp, subgalea aponeurotica, or intracranial electrodes) to "prime" the brain to a state ready for learning at any time. "Prime" may occur in an open loop based on any initiation or psychophysical task state, or may be conditional on the detection of a "trigger signal" indicating "ready to learn." "Prime" is considered a spatially nonspecific, generalized diffuse modal boost. "Commands" are specific data, e.g., text or face-name association or brief description or fact. This can be uttered or displayed. Immediately after the command, there may be a teach-back and "labels," e.g., a vibratory tactile pattern or an olfactory cue. The idea is that the user can intentionally and voluntarily deploy these "labels" to help retrieve learned "command" episodic memory content in the future. This approach can be combined with the motor enhancement techniques described herein.

[0071] Multimodal feedback Primary sensoricortical interference stimulation involves delivering current to two or more conductor contacts so that wave interference targets a specific target without affecting the intermediate position. "Smart" refers to considering ongoing intrinsic (or steady-state induced by fluctuating external stimuli) activity, either locally as recorded directly from the contacts or inferred by low-resolution tomographic inference procedures, so that a smaller amount of total current (as a spatiotemporal integral, i.e., amplitude-duration inversely proportional) is required to achieve a given targeted electrical stimulation effect. The idea is that electrical stimulation (scalp, subgalea aponeurotica, intraosseous, epidural, subdural, intracranial, intracerebral, intravascular, transcutaneous peripheral nerve, neuromuscular FES) can induce specific visual / auditory / tactile / proprioceptive or other novel sensations, either alone or in combination, by delivering stimulation at specific electrodes in conjunction with ongoing vibrational activity. For example, if a person uses the Freedom System interface to take notes, junctions in the primary motor cortex, premotor cortex, and possibly the subgalea aponeurotica above the parietal planning cortex, or junctions in specific peripheral nerves, decode neural activity associated with predefined / pre-calibrated neural gestures ("neural graffiti") so that the user can get feedback and see what the person was gesturing, and another set of junctions generate electrical currents (including those in the mastoid process and shoulder and other positional reference areas) to induce visual phosphenes constructed to form, for example, letters and words. This can also be done in conjunction with ongoing activity in the visual cortex, e.g., a specific phase of intrinsic oscillations at a particular frequency, e.g., the timing peak of an interfering stimulus to hit 90° at a 10Hz oscillation. Optimal frequency / phase timing can also be explored in calibration sessions and updated daily, weekly, or continuously as a moving average or other ongoing self-calibration rule.If endogenous activity is insufficient, the system can also intentionally add input through external / quasi-external stimuli, such as flashing lights (e.g., staring at a cell phone screen flashing at 8.3 Hz, then the stimulus interference rides on that 8.3 Hz wave, resulting in the appearance of an image on the screen that can only be seen by the brain but is not physically on the screen; or similarly stimulating the median or sural peripheral nerve at 8.3 Hz). However, in the case of a completely self-contained system, steady-state stimuli can originate from implanted tactile devices such as vibrating motors or speaker diaphragm membranes, or buzzer bracelets, buzzer rings, or buzzer earrings. The system can combine intracranial and extracranial electrical stimulation for this cumulative "load." That is, low-amplitude electrical stimulation applied through the skin at one or more locations can bias the activation of the sensory cortex, even if not consciously perceived, and consequently, lower currents applied directly to that cortex from subgalea aponeurotic electrodes were needed to induce perception. If the computer knows the pattern, it can also have variable frequencies (random or sweep) in addition to a constant steady-state frequency, which can add interference stimuli to specific phases of that known input variation pattern.

[0072] "Phosphene" refers to the visual perception of dots of light or flashes in the absence of visual stimuli. Electrical stimulation of the visual cortex of the brain causes a person to perceive phosphine. This type of "induced hallucination" can also be done for other types of sensations, such as auditory and tactile stimuli. Electrical stimulation of the somatosensory cortex may, for example, cause the sensation of water dripping on the skin, and electrical stimulation of the auditory cortex may cause the perception of distinct timbres. Electrical stimulation of the vestibular nerve (such as an electric current passing through the mastoid process) can induce a vivid sensation that the world is tilting or rotating. Similarly, stimulation of the trigeminal nerve (of the mouth or face) can cause a distortion of the sense of balance. Stimulation of higher cortical areas, such as the ventral or lateral parts of the temporal lobe, can induce much more elaborate sensations, such as the re-experience of specific shapes, faces, words, or even specific memorized events. From now on, phosphene can be considered as indicating any induced experience, whether it be a flash of light or light, a timbre or tactile impression or a sense of balance.

[0073] The methods and systems described herein can be implemented to induce specific combinations of phosphenes as a means of feedback from a computer. That is, the system attempts to transmit images, sounds, touch, limb position, and other perceptions in a manner similar to viewing an image display / computer screen, listening to audio, or using a haptic feedback device. This approach can be considered the brain's equivalent of what retinal, cochlear, and auditory brainstem implants do (primarily via neocortical stimulation, but with "temporal interference" and "short cross-pulse" techniques enabling sensory thalamic / LGN / MGN / VP and other subcortical activations). It should be noted that induced perceptual representations, such as those from a cochlear implant, are part of a closed-loop system that the individual learns to interpret, and therefore may not require the same spatiotemporal resolution as might be necessary. Therefore, phosphene-based visual and auditory feedback does not necessarily need to produce vivid 3D images or sounds; in fact, producing overly vivid perceptual representations can be counterproductive, as it may interfere with the ongoing experience of real, existing visual, auditory, and other sensory inputs. Instead, induced phosphene-feedback systems are intended to be gentler, less disruptive overlays that convey enough information to make the closed-loop system useful, for example, by informing the user that they have entered data successfully or by prompting them to repeat the text they just entered.

[0074] The information transmitted by phosphene may include the following: • Spatial localization cues, such as map paths overlaid on a scene, or the location of building frameworks and internal wiring / piping, or diagnostic information overlaid on a patient a doctor is examining, fatigue fracture of a material a technician is holding, or radiation / X-ray / infrared data of the sky an astronomer is observing—a kind of "hallucinatory" or "augmented reality" that allows you to see things that would be invisible to the naked eye / ear / hands. • The "visual" text of language, the "auditory" conversation, and the "tactile" impression. Thus, if there is text on a computer screen or smartphone, it will be delivered via this phosphene-perceptual representation system rather than being visually displayed on the screen or played as audio. This can include emojis, hieroglyphs, icons, gestural language gestures, ideograms / idiophones similar to Chinese characters, i.e., certain visual / auditory / spatial patterns that users can unpack and interpret more compactly both word by word and at both grammatical and narrative levels, the symbol "&" for meaning "and", and notational arc shapes for grouping phrases, or indentation for paragraphs. • Images and sounds, touch, taste, smell, balance • In the case of somatosensory phosphenes, it is important to note that the same electrode array that can be used to deliver current to the somatosensory cortex and ventroposterior thalamus can also deliver current to the scalp nerves, and these can be additionally used to induce sensory representations.

[0075] Exercise strengthening A key innovation of the Freedom system is its ability to combine output effectors in diverse ways. Therefore, a given output effector may be used to achieve specific functional movements in reality, or to enhance motivation, plasticity, and arousal, allowing users to learn more quickly and effectively how to deploy the Freedom system or its components for specific tasks.

[0076] Conductive contacts placed in the scalp, earlobe (passing current through the auricular branch of the vagus nerve), around the spinal cord (implanted or percutaneous epidural spinal root stimulation along the vertebrae, either on the skin of the back), peripherally (such as the median nerve at the wrist or the sural nerve at the ankle), or implanted (such as subgalea aponeurotica, intraosseous, subdural, intracortical, or deep locations) can be used to selectively activate the brainstem, sensory pathways, or the motor cortex itself. The timing of anode (positive, excitatory) stimulation can be adjusted to coincide with commands to imagine, attempt, or perform specific actions (e.g., wrist flexion or picking up a pen), and then subsequently while freely performing everyday functions; there is compelling data that anode stimulation to the primary motor cortex enhances functional movement after stroke, and this is expected to apply to other neurological conditions and other anatomical targets (cerebellum, including the premotor cortex, supplemental motor cortex, ventrolateral thalamus, and deep cerebellar nuclei). Unlike tests in which electrical stimulation is continuously applied during physiotherapy, the Freedom system can trigger electrical stimulation to one or more sites (such as the scalp, brain, or peripheral nerves) at precise times in a given functional movement or task, for example, when flexion begins, or during the bring-to-mouth phase of bringing an object to the mouth, or at a specific stance in a gait cycle or bicycle wheel position. Similarly, the Freedom system can dynamically adjust stimulation parameters (such as the contact point between source and sink, frequency, and amplitude) in different phases of specific functions and movements.

[0077] Buffer Feedback Real-time decoded neural gestures for inputting data stored in a buffer in a chip implanted in the body or on the body (e.g., behind the ear, or on glasses or the wrist), or in a laptop or phone or other nearby receiver. Feedback may include audio to an in-ear speaker, images on a screen / display, and most importantly, tactile feedback as electrical stimulation patterns via subgalea aponeurotic leads (including interference patterns to the sensory cortex as visual / auditory / tactile "phosphenes", to the visual cortex, to the auditory cortex).

[0078] Enhanced frequency tagging Frequency tagging is defined as a technique in which inputs such as visual (text on a screen, specific images), audio (spoken words, music, timbre, etc.), and tactile (vibration tactile devices) are set to fluctuate at specific frequencies, resulting in them inducing steady-state evoked potentials in the brain. This allows neural and muscle recordings (scalp EEG, subgalea aponeurotica EEG, intraosseous EEG, epidural EEG, subdural EEG, deep EEG, micro-LFP, single-unit recordings; from surface EMG, implanted EMG, etc.) to detect these frequencies (from fast Fourier transform, wavelet convolution, Hilbert transform power and phase, etc.) and infer the presence, timing, and location of the fluctuating stimulus processing. The premise of this electrical stimulation interference method is that electrical stimulation (delivered non-invasively via tDCS / tACS / tSOS from the scalp, remotely via rTMS, or by currents delivered through subgalea aponeurotica, intraosseous, subdural, deep, intracortical contacts; or implanted in peripheral nerves, perisar, epidural, cutaneous EMG, or perimuscular regions) selectively enhances the circuits and synapses involved in frequency-tagged stimuli at the time those stimuli are presented. Electrical stimulation coupling to frequency-tagged stimuli can also be based on simple timing that occurs concurrently (i.e., electrically stimulating the brain when an agitation stimulus is presented), and more effectively, when the precise timing and amplitude characteristics of the electrical stimulation match one or more parameters of the stimulation frequency (e.g., 3.26 Hz tDCS concurrently with a visual image flashing at 3.26 Hz).This can be achieved by rough timing or precise phase locking (e.g., the phase of a 3.26 Hz electrical oscillation matches the phase of an external stimulus 3.26 Hz oscillation, or the relationship between these two precisely cancels out, e.g., using 45, 90, or 180 degrees, or using phase precession sliding phase relationships), and also by using harmonics and other frequency relationships, The frequency-tagging-stimulus-interference algorithm may also use phase-amplitude coupling, so that the frequency of an electrical stimulus may depend on the amplitude of the fluctuations of the external stimulus, or the amplitude of an electrical stimulus may depend on the frequency of the external stimulus. If the inventors define an oscillatory feature as any parameter of an oscillator that can be set or observed, such as the polarity of the oscillatory, the waveform, the anatomical location of the oscillatory (known or inferred, e.g., by LORETA), the direction of the traveling wave, presence / absence, onset / canceling timing, power at a particular frequency, the frequency at which the peak power was, or the phase of the oscillatory bandpassed at any particular frequency, then the inventors can assert a contingent rule that any particular oscillatory feature of an external stimulus (flashing stimulus) or any particular oscillatory feature of an electrical stimulus depends on a particular oscillatory feature of neural activity recorded at one or more locations. For example, when a power increase exceeding a predefined threshold (such as two standard deviations above a baseline recording from that region) is detected for 3-4 Hz activity recorded from two or more contacts, this may trigger a high-frequency electrical stimulus in the same or a different set of contacts, or a different frequency or power of an external stimulus vibration.More precise triggers (for use in random rules) may include measures of inter-contact coherence, phase alignment, mutual information, and phase-amplitude coupling. In addition to externally driven steady-state potentials, the system can be timed so that electrical (or optical) stimuli are aligned with existing intrinsic oscillations in a specific frequency band at a specific phase.

[0079] Cortimo's operating principle Figure 4 shows an exemplary embodiment of a system for promoting motor function (Cortimo software suite) according to the claimed invention. The system shown in Figure 4 can be an example of the motor function promotion system in Figure 2 and can highlight the operating principle of the Cortimo software suite and how some software components are incorporated. Briefly, brain signals are recorded, amplified, pre-filtered, and digitized using a neural signal processor (NSP) 425. Additionally, the central software enables functionalities such as signal processing, data filtering, spike detection, and data storage. The NSP 425 streams data in real time to a dedicated UDP port, and the data is stored in a data storage device 430 (e.g., a buffer) on the UDP port, which the Cortimo suite can access using an API.

[0080] The Cortimo suite can read brain signals from a data buffer at time intervals selected by the user (ranging from 50ms to 100ms). After the data is queried from the buffer, the Cortimo suite executes dedicated brain-computer interface signal processing and decoding algorithms to provide brain-derived control for external applications. The first stage of the algorithm prepares and unifies spike timestamps derived from the row voltage channels and NSPs to obtain a data stream that can be processed further at any time. In this stage, the Cortimo suite uses timestamps derived from high-precision NSP clocks to synchronize the data stream and various software components.

[0081] The Cortimo system is designed to leverage two key brain signaling components: spike counts and local field potentials (LFPs). Following data synchronization, the signals are split into user-defined time windows, processed to reduce noise and artifacts, and then feature extraction is performed. Subsequently, depending on user-selected analysis parameters, the two feature streams and real-time Cortimo data are saved to dedicated files for offline analysis and machine learning training, if necessary.

[0082] Real-time feature extraction is the highest stage of the algorithm, as the extracted features are then fed into a classification model (decoder) to decode the brain's intentions. Therefore, selecting the most prominent features greatly benefits the decoding performance of the brain-computer interface (BCI). The decoded output is then translated into motor control commands and transmitted via an internal UDP communication layer to the VR Arm application and Bluetooth controller. The Cortimo decoding approach is designed to be flexible, fully utilizing available training data and current BCI performance. Traditionally, BCI applications were designed for the optimal implementation of a single, predetermined decoder (e.g., a classification method), despite allowing for parameter retraining and adaptive feedback loops. The Cortimo suite implements different solutions, and in fact, the system provides a set of seven different classification models that can be deployed at any time after a short period of training data collection. In other words, Cortimo allows the operator to identify, select, and test different decoding approaches with just a few interactions with a graphical interface. This design choice addresses one of the main limitations of current BCI systems: the lack of generalizability of BCI system performance across different sessions and subjects. The Cortimo decoder can be quickly trained using an embedded offline tool that can use either raw data files or Cortimo-generated feature binary files to perform a rapid optimization procedure to identify the best features and best-performing classification models for specific datasets used for training. Furthermore, these models and their parameters can be stored in files and easily loaded into the Cortimo suite for real-time use.

[0083] Training / Test Protocol The Cortimo suite provides a complete set of training and testing protocols that leverage the functionality and real-time processing capabilities of the VR Arm application. In addition to providing realistic models of arm kinematics for practice and rehabilitation, patient visual feedback, and motion targets, the VR Arm application collects critical, relevant real-time information that is relayed back to the Cortimo software for file storage. These files provide foundational data such as timing information, data labeling, and arm kinematics, which are used for the offline machine learning methods of the Cortimo classification model. Furthermore, the VR Arm application represents a valuable tool for designing and implementing innovative rehabilitation protocols that can be fully automated and instrumented to ensure optimal training and patient adherence to treatment methods.

[0084] Closing the feedback loop In contrast to the most common BCI systems, whose performance is evaluated using offline tools, the Cortimo suite aims to achieve optimal real-time use and performance. Therefore, system performance that provides precise real-time feedback and the shortest latency is crucial. The Cortimo system is designed to ensure that all relevant feedback information is updated in each computation cycle. This is achieved by utilizing the streaming capabilities of the wearable device and integrating the wearable device backend into the Bluetooth controller software module. This module provides a stable and consistent communication channel between the wearable device and other components of the Cortimo suite, thus creating a reliable closed-loop system. In addition, the Bluetooth controller software module implements control logic for the wearable device motors and proposes two distinct control strategies: moving the motor in a specific direction (flexion / extension) at regular time intervals of 100ms based on the patient's brain intention to move their arm and hand, thereby implementing separate control strategies; and using patient-derived signals to send the motor to a specific position represented by joint angles with a single command, thereby implementing a continuous control strategy.

[0085] Cortimo Matlab Suite The real-time Cortimo Matlab Suite 410 can collect data from Multiport Central software using a proprietary API. The API and Cortimo applications are embedded using the Matlab programming environment. However, in some cases, the API and Cortimo applications can be embedded using other programming environments, such as Python or C++-based environments. The Cortimo suite can also synchronize and integrate neural data and classification outputs with kinematic data derived from wearable device joint motors. This bidirectional communication is implemented using asynchronous UDP localhost datagrams exchanged between Matlab tools and a C++ Bluetooth controller software module (BCSM). After deriving brain-derived arm control, the Cortimo application sends outputs to the Bluetooth controller software, which then provides an interface for controlling the wearable device and simultaneously receiving feedback.

[0086] Bluetooth Controller Software Module (BCSM) The Bluetooth Controller Software Module (BCSM) 415 implements bidirectional communication with Cortimo software using local host asynchronous UDP communication and bidirectional communication via Bluetooth, and the wearable device firmware runs in an embedded circuit mounted on the power brace. More specifically, the Bluetooth Controller Software Module (BCSM) is responsible for the following: • Manage Bluetooth communication with wearable device firmware; • Generate real-time commands for wearable device firmware; • Receive and manage real-time feedback from wearable devices, such as movement position, battery life, EMG readings, and motion limit range; and • Adjust the settings of your wearable device if necessary.

[0087] The interface between the Bluetooth controller software and the wearable device is implemented using a (proprietary) wearable device API developed in the C++ programming language.

[0088] Virtual reality 3D Arm application VR arm 420 is a standalone Windows application that implements realistic and physically precise arm and hand models. This model is developed to mimic the behavior and control of wearable devices, and can be controlled using similar input commands. Furthermore, the VR application allows for real-time visualization of the wearable device's kinematics, thus providing a valuable tool for visual feedback. Another important aspect of the VR application is its ability to implement training and rehabilitation protocols for both wearable and non-wearable device use. The application has a set of GUIs that can be used to configure specific training protocols for automating rehabilitation sessions. For example, an operator can choose to display different types of motion or specific target arm positions to guide the subject during a session. VR arm is also capable of collecting BCI and rehabilitation session data and parameters. This information is relayed back in real-time to the Cortimo Matlab software and used to close the feedback loop between the system and the subject. The VR application uses the local host UDP communication protocol on a dedicated and secure port to exchange real-time data with both the Cortimo Matlab suite and the Bluetooth controller software module.

[0089] Cortimo Suite Graphical User Interface (GUI) This section introduces the main graphical user interface (GUI), which represents a key aspect of the Cortimo software. It is essential to emphasize that the GUI is designed to allow users to easily, intuitively, and with minimal training, manage and run comprehensive Cortimo BCI sessions. The GUI enables complete customization and modification of analyses and protocols without the need for programming skills or code modification. In other words, the code behind the GUI handles all parameters, allowing Cortimo to provide unique user-friendly functionality and flexibility. Users can easily select the most appropriate settings and change them at any time without compromising system behavior or performance. Furthermore, the GUI provides context-driven feedback at each stage to help guide software stakeholders in setting up the optimal BCI session. The Cortimo suite deploys several GUIs, which can be managed simultaneously, allowing control over all BCI analyses and settings. In addition, the Cortimo suite provides patient-specific views that simply display simplified visual feedback for the patient during execution. These patient views differ from the operator GUI, which also displays and modifies all relevant analysis parameters and settings.

[0090] Main GUI The main GUI provides all the necessary controls for managing BCI sessions, connecting main software components, managing protocol training and testing, and displaying all the required information. Figure 5 shows the main Cortimo GUI.

[0091] BCI Analysis Parameter Selection GUI The parameter selection GUI is directly accessible from the main GUI and provides all the necessary functionality to select and manage BCI analysis parameters, feature extraction, and decoding techniques. Furthermore, the parameter selection GUI allows operators to easily train, retrain, and load pre-trained decoding models. This GUI can be used to create new decoding approaches and import them directly into real-time Cortimo operations. The parameter selection GUI is shown in Figure 6.

[0092] FFT Angle GUI The FFT GUI can include a real-time display of acquired signals. The FFT GUI can enable rapid, immediate evaluation of acquired system performance and signal characteristics. The FFT window can also be used for rapid evaluation of Cortimo input filters and their performance. Figure 7 shows the FFT GUI.

[0093] Cortimo VR Arm GUI The Cortimo VR Arm application can include two GUIs displayed on two separate screens: one for the system operator and one for the patient. Figures 8 and 9 show the VR Arm operator GUI.

[0094] Operator View The Operator View compiles all relevant information from the VR application, as well as the training / test and rehabilitation protocols to be administered to the patient. This user-friendly GUI allows for quick setup and management of all delivered protocols.

[0095] Patient View The patient view displays a simplified version of the operator view. This GUI can display visual feedback for the patient during a BCI session. Application controls and additional information can be omitted, allowing the patient to fully concentrate on the movement of either the VR arm or the wearable device.

[0096] System behavior An overview of system behavior allows for both identifying the main system components and their interactions, as well as verifying system output. A more detailed description of the Cortimo architecture follows the high-level overview shown in Figure 10.

[0097] The block diagram in Figure 10 presents the main components required for the Cortimo BCI suite. Brain signals can be collected using the Neuroport acquisition system. After signal preprocessing and digitization, the resulting digital voltage channel and spike count data can be collected by the Cortimo central application via User Datagram Protocol (UDP) connectivity and a dedicated API. The Cortimo suite can be the heart of the BCI system, capable of advanced real-time signal processing and feature extraction. The extracted features can then be fed into a patient-specific decoder that can be quickly retrained using newly available datasets. The decoder can translate the patient's brain signals into user movement intentions and transmit these commands to a wearable device, such as MyoPro, via a Bluetooth Controller Software Module (BCSM). The BCSM manages bidirectional wireless communication between the PC application running on the power brace and the firmware. This communication protocol can be specifically designed for the Cortimo BCI application and utilizes a proprietary API. Finally, the virtual reality application can exchange data with other entities in the Cortimo suite for both system testing and training. VR applications can also include a realistic visual interface between the patient and the BCI operator, providing real-time feedback on arm movements and rehabilitation protocols.

[0098] Figure 11 shows the high-level design of the Cortimo suite and its main components. The three main software modules are further broken down into key logical attributes that define their behavior. Figure 12 shows the mid-level Cortimo application architecture. Furthermore, the three main software components of the Cortimo suite can operate asynchronously to maximize hardware / software performance in runtime. Data communication between different data streams and modules can be performed asynchronously. Thus, the Cortimo application can act as a central point and perform data synchronization. Each system component can generate data and stream it at different rates. This data and their corresponding sampling rates can be stored in dedicated low-level hardware / software circular buffers. Then, at regular time intervals, the Cortimo application can query all low-level buffers, process all available data streams, and store them in dedicated high-level software buffers. Subsequently, the data streams can be unified using NSP absolute timestamps and finally stored in binary data files for offline analysis if necessary. This approach can rely on absolute timestamps derived from the high-precision clock of the neural signal processor.

[0099] Cortimo application Figure 13 shows a class diagram of a Cortimo application. A Cortimo application can read NSP data buffers in pseudo-real-time. This API can include a binary library, cbsdk.dll, Matlab-specific mex files, and cbmex files. The main class can be CortimoBCI_8, which runs code behind the scenes for the main GUI and main Cortimo functionality, using a startTimer class for setting up key parameters, as well as an updateDisplay class that can represent the main application loop triggered at user-defined time intervals (e.g., in the range of 50ms to 100ms). updateDisplay can define the real-time behavior of the software and can perform processing and decoding of brain signals. The Cortimo suite can be the core of the BCI system and may also feature an additional GUI as well as additional code to manage data communication, synchronization, and storage. Another important component of the Cortimo suite can be offline training capabilities for classification models defined within the classifier class. In other words, the TrainingCortimoClassifier_1 and TrainingCortimoClassifier_UsingOnLineFeats classes can perform machine learning training by carrying out all the necessary steps and can generate new decoders and configurations. These classes can implement two important but different training strategies. The former class allows the user to select raw brain data to be used for feature and classification optimization, while the latter allows the user to select a Cortimo binary file containing online extracted features to be used for classification optimization. In other words, the latter approach can utilize already extracted features (in real time), while the former approach allows for the selection and extraction of new features. Figures 14 and 15 show the main classes for functionality of the Cortimo suite during execution.

[0100] Bluetooth controller software module The Bluetooth controller software module can be a standalone application that runs in the background, allowing the Cortimo suite to manage and communicate with wearable devices by directly speaking with the firmware embedded in the device using native command strings defined in the manufacturer API. Figure 16 shows the main class diagram for the Bluetooth controller module.

[0101] The main class for interacting with wearable device firmware is MyomoIO, which can act as a software gateway for bidirectional Bluetooth communication with the device.

[0102] The main application class is MyoDev. This class defines the application's main loop and can perform regular callbacks (control and communication) to all other classes. It is within this class that the Bluetooth controller software module can access the wearable device backend. The Control_Logic class handles the control strategy for the wearable device, receiving input commands from the Cortimo suite and translating them into command strings that the wearable device firmware can interpret and execute. To ensure patient safety and comply with safety mitigation factors required by the FDA, Cortimo communication with the wearable device can implement commands deemed safe by the wearable device API, and the system is designed to ignore incorrect or defective commands and can operate within the scope of usage instructions specified by the wearable device manufacturer. Furthermore, the Control_Logic class can implement different types of motor control strategies and can translate wearable device backend calls into real-time kinematic feedback information that the Cortimo suite can consume.

[0103] The `udp_dataclass` class can implement bidirectional communication with other Cortimo software components. This class can asynchronously execute and leverage multithreaded processes for performance optimization. Figures 17 and 18 show the main classes for managing the behavior of Cortimo software components. Figure 17 shows the main class handling the communication protocol for wearable devices, and Figure 18 shows the main class managing and controlling wearable devices in real time.

[0104] Virtual Reality Arm Application The VR Arm application is a multithreaded application that deploys the capabilities of the latest generation Unity 3D engine for handling virtual arm kinematics and physics in real time. The application consists of several classes that run concurrently and are executed simultaneously at regular time intervals. The main class is shown in Figure 19, and its fields and methods are extended in the following diagrams.

[0105] Figure 20 shows the main classes responsible for handling the graphics, characteristics, behavior, and performance of a VR ARM application. These classes are typically found within a Unity3D engine application, although they may be auxiliary to the execution of the ARM application. These classes form the application framework and can handle runtime events and interrupts.

[0106] Figure 21 shows the main classes that control the behavior and characteristics of the virtual arm and hand in the application. For example, the rotation_script class can perform all operations and create a bidirectional interface between motion commands and virtual arm / hand control. Motion commands can be received from operator input, such as GUI commands or keyboard keys, or from the output of the Cortimo suite motion control layer. After the motion commands are interpreted, they can be applied to a physically precise 3D arm model with a specific rotation / hand movement class. The real-world classes used for VR motion can depend on the operator's selection, protocol, and the type of motion that can be selected at runtime.

[0107] Figure 22 shows the main class that manages training and testing protocols for VR applications. The Target_Script class can implement and manage training and rehabilitation protocols. Operators can choose from several types of protocols using a dedicated GUI, and the Target_Script class is responsible for their execution, timing, and feedback data collection.

[0108] Figure 23 shows a class of other Cortimo software components that are deployed to handle multiple real-time bidirectional UDP communication channels. UDP communication can occur asynchronously, allowing the VR Arm application to run independently of the presence of the wearable device. In other words, the VR application can be standalone software that can be used in conjunction with brain-derived or other control signals for training and testing both system and subject performance. For example, a novel decoding approach or parameter set can be tested without the need to wear (or use) an external power brace. Furthermore, it is worth noting that the VR Arm application is a valuable and unique tool for easily and automatically collecting the underlying training data for the machine learning algorithms on which the Cortimo decoding approach relies.

[0109] Application thread As previously noted, Cortimo can contain three multi-process and multi-threaded main applications. These components can operate asynchronously, maintaining regular, unmanaged communication channels that enable near real-time data exchange. The components can run independently, relying on separate main threads. This partitioned approach allows the Cortimo system to handle execution time issues using a set of warning messages that can achieve two goals. First, the Cortimo system can efficiently handle internal crashes without freezing the system or stopping the BCI experiment session. In case of component malfunction, the system can attempt to isolate and resolve the problem. If additional input from the system operator is required, the system can communicate highly specific instructions using clear, easy-to-understand graphical elements. Second, the main Cortimo system can recover from multi-component problems while ensuring patient safety or data integrity.

[0110] Connection thread The Cortimo system can also include a set of connection threads. These connection threads can ensure real-time asynchronous communication between components with minimal system latency and workload. The connection threads run in the background and provide real-time data exchange performance.

[0111] Physical View A physical view can present the Cortimo system hardware components on which the software modules are executed. This view helps to highlight the physical connectivity between components. Figure 24 shows the physical layer of the Cortimo suite along with the system's main components. Figure 25 shows a block diagram of the Cortimo system, including the software components and inter-component connectivity.

[0112] Use Case View The use case view can show user actions on system components. Figure 26 shows the communication routes between actors and entities involved in a Cortimo BCI session.

[0113] Cortimo Component Communication Protocol As noted in the section above, internal communication between Cortimo components can be based on the UDP protocol. Each Cortimo software component can act as a UDP listener and sender independently of all other software processes, performing data exchange with high time precision and minimal latency. The use of both low-level and high-level data buffers, combined with accurate timestamps and metadata, can guarantee data integrity even in the case of latency caused by hardware or software workload. Furthermore, the use of a multithreaded approach to software design can provide a valuable tool for performance optimization, especially when different components are executed asynchronously.

[0114] To ensure data protection, each software process can establish a dedicated UDP communication port along with other software components, and these connections can be kept open throughout the execution of the code. If the Cortimo suite is shut down, or if the operator chooses to shut down the connection using the GUI, the local port can be closed.

[0115] VR Arm application communication parameters A VR arm application can have different modes: 1) the application can be controlled by commands sent by Cortimo; 2) if the application can be controlled by a Bluetooth controller, the application can be linked to the real-world location of a wearable device; and 3) the application can be used as a training / testing tool. In this case, the VR arm can be synchronized with other signals and applications via the Cortimo suite, which can provide a data stream that can be used to train / test the system. In this operating mode, the VR arm can respond to control commands sent from the Cortimo suite in a complete data stream, or the application can write the latest protocol information to a data buffer at regular time intervals. In this case, the Cortimo suite can read the information from the buffer at variable time intervals and store the data stream in a binary file.

[0116] Cortimo communication parameters The Cortimo software can either send / receive data from external applications or read updated information from a UDP buffer. These different data access modes can be kept separate using different UDP ports. The MATLAB Cortimo suite can also communicate with two applications: Bluetooth controller systems and VR arm applications.

[0117] Wearable device command layer To establish an efficient communication channel between the Cortimo suite and wearable devices (e.g., MyoPro devices) while ensuring patient safety and data protection, a dedicated command protocol layer can be implemented. The Bluetooth controller software module can interpret commands received from other Cortimo applications and forward a subset of validated commands to the wearable device firmware. Furthermore, the wearable device API commands and protocols can be closed to external applications. Thus, the Bluetooth controller software module can provide a validated and secure interface layer between the wearable device firmware and the Cortimo suite.

[0118] In some cases, the internal Cortimo command layer for a wearable device may consist of two ASCII characters embedded in the data packet that can be interpreted by the wearable device backend.

[0119] In addition, during wirelessly controlled brace operation, the communication layer can use a sequence of characters to send real-time commands to the wearable device motors. For example, a first character can determine the desired movement of the hand motor, while a second character can determine the desired movement of the elbow motor.

[0120] Brace motor controller The Bluetooth controller software module may include dedicated classes that translate Cortimo commands into simpler instructions to be sent to the wearable device firmware. Furthermore, the control approach, whether separate or continuous, can determine how these interface classes operate. Specifically, in the case of separate controls, the Control_Logic class can read motion requests from the BCI decoder based on its value and act on the current motor position (received in real time by the interface) by applying separate increments or decrements. In addition, the Control_Logic class can ignore motion requests that fall outside the range of motion of the device; for example, if a brace is already fully bent, requests for further bending can be ignored.

[0121] For continuous trajectory control, the Control_Logic class can receive the BCI decoder output as a desired angular position. This position can be converted into a total range of motion as a motion percentage before being transmitted to the wearable device firmware and actuators.

[0122] Cortimo output motion commands The Cortimo suite can output a verified sequence of commands to be sent to a wearable device. Based on the selected BCI motion control strategy, these commands can be either separate commands or a sequence of commands representing precise joint angle rotations expressed across the total range of motion percentages. The commands can be stored as two doubles in the Cortimo application and sent as two consecutive ASCII characters to the Bluetooth controller software module and VR application. After reception, the two ASCII characters can be converted to the required positions and stored in an array of two integer variables.

[0123] File data structure The Cortimo suite can generate several types of output files. These files can store information about different data streams and system components. They are essential components of offline analysis for system optimization and machine learning algorithm training. The output files can be grouped into three categories: NSP brain signal raw data files, Cortimo binary files, and Cortimo configuration files.

[0124] NSP files can contain all data and information derived from neurophysiological acquisition systems (e.g., Blackrock NSP acquisition system, electrocardiogram system, skin electroreactivity system, etc.). They can be raw data files and can be manipulated and used to rerun, simulate, or perform offline analysis of BCI data. These files can store two types of time series: continuous local field potentials (LFPs) and timestamps of neuron spikes detected in real time by the NSP system.

[0125] Cortimo binary files can store information collected by the Cortimo suite during real-time BCI sessions. Specifically, data from all system components can be integrated using unique timestamps from the NSP and stored in binary files. These binary files can be loaded into offline analysis tools, which provide crucial data for training decoders, evaluating performance, and identifying optimal BCI parameters.

[0126] The Cortimo configuration file can store information and settings for the Cortimo decoder and Cortimo application. Configuration files can be generated for both offline and real-time analysis, and they contain information and parameters for the BCI. These files can be reloaded multiple times, and they can provide important information about different BCI sessions and classification performance.

[0127] Binary file data structure The Cortimo suite can generate proprietary binary data files crucial for collecting all data and information generated during a BCI session. The system can generate different types of files depending on the user-selected options and settings of the BCI application.

[0128] Training data file Cortimo BCI can store training data files for each BCI session.

[0129] LFP Maker File Cortimo BCI can store LFP features extracted in real-time sessions. Furthermore, the system can store settings and parameters related to the extracted features for further processing.

[0130] Spike velocity feature file Cortimo BCI can store spike velocity features extracted in real-time sessions. Furthermore, the system can store settings and parameters related to the extracted features for further processing.

[0131] Cortimo Decoder Settings The Cortimo suite allows for the saving and loading of system and classification settings. These parameters can be stored in dedicated files and loaded using a GUI. The GUI assists the operator in monitoring BCI analysis parameters, extracted features, and implemented classification methods. In addition to providing visual feedback for BCI settings, these files can store the information necessary to run a BCI session. For example, after launching the Cortimo system, one of these files can be loaded to run a complete BCI session end-to-end without any additional user input.

[0132] Decoding Strategy Analysis parameters and strategies can be changed in real time and implemented with few clicks via a dedicated GUI. This enables fast and reliable performance optimization and retraining. This is an innovative approach compared to traditional BCI algorithms where the overall signal processing cascade and classification are hardcoded into the system, requiring modification of software programming tools. The main components of Cortimo's decoding strategy are feature extraction, feature optimization, and classification optimization / training.

[0133] Feature extraction During execution, the Cortimo system can perform feature extraction based on the operator's preferences. Specifically, the operator can choose whether to derive features from LFP, spike velocity, or both. Furthermore, the operator can select the specific NSP channels to be used, the specific time analysis window, and the frequency parameters being extracted.

[0134] Additional feature extraction can be implemented by combining features and spikes from the LFP. This approach can extract a certain number of features from the LFP and combine them with a certain number of features extracted from the current spike rate and the average of the past 10 spike rate bins. This approach can provide feature set updates at a rate chosen for the LFP, but it can also take into account faster data provided by spike counts.

[0135] Feature optimization The Cortimo suite can perform feature extraction at runtime. These features can be directly selected and used by the operator to quickly train a new classifier and run the system.

[0136] Alternatively, the Cortimo suite can run a complete offline optimization routine that works with both raw data files and binary files containing features. This offline optimization toolset can ensure that the system can be consistently retrained as more and more training datasets become available, and thus the optimal features and classifiers can be selected. This optimization procedure loads the raw data files and, if selected by the operator, the training data files, quickly reviews them, and generates a set of performance metrics and configuration files that can be loaded into the Cortimo system for use in execution time.

[0137] Classification Optimization Cortimo can deploy classification models ready to be trained using either extracted online features for rapid training and testing, or raw NSP data files with Cortimo BCI training labels for slower but more precise training and testing. Both options can generate configuration files ready to load into the Cortimo runtime GUI. Furthermore, optimization routines can be run when new datasets become available, thus allowing for the retraining of a suitable and optimal model without the need for coding or modification of the Cortimo application structure. Different classifiers can be easily trained, tested, and reloaded, ensuring that Cortimo applications can quickly switch to and implement different decoders. This level of flexibility allows for rapid system adjustments to specific experimental and environmental conditions. For example, the seven retrainable classification algorithms supported by the Cortimo suite are Linear Discriminant Analysis, Coarse Decision Tree, Quadratic Support Vector Machine, Linear Support Vector Machine, Medium KNN, and Ensembled Bagged Tree.

[0138] These models can be trained using data collected by Cortimo. Specifically, NSP raw data files or binary feature files can be efficiently combined with training / label binary data files collected by the BCI system. By using unique and highly accurate timestamps for automatic labeling of brain-derived signals, the system can group the data into specific epochs, which can then extract features from the epoched signals and associate them with the correct output labels. At the end of this process, a large training dataset is available for optimizing the classification models described above. After these models are trained, a three-fold cross-validation method can be applied to evaluate the performance of each trained model and selected features. Finally, the Cortimo system can generate a report with a summary performance evaluation metric. The operator can choose whether to adopt the best-performing decoder or other decoders suggested based on various considerations. Based on the operator's selection, a suitable configuration file containing all the information necessary to implement the decoder is generated and loaded into the Cortimo GUI. After the new or existing configuration file is loaded, the Cortimo BCI is ready for runtime use.

[0139] Furthermore, the Cortimo suite can implement separate decoders, one for each joint to be controlled (e.g., hand and elbow). These decoders can be different and depend on different extraction features. In other words, two joint commands can be derived using parallel but independent signal processing and decoding algorithms.

[0140] Component integration and safety All Cortimo software components are designed not to alter the safety and operating principles of existing FDA-approved software components. Furthermore, the system is designed to operate only when all components are functioning correctly. In other words, a malfunction of just one built-in component can cause the software real-time control to trigger a series of warnings and messages to the operator to manage the situation.

[0141] Manuscript 1 Stroke is the leading cause of physical disability, with a worldwide prevalence of 42 million people in 2015, affecting over 4 million adults in the United States alone, with 800,000 new cases annually. Stroke results in permanent motor impairment in 80% of cases, and half of stroke survivors require long-term medical care. Brain-computer interface (BCI) technology offers a potential solution to restore functional independence and improve health in those affected. Over the past decade, intracortical BCI technology has continued to advance, with multiple groups demonstrating the safety and efficacy of this approach to derive control signals and restore communication and control. Simultaneously, wearable robotic orthotic technology can benefit patients with weakened limbs. This single-patient pilot clinical trial sought to demonstrate that a commercially available powered arm orthosis could be linked to the cerebral cortex in an adult with the most common form of chronic stroke. A direct pathway from the brain's motor center to the orthosis can also inspire the paralyzed limb to enable useful hand and arm function.

[0142] Several signal sources have been added to provide commands to move the paralyzed limb. Electromyographic (EMG) control of powered braces or functional electrical stimulation (FES) of muscles has proven problematic because the user could not generate sufficient or reliable activity to provide good control signals, or because recorded voluntary muscle activity (intended to generate commands) was counteracted by the stimulator's influence. Contralaterally controlled electrical stimulation (activity from the healthy arm triggering stimulation in the paralyzed arm) is a useful therapeutic intervention to improve function in the weaker limb, but it is unclear how this artificial command source can be generalized to a continuously wearable device that allows independent arm movement. Several groups are exploring scalp EEG, which is closer to the command source, to derive control signals to drive robotic braces, and in some cases FES. While using EEG-derived signals may be promising in rehabilitation therapy, this would likely not be feasible for routine independent function because skin sweat and hair can fluctuate impedance and impair signal quality. Even routine application of a subset of contact points to the same skin area can lead to skin damage and cellulitis. Furthermore, EEG signals are limited to commands that can be easily and reliably derived from available signals. In contrast, intracortical interfaces provide a rich source of high-resolution, multidimensional control signals. This is because they are the origin of such signals in healthy adults, non-human primates, and individuals with spinal cord or brainstem disorders.

[0143] Despite the vast majority of strokes involving cerebral white matter and even direct parenchymal damage, intracortical neuromotor orthoses have not been tested in individuals with strokes above the midbrain. It is unknown whether the motor cortex remains a reliable signaling source in this large population. Proof of concept that brain-computer interfaces based on the implantation of microelectrode arrays into the intact cortex above subcortical strokes may also restore behaviorally useful independent voluntary movement could, in principle, lead to the development of fully implantable medical devices that could reverse the motor deficits caused by stroke.

[0144] method Approval for this trial was granted by the U.S. Food and Drug Administration (exemption from investigational device regulations) and the Thomas Jefferson University Institutional Review Board. Participants described in this report have given permission for photographs, videos, and some of their protected medical information to be published for scientific and educational purposes. Following informed consent and completion of medical and surgical screening procedures, two MultiPort (Blackrock Microsystems, UT) devices, each containing two 8x8 platinum-end microelectrode arrays connected to titanium pedestal connectors, were implanted in the cortex of the precentral gyrus using air intubation technology. Details of the human surgical procedure are in preparation for publication and followed other similar studies. The trial selection criteria are available online (see Clinicaltrials.gov, NCT03913286). This trial was designed so that the implantation phase would last up to three months (Figure 27).

[0145] participants The participant was a right-handed male aged 35-40 years who experienced an acute-onset right hemisphere stroke presenting as high-density left hemiplegia and expressive aphasia. The time of onset was unknown, and due to hypertension at the time of examination, the participant was not a candidate for thrombolysis. CT angiography showed occlusion of the right posterior cerebral artery and severe stenosis of the left posterior cerebral artery proximal to P2. Brain MRI showed acute infarction in the right basal ganglia / coron radiata and right occipital lobe. He was initiated with a 3-week dual antiplatelet therapy, then switched to atorvastatin and 81 mg of aspirin once daily in combination with an antihypertensive drug. He had left hemiplegia, dysphagia, left ipsilateral hemianopsia, and high-density left visual neglect, and was transferred for inpatient rehabilitation. After 3 months, his aphasia and dysphagia had recovered, and he learned to walk independently, although he still had persistent left foot drop. Neuroimaging studies showed evidence of multiple strokes and a history of asymptomatic stroke. The participant had a generally good health status and no known stroke risk factors such as diabetes or smoking. Despite a history of loud snoring, the participant had not been evaluated for obstructive sleep apnea. Transthoracic and transesophageal echocardiography were normal, as was the serial hypercoagulation panel. The participant was adopted, and their biological family history was unknown. The participant was considered to have had an embolic stroke of unknown origin. Although serial electrocardiograms since the stroke were normal, the participant was scheduled for loop recording to investigate possible paroxysmal atrial fibrillation. The participant had a learning disability and was estimated to have had mild cognitive impairment prior to the stroke. Formal screening neurophysiological tests identified neurocognitive problems (full-scale IQ 59). Furthermore, it was concluded that appropriate informed consent was provided, the participant remained fully capable of participating in the trial, and thus met the requirements and needs of the trial. The participant provided both oral and written informed consent to participate in the trial and to share their personal information with the public. The participant was working full-time at the time of the stroke and had not been able to return to work since the stroke.

[0146] Preoperative fMRI Participants underwent MRI using a 3T Philips Ingenia MRI scanner. 1 mm isotropic 3-D T2 FLAIR images were acquired for structural positioning. A single-shot echoplanar gradient echo imaging sequence was used with 80 volumes, repetition time (TR) = 2 s, echo duration (TE) = 25 msec, voxel size = 3 × 3 mm², slice thickness = 3 mm, and axial slices = 37. Participants were asked to visualize the movement of their paralyzed left hand during the MRI. Each exercise trial consisted of a block plan characterized by repetitions of 20 s rest blocks and 20 s activity blocks. This block plan was repeated over four scans totaling 240 s. Visual stimuli included a 20 s video showing a 3D model of the limb at rest, followed by a 20 s video of the limb performing the desired task. The exercise tasks included hand opening / closing or arm extension at the elbow and were either "active" (participant performed or attempted to perform the motion) or "passive" (physician moved the participant's arm by hand). In active tasks, participants were instructed to focus on following movements in a video or on following movements related to a paralyzed limb. Task prioritization was based on the assessment of participants' abilities, specifically the examination of BOLD activations observed during pre-training and scans. Post-processing, including motion correction, smoothing, and generalized linear model estimation, was performed using SPM software (www.fil.ion.ucl.ac.uk / spm) and Nordic brain EX software (NordicNeuroLab, Bergen, Norway). Statistical maps were overlaid on 3D T2 FLAIR images for visualization of activations.

[0147] Artimo System "Cortimo" is the designation provided to the FDA to represent the entire system (Figure 28), which included two percutaneous Multiport (Blackrock Microsystems), each containing two multi-electrode array sensors, cabling, amplifiers, software, and a powered MyoPro brace. Each sensor is an 8x8 array of silicon microelectrodes protruding 1.5 mm from a 3.3x3.3 mm platform. At the time of manufacture, the electrodes had impedances ranging from 70 KOhm to 340 KOhm. The arrays were implanted on the surface of the MI arm / hand cortex guided by preoperative fMRI, and the electrodes were penetrated into the cortex to attempt to record neurons in layer V. The recorded electrical signals are transmitted externally via a Ti percutaneous connector securely fixed to the skull. During recording sessions, cabling attached to the connector routes the signals to an external amplifier and a computer that processes the signals, converting them into different outputs, e.g., servomotor positions on the MyoPro brace or screen positions on the nerve cursor. Currently, this system must be set up and managed by an experienced technician.

[0148] MyoPro Brace MyoPro (Myomo, Inc, Cambridge, MA) is an FDA-approved myoelectric arm orthosis designed to assist a paralyzed arm. The rigid brace incorporates metal contacts mounted on a soft stretch that can be adjusted so that the contacts rest proximal to the biceps and triceps muscles and distal to the wrist flexor and extensor muscles of the paralyzed upper limb. Sensors continuously record the root mean square of underlying muscle activity. A threshold is manually set so that a signal exceeding it triggers one of the MyoPro motors. Since the participant had residual elbow flexion and extension strength, the elbow motors were set so that biceps activation triggered elbow flexion and triceps activation triggered elbow extension. MyoPro was configured to use either electromyographic control or BCI-based control to open the hand. Since the participant was unable to voluntarily extend the wrist or spread the fingers, the electromyographic mode was set so that the default state was the open hand and that it could only be closed by activating sufficient wrist flexor activity.

[0149] Session Record Study sessions were scheduled five days a week in temporary accommodation adjacent to the hospital, provided to participants. Sessions could be canceled or terminated early at the participant's request. Sessions began simultaneously with nerve recording and spike identification. Initial sessions included filter construction and structured clinical endpoint (cursor control) trials, while in the final month of the trial, using an "untrained" algorithm, participants proceeded directly to BCI-controlled hand actions after the patient's cables were connected. This was followed by computer task performance, orthotic control, and occupational therapy practice. The electrodes and nerve signals selected immediately before filter construction remained constant for any given session of orthotic control trials.

[0150] Decoder filter construction Units were extracted using an automated thresholding approach on a per-electrode channel basis based on the root mean square multiplier. For each session, linear filters were constructed using single signals and multi-unit data or high-frequency (100–1000Hz) local field potentials derived from multiple channels (20–30), and these real-time multidimensional neural features were converted into one-dimensional or two-dimensional (position or velocity) output signals. Motor activity and motor imagery approaches, including imagining opening and closing a paralyzed hand, passively flexing and extending the elbow, passively opening and closing the hand, and observing the up-and-down movement of a computer cursor displayed on a monitor without any specific commands, were tested for filter construction. Training data for constructing linear filters was collected by having participants fixate on a target cursor moving slowly up and down on the screen for one minute (5 seconds to move from top to bottom or vice versa at a 20° field of view). After constructing this preliminary filter, a new one-minute retraining session was conducted, this time with a manually controlled target cursor accompanied by a neurally controlled predicted cursor by the participant. Using this additional training set, a second filter was constructed, and the filter was then tested in a simple target-grabbing game where the predicted output y-position was discretized into zones such that the animated fairy rises a fixed distance (1 cm) at the top of the screen and descends the same fixed distance at the bottom of the screen.

[0151] BCI orthotic device use Next, separate outputs were used to control hand opening via the MyoPro hand brace motor. The up-and-down mapping on the screen was translated into the closed-and-open hand position. Participants then performed a series of functional tasks, including grasping and then dropping an object, the Action Research Arm Test, and a variation of the Jebsen Taylor Object Transfer Test. These were tested both while the participants were seated and standing.

[0152] "Untrained" Mapping When participants attempted to apply excessive force to the orthosis motor with their remaining finger flexion force, a novel "untrained" approach was developed to calculate spectral power in the high-gamma band (100–500 Hz) using a rolling 1-second baseline of the LFP signal. Specifically, a 1-second continuous LFP voltage was used to calculate the average spectral density estimate in the 100–500 Hz frequency band using non-overlapping frequency bins with a 50 Hz width. Spectral density was calculated using the Matlab periodogram method. A 1-second rolling window with 50% overlap was used to update values ​​every 500 ms. Real-time spectral features derived from the 20 most neurally tuned channels were averaged across the channels to produce a single high-gamma band value every 500 ms software update. Closing of the orthotic hand was 0.5–3V. 2 10V from the idle baseline in the range of / Hz 2 This is triggered by an increase in average spectral power up to a value exceeding / Hz, and real-time values ​​exceeding this threshold cause the hand movement to close.

[0153] Occupational and physical therapy simultaneously As participants were discharged from acute rehabilitation 60 days after their initial stroke, they were enrolled in outpatient physical and occupational therapy. Prior to device implantation, participants completed a 6-week course of occupational therapy screening. After device implantation, participants continued occupational therapy twice a week and physical therapy once a week. Occupational therapy focused on postural training while sitting and walking, while wearing and removing the MyoPro, and while using the MyoPro for functional activities. Scheduled functional electrical stimulation (e.g., pliers grip program; XCite, Restorative Therapies) and vibration therapy were used for spasticity management. Physical therapy exercises included scapular mobilization, range of motion progression, weight-bearing, intensive use of game-related activities to promote left UE volitional control, and aerobic endurance exercises.

[0154] result Participants received intracortical implantation in the fall of 2020 and had the implant removed in January 2021, three months later, according to the planned three-month duration of the trial. Throughout the course of the trial, participants experienced three mild and one serious device-related adverse events, all of which were treated, recovered from, and reported to government regulatory authorities. The serious adverse event was a scalp infection at the left pedestal site, one week prior to the date of device removal, despite a topical antibiotic regimen and regular cleaning. This infection was anticipated and described as a potential risk in the informed consent form and consent interview. The left pedestal site had been problematic since the time of the initial surgery because it was impossible to re-approximate it precisely with a skin flap, leaving the base of the pedestal exposed. This area was protected, granulated, and new skin grew. Participants were asymptomatic and fever-free, and the infection was detected only by careful visual inspection. Prior to device removal, participants were treated with antibiotics twice daily for 7 days. Skin cultures taken from the pedestal site at device removal revealed the presence of pansusceptible Staphylococcus lugdunesis and Staphylococcus capitis, as well as yeast, and appropriate antimicrobial treatment was provided. No growth of organisms was observed in cultures taken from adjacent bone. The only macroscopic evidence of infection at device removal was a small area (approximately 2 cm²) of skin adjacent to the right pedestal. 3 The patient presented with erythema and fragile tissue. The participant was discharged and returned home. The participant remained in the Cortimo trial for ongoing neurosurgical follow-up and surveillance, as well as to track any further performance improvements in the use of MyoPro electromyography in ongoing outpatient occupational therapy.

[0155] Preoperative anatomical and functional neuroimaging Preoperative imaging revealed an older infarct in the adjacent white matter, including the right lenticular nucleus and corona radiata, with a portion of the posterior limb of the internal capsule progressing since the 2019 acute stroke imaging MRI, along with a large older right PCA infarct (Figure 29). In addition, a small area of ​​band-like signaling abnormalities with subcortical white matter and medial aspects of the hand knob area in the right gyrus precentral fossa was identified. This may reflect retrograde neuronal degeneration. In the DTI, there was extensive loss of anisotropy in the area of ​​the right corticospinal tract from the older infarct. The “conceived” left-hand motor paradigm and passive motor paradigm aided the diagnosis with good agreement. Following the challenge of hypercapnia, a bold signal was evident in the gyrus precentral fossa. In the “conceived” left-hand motor paradigm, activation was observed in the expected location along the central sulcus, with lateral aspects of the hand knob area in the gyrus precentral fossa and adjacent portions of the postcentral gyrus (Figure 29C). In the passive left elbow movement paradigm, activation was observed along the central sulcus, showing good agreement with the previously described "imagined" motor task, with slightly greater posterior and superior extension of activation, reflecting the prominent sensory component of this passive movement paradigm. 3D brain models were printed using 3D FLAIR sequencing to enable 3D visualization of the surgical field for more precise preoperative planning (Figure 29D).

[0156] Neurological recording Clear single units were recorded from 87 of the 256 channels (Figure 30). Neural activity correlated with actual and attempted movements in the paralyzed left arm, as well as the intact right arm. Discharge rates of various units appeared to correlate with specific residual actions, including wrist extension, which gradually occurred over the course of the 3-month duration (Figure 31). By taking spike counts recorded every 200 milliseconds in each channel, running them through a leakage integrator, and then summing the outputs of these leakage integrators across all channels, the inventors were able to visualize cumulative cross-array firing rate activity compared to forearm electromyographic activity (Figure 31). Of the 256 electrodes in each session, the inventors identified 40 channels that were ultimately used for neural decoding. These channels were used to extract neural features encoding hand and elbow flexion and extension. Two main hand opening / closing decoding approaches were used: 1) separate two-state classifiers based on the continuous output of one-dimensional linear filters; and 2) an "untrained" threshold crossover approach with rolling baseline normalization.

[0157] Orthotic control Left upper limb scores on the Action Research Arm Test (ARAT) were 0 without an orthosis under electromyographic control, 5 with an orthosis, and 10 with an orthosis under direct brain-controlled control. In one component of the Jebsen-Taylor standardized test for hand function, the goal was to pick up five cans one at a time and move them a few inches forward on a table (typical times were 3.23 seconds for empty soup cans in subtest 6 and 3.30 seconds for filled cans in subtest 7). Because the design of the hand orthosis hindered the ability to grasp the soup cans (e.g., the brace only assists the thumb and the next two fingers), the participant performed a modified version of the test. It took the participant 146 seconds to pick up, move, and release five medicine bottles using electromyographic control, and 95 seconds to perform the same task under BCI control. Another task involved grasping an object (e.g., a stress ball or whiteboard wipe) with the right hand, placing it in the paralyzed left hand, then extending the left arm and lowering it towards the floor to drop the object into a large box. This process was then repeated five times in a row. Participants were seated and performed both of these tasks. In the two pick-up-drop-five trials under electromyography (EMG) control, participants' completion times were 128 and 222 seconds / trial, and in the five trials of the same task under BCI control, the times were 81, 106, 137, and 214 seconds. In addition to measuring the total time to perform these grasp-move-release tasks, the inventors also quantified the time taken to release the object after the hand reached the target position. The release time was faster under BCI control than under EMG control (p=0.04, two-sample t-test).

[0158] Motor outcomes Motor measurements were tracked over time when participants were not connected to a BCI or were wearing a MyoPro orthosis, demonstrating that the implantation procedure did not reduce residual strength in the paralyzed left arm and left leg. In fact, muscle strength increased in the left arm. Sequential neurological tests since the time of stroke showed the absence of voluntary wrist or finger extension on manual muscle tests (0 / 5), while at 2 months from the start of the trial, participants began to consistently exhibit voluntary wrist extension against gravity (3 / 5), and were sometimes able to slightly voluntarily extend their fingers (2 / 5). One month prior to device implantation, the Fugl-Meyer upper limb score was 30 (out of 66) for the left upper limb; this increased to a score of 36 four weeks after implantation of two Multiports and to a score of 38 seven weeks after implantation. Despite participants not receiving botulinum toxin injections or any type of antispasticity medication during the clinical trial, spasticity gradually decreased over time, as reflected in a progressively decreasing number on a series of measurements of the modified Ashworth scale for spasticity, in addition to internal and external rotation of the shoulder, for passive flexion and extension movements of the fingers, wrist, and elbow.

[0159] Consideration This pilot study demonstrated that ensemble single-unit activity maintains activity in the ipsilateral cortex of a lesion site superior to chronic subcortical stroke. To the inventors' knowledge, this is the first report of intracortical recordings in the ipsilateral cortex of a lesion site for upper midbrain stroke. This study demonstrated that single-neuron movement-related activity can be decoded and used to control a powered orthosis that restores functionally useful voluntary upper limb movement. Importantly, this brain-computer interface system can be used simultaneously with residual non-injured movement and movement in the limb that has a gradient from non-injured voluntary movement to the absence of voluntary movement, as is common after stroke. An electromyographic approach based on wrist flexion enabled voluntary hand release, but this approach triggered increased muscle tone, subsequently delaying orthosis use (because movement was counteracting abnormal tension); the BCI control mode essentially avoids this problem, allowing the motor to operate more smoothly and quickly. Electromyography recordings demonstrated that participants continued to engage their wrist flexor muscles during BCI control, while the amplitude decreased from abnormally elevated levels to more normal levels.

[0160] This trial was not intended to restore voluntary motor control in the hemiplegic upper limb in the absence of any device use, but even if it were, the inventors found improved strength and reduced spasticity. This suggests that the implantation of four arrays into the ipsilateral cortex of the lesion site did not exacerbate pre-existing hemiplegia (i.e., it did not worsen weakness in the hand or arm). In fact, hand function improved after the intervention. One possible explanation for the unexpected improvements in voluntary wrist and finger extension is intensive practice. Another, more speculative explanation for the participants' improved forearm function is that daily exercises of ipsilateral cortical activity for BCI-orthosis control promoted plastic drive responses that normalize or compensate for abnormal motor synergies.

[0161] Although the limited number of trials across various tasks reduced the statistical power to compare electromyography control with BCI control, qualitatively, there appeared to be a tendency toward faster control in BCI mode. This may be due to the fact that triggering orthotic action from direct cortical recordings does not activate the abnormal forearm synergy in the same way that electromyography control does. Spasticity may represent abnormal plasticity and a lack of corticoreticular facilitation at the medullary inhibitory center, resulting in reduced inhibition from the dorsal reticulospinal tract in the spinal extension reflex. The medial reticulospinal and vestibulospinal tracts are not in opposition, leading to hyperexcitability of the extension reflex. In electromyography mode, where hand closure is triggered by the activation of residual wrist flexor muscles, this hyperexcitability is inevitably triggered, resulting in the orthotic motor having to "struggle" to open the hand, slowing down the process. In BCI mode, any remaining wrist flexor and extensor muscle activity is involved, however to a lesser degree, resulting in less abnormal tension and allowing the orthotic motor to perform hand actions more easily and quickly.

[0162] This pilot study suggests that usable control signals are present in ipsilateral cortical activity at the lesion site. To be clinically scalable, future devices must be fully implantable to minimize infection risk and allow for mobility. With the advent of fully implantable BCIs (i.e., without percutaneous connectors), a wider range of stroke survivors may also benefit. An option that may even gain broader clinical adoption, as demonstrated in at least one person with chronic stroke, would be to combine direct cortical control with implantable functional electrical stimulation in the paretic arm. Direct cortical-driven peripheral muscle stimulation, when deployed continuously in daily life, may have both rehabilitative and direct functional benefits. Fully implantable brain-computer interfaces may represent an opportunity for medical devices to help stroke patients break their recovery steady state and achieve greater functional independence.

[0163] Equivalents While preferred embodiments of the present invention have been described using specific terminology, it should be understood that such descriptions are for illustrative purposes only and may be modified and altered without departing from the spirit or scope of the following claims. The inventors further require that the scope granted to those claims conforms to the broadest interpretation available under the law in effect on the filing date of this application (and any application for which this application takes priority), and that no change in law (by statutory or judgment) after the priority date of this application may narrow the scope of the attached claims.

[0164] Inclusion by reference The entire contents of all patents, published patent applications, and other references cited herein are expressly incorporated herein by reference.

Claims

1. A method for controlling the operation of a device for assisting the initiation of a patient's movement, comprising a set of nerve sensors and one or more output effectors, A computer having a set of nerve sensors and a nerve signal analyzer electrically connected to one or more output effectors and equipped with a nerve signal processor and a classification model, performs the following: The neural signal processor includes the steps of receiving a set of neural signals from a set of neural sensors, Using the neural signal analyzer, the steps include: extracting a set of features from a set of neural signals; The steps include using the neural signal analyzer to input a set of features into a classification model, The process involves using the neural signal analyzer to determine the user's attempted activity from a classification model, and The steps include using the neural signal analyzer to transmit a set of stimulus signals to one or more output effectors according to the set of activity and neural signals attempted, and Methods that include...

2. The computer further comprises the step of training the classification model using the neural signal analyzer, The training step is, Receiving a set of training nerve signals from the set of nerve sensors, Receiving input indicating actions performed by the trainer, Extracting a set of training features from the set of training neural signals, Mapping the set of training features to directive actions including, The method according to claim 1.

3. The method according to claim 2, wherein the step of training the classification model further includes determining a feature threshold from the mapping, and the user's attempted activity is further determined from features among the set of features that reach the feature threshold.

4. The method according to claim 2, wherein the trainer includes the user, a physical therapist, an occupational therapist, or a combination thereof.

5. The computer performs the following: A step of identifying a set of proportional values ​​between the set of neural signals and the user's attempted activity, A step of generating the set of stimulus signals according to the set of proportional values. The method according to claim 1, further comprising:

6. The method according to claim 5, wherein the step of identifying the set of proportional values ​​is brought about by a multilayer perceptron network, a convolutional neural network, a genetic algorithm, a binary particle swarm optimization process, a generative adversarial network, a polynomial and radial basis kernel support vector machine, a Kalman filter, a generalized linear mixed model, a particle filter, a random forest algorithm, a rotation forest algorithm, or a combination thereof.

7. The method according to claim 1, further comprising the step of the computer performing the step of identifying an activation pattern from the received neural signals, wherein the step of determining the attempted activity follows the identified activation pattern.

8. The method according to claim 1, further comprising the step of modifying the classification model according to the set of features, which is performed by the computer.

9. The method according to claim 1, wherein the set of nerve signals includes scalp EEG, subgalea aponeurotica EEG, intraosseous EEG, epidural EEG, subdural EEG, intracortical LFP, deep EEG, single-unit recording, heart rate, heart rate variability, respiratory rate, electrocutaneous conductance, blood glucose level, pupil diameter, extraoculogram, electromyography, positioning of user body parts, user kinematic and dynamic signals, voice signals, keyboard entries, mouse clicks, joystick use, or a combination thereof.

10. The method according to claim 1, wherein the output effector includes a set of electrical contacts, an electrical prosthesis, a brain-computer interface, or a combination thereof.

11. The system receives a set of brain signals from the user. Digitizing the set of brain signals, Digitized brain signals are stored in a buffer. A neural signal processor configured in such a way; Search the buffer for digitized brain signals, From digitized brain signals, we identify sets of spike counts, local field potentials (LFPs), or combinations thereof. Extract a set of features from the spike count set and LFP, Input the set of features into the classification model, The movement patterns attempted by the user are identified using a classification model. It generates motor control commands according to the movement of the attempted motion, and Transmits motion control commands. A neural signal analyzer configured as follows: It receives a motor control command, and It generates the corresponding movement according to the motor control command. Rehabilitation assistive devices and A system that includes this.

Images