An implantable cerebrospinal interface system

Abstract

This invention relates to an implantable brain-spinal interface system, comprising: a signal acquisition unit, an overall control unit, a device unit, and a neuromodulation stimulation coupling unit. The overall control unit includes at least a control strategy model and a controller. The controller, based on the patient's motor intention (with or without motor intensity) output from the control strategy model as input, outputs stimulation parameters, including channel number, polarity, waveform, frequency, amplitude, pulse width, and combinations thereof. The stimulation output module is configured to output the stimulation parameters configured by the controller. The implantable brain-spinal interface system of this invention controls gait parameters based on the patient's bioelectrical signals, i.e., based on the patient's intention, thus facilitating the remodeling of the brain-spinal cord-muscle neural circuit, particularly the patient's ability to control muscle activity intensity.

Description

Technical Field

[0001] This invention relates to the field of neuromodulation technology for spinal cord injury repair, and in particular to an implantable brain-spinal interface system. Background Technology

[0002] Motor function impairment can be caused by various factors, including spinal cord injury, stroke, and other neurological diseases (such as Parkinson's disease and multiple system atrophy). Among these, motor function impairment caused by spinal cord injury is characterized by bilateral involvement, poor prognosis, and severe impact on quality of life, and is considered a relatively serious form of motor function impairment.

[0003] Spinal cord injury (SCI) is caused by external impact, disease, or congenital defects, resulting in damage to the structure or function of the spinal cord. Its core pathological manifestation is the physical interruption or dysfunction of neural pathways. As a crucial pathway connecting the brain and the peripheral nervous system, the spinal cord is responsible for transmitting motor commands and sensory feedback. When the spinal cord is damaged, motor signals from the brain cannot be transmitted to the motor neurons of the lower limbs, leading to the inability of muscles to contract voluntarily and the loss of the patient's ability to walk. Furthermore, the axons of neurons at the site of injury have extremely weak regenerative capacity, and the local microenvironment (such as glial scar formation and inflammatory responses) further inhibits nerve repair, resulting in chronic, irreversible paralysis. Spinal cord injury is a disease with an extremely high rate of disability worldwide. In addition to causing loss of motor function, SCI is also accompanied by urinary system dysfunction, pressure sores, respiratory complications, and other issues, significantly reducing quality of life and imposing a heavy medical burden on families and society.

[0004] Current treatments for paralysis caused by spinal cord injury mainly include drug therapy, physical rehabilitation training, and neuromodulation techniques, but all have significant limitations:

[0005] Drug treatment, such as neurotrophic factors and anti-inflammatory drugs, can only relieve local inflammation or promote limited axon regeneration, but cannot rebuild functional neural connections, resulting in poor clinical efficacy.

[0006] Physical rehabilitation: It helps patients move passively through mechanical assistive devices (such as exoskeleton robots) or manual assistance training, but it relies on external equipment and cannot restore the ability to control themselves.

[0007] Neuromodulation: Primarily epidural electrical stimulation (EES). This involves implanting electrodes to electrically stimulate the spinal cord, activating motor neurons and inducing muscle contraction. While EES can induce muscle contraction in some patients, it relies on a pre-set stimulation pattern and cannot be controlled by natural will. Because the stimulation is started / stopped manually or externally, rather than through the patient's own will, its effectiveness in restoring motor function has limitations.

[0008] Brain-computer interface (BCI) technology decodes neural signals from the brain and converts them into executable commands, enabling direct control of external devices or the nervous system. This control is directly exercised by the patient's own will. The general process is signal acquisition → signal decoding → command execution. Common command execution components include external devices such as prostheses, robotic arms, and exoskeletons, hence the name brain-computer interface. However, when using external devices to restore a patient's mobility, the limb movements are not controlled by the patient themselves, which raises a problem. After nerve damage, the rehabilitation of voluntary motor function relies on neural remodeling, meaning that the neuronal groups associated with a particular movement need to relearn how to coordinate. This is a target-specific remodeling process. It involves a closed loop: brain command → descending tract → spinal cord → motor neurons → muscles → brain feedback. Remodeling is completed within this closed loop. However, using external devices to drive limb movements often misses this complete loop, such as the spinal cord → motor neurons → muscles links. Therefore, using external devices to drive limb movements is often less effective in promoting neural remodeling and restoring the patient's voluntary motor ability.

[0009] In Chinese patent application CN105792886B, the control algorithm is developed based on the linear relationship between the frequency of electrical stimulation applied in the epidural and subdural spaces and gait-related parameters. For example, there is a linear relationship between the frequency of EES and gait-related parameters—particularly foot height. The control algorithm includes feedback and feedforward components. The control part does not perform analysis and output of the patient's movement intensity; therefore, a feedback component is needed to control the movement intensity. Summary of the Invention

[0010] The purpose of this invention is to propose an implantable brain-spinal interface system that can directly output the patient's movement intensity without having to control the movement intensity through feedback components.

[0011] To achieve the above objectives, the present invention provides an implantable brain-spinal interface system, comprising:

[0012] A signal acquisition unit is placed at a specific location to acquire the patient's bioelectrical signals. The signal acquisition unit is a single-channel or multi-channel first electrode.

[0013] The overall control unit detects the bioelectric signals acquired by the signal acquisition unit and runs the control strategy in real time, and outputs stimulation parameters to the neuromodulation stimulation coupling unit in real time through the device unit; the overall control unit includes at least a control strategy model and a controller, the controller outputs stimulation parameters based on the patient's movement intention and whether or not the movement intensity is included as input according to the control strategy model output; the stimulation parameters include the number of channels, polarity, waveform, frequency, amplitude, pulse width and their combinations.

[0014] The device unit is configured to receive commands output by a controller and generate a target motion result through electrical stimulation and / or a power device; the device unit includes a signal processing module, a computing module, a power supply module, and a communication module, and also includes at least one of a stimulation output module and a power module; the stimulation output module is configured to output stimulation parameters configured by the controller.

[0015] A neuromodulation stimulation coupling unit is configured as a second electrode to release the stimulation parameters.

[0016] Preferably, the first electrode is composed of a human body contact part, a conductive part, and a support part.

[0017] Preferably, the location method for the specific location includes empirical location or inspection location.

[0018] Preferably, when the bioelectric signal originates from the examination location, the specific location is selected at the part where the signal response and target movement are most correlated.

[0019] Preferably, the signal acquisition unit is placed in the brain parenchyma.

[0020] Preferably, the material of the human body contact part is a platinum-iridium alloy, the material of the conductive part is a platinum-iridium alloy or a nickel-cobalt alloy, and the material of the support part is silicone or epoxy resin.

[0021] Preferably, the number of channels is 2-1024.

[0022] Preferably, the number of channels is 8-64.

[0023] Preferably, the calculation time for each input and output of the overall control unit is ≤150ms.

[0024] Preferably, the overall control unit further includes a preprocessor responsible for the initial processing of the raw time-domain data and a decision unit configured to generate classification results.

[0025] Preferably, the movement intention is a specific action, the activity of a specific joint, and / or the activity of a specific muscle.

[0026] Preferably, the overall control unit is responsible for outputting all current motion intentions by one or a group of models, or by multiple or more groups of models.

[0027] Preferably, the control strategy model includes at least a first sub-model and a second sub-model; the first sub-model is configured to focus on spatial features, and the second sub-model is configured to focus on temporal features, wherein the spatial features are reflected in positioning, and the temporal features are reflected in the dynamic evolution of motion.

[0028] Preferably, the first sub-model uses a graph convolutional network model, a convolutional neural network model with spatial attention mechanism, or an adaptive spatial filter model; the second sub-model uses a Transformer network model, a temporal convolutional network model, an LSTM model, or a gated recurrent unit model.

[0029] Preferably, the controller directly outputs stimulation parameters or stimulation strategies, wherein the stimulation strategy is the selection of stimulation parameters; the stimulation parameters and stimulation strategies are configured to correspond to stimulation targets, and the stimulation targets are configured to correspond to motor intentions.

[0030] Preferably, the exercise intensity is configured by at least one of frequency, pulse width, and amplitude, and the control parameter space has a minimum threshold and a maximum threshold. The maximum threshold is the value of one of frequency, pulse width, and amplitude, or the value of at least two of them, or the sum of at least two of them, or the product of at least two of them. The minimum threshold is determined by observing the exercise response or by electromyography.

[0031] Preferably, the controller includes an anti-disturbance module, which is based on a filtering algorithm or a weighted rule.

[0032] Preferably, the controller has a security module.

[0033] Preferably, the overall control unit has a feedback component, which consists of a sensor and a calculator.

[0034] Preferably, the signal processing module, computing module, power supply module, communication module, and stimulation output module are integrated into a single hardware device and implanted subcutaneously in the patient; the power module is located outside the patient's body, and the communication module is communicatively connected to the power module.

[0035] Preferably, the second electrode has ≥4 channels.

[0036] Preferably, the second electrode is positioned around the spinal cord, in the epidural space, or in the lumbosacral segment of the spinal cord.

[0037] Preferably, the second electrode is located in the brain parenchyma, outside the dura mater, under the dura mater, or in a motor-related area of ​​the brain.

[0038] Preferably, when the second electrode is positioned in a motor-related area of ​​the brain, the second electrode is shared with the first electrode.

[0039] Preferably, the second electrode is a sheet electrode, a columnar electrode, a macro electrode, or a micro electrode.

[0040] Based on the above technical solution, the advantages of the present invention are:

[0041] The implantable brain-spinal interface system of the present invention controls gait parameters based on the patient's bioelectric signals, that is, based on the patient's own intentions. Therefore, it is more conducive to the remodeling of the neural circuits of the patient's brain-spinal cord-muscle, especially to the remodeling of the patient's ability to control the intensity of muscle activity.

[0042] The control system of the present invention can directly output the patient's exercise intensity, thus eliminating the need for feedback components to control the exercise intensity, reducing system cost, system complexity, and usage cost, and improving user compliance. Attached Figure Description

[0043] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0044] Figure 1 This is a schematic diagram showing the region most strongly associated with the lower limbs as revealed by a patient's motor fMRI scan.

[0045] Figure 2 A workflow diagram of a preferred embodiment of the overall control unit;

[0046] Figure 3 A diagram showing the accuracy comparison between using the first and second sub-models individually and in combination.

[0047] Figure 4 A schematic diagram comparing the controller output with and without the anti-disturbance module during continuous stepping for the patient;

[0048] Figure 5 A diagram showing the relationship between stimulus amplitude and hip joint angle during the control process;

[0049] Figure 6 A schematic diagram of a stick figure modeling the stimulus amplitude and lower limb posture feedback from sensors during the control process;

[0050] Figure 7 This is a graph showing the stimulation amplitude corresponding to the recovery time and normalized motor threshold of three patients in the clinical pre-experimental data of closed-loop epidural electrical stimulation based on EEG.

[0051] Figure 8 This is a comparison diagram showing the actual muscle strength effects on patients by the implantable brain-spinal interface system of this invention and simple epidural stimulation.

[0052] Figure 9This invention provides a graph showing the electroencephalogram (EEG) signals, true intentions, electromyography (EMG) of important muscles, and the synchronization of movement and time in both lower limbs of patients with ASIA score A and AISA score B who are undergoing treatment and rehabilitation training using the implantable brain-spinal interface system of this invention. Detailed Implementation

[0053] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0054] This invention provides an implantable brain-spinal interface system for the rehabilitation treatment of patients with severely impaired motor function, comprising:

[0055] A signal acquisition unit is placed in a specific location to acquire the patient's bioelectrical signals. The signal acquisition unit is a single-channel or multi-channel first electrode.

[0056] The overall control unit detects the bioelectric signals acquired by the signal acquisition unit and runs the control strategy in real time, and outputs stimulation parameters to the neuromodulation stimulation coupling unit in real time through the device unit. The overall control unit includes at least a control strategy model and a controller. The controller outputs stimulation parameters based on the patient's movement intention and whether or not the movement intensity is included as input, according to the control strategy model. The stimulation parameters include the number of channels, polarity, waveform, frequency, amplitude, pulse width and their combinations.

[0057] The device unit is configured to receive commands output by a controller and generate a target motion result through electrical stimulation and / or a power device; the device unit includes a signal processing module, a calculation module, a power supply module, and a communication module, and also includes at least one of a stimulation output module and a power module; the stimulation output module is configured to output stimulation parameters configured by the controller.

[0058] A neuromodulation stimulation coupling unit is configured as a second electrode to release the stimulation parameters.

[0059] Specifically, the signal acquisition unit can be a single-channel or multi-channel electrode that acquires the patient's bioelectrical signals and connects to the overall control unit to transmit the signals. The signal acquisition unit is placed in a specific location, and there are two positioning methods: empirical positioning and examination positioning.

[0060] In some embodiments, the bioelectrical signals originate from empirically located sites, such as the primary motor cortex, premotor cortex, sensory cortex, or sensorimotor cortex of the brain. Preferably, it can be the primary motor cortex, or a combination of the primary motor cortex and sensory cortex.

[0061] In some embodiments, the bioelectrical signals are derived from sites located by examination, such as motor fMRI or motor magnetoencephalography (MEG).

[0062] In a preferred embodiment, the bioelectrical signal originates from the site where the signal response and target movement are most strongly correlated during examination and localization. For patients with severely impaired motor function, the functional areas responsible for their primary functions may differ from medical experience. For example... Figure 1 As shown, fMRI examination of this patient revealed that the area most strongly associated with the lower limbs was the premotor zone, rather than the primary motor zone, and the association was highlighted.

[0063] Preferably, the signal acquisition component should be placed within the brain parenchyma. Currently, signal acquisition electrodes are divided into electrodes that do not penetrate the dura mater and electrodes that do penetrate the dura mater. EEG signals are emitted by nerve cells in the brain parenchyma. For electrodes that do not penetrate the dura mater, the electrical signal needs to pass through cerebrospinal fluid and the dura mater to reach the electrode, resulting in a relatively small signal amplitude and weak anti-interference ability. Furthermore, the positional accuracy of the electrical signal decreases after conduction. In particular, epidural electrodes that do not penetrate the dura mater require fenestration in the skull, the size of which is related to the electrode size, leading to significant damage. Additionally, when epidural electrodes come into contact with the dura mater, the dura mater is prone to edema and hemorrhage.

[0064] In some embodiments, the signal acquisition part may be an electrode, which is referred to as the first electrode in this invention.

[0065] In some embodiments, the first electrode comprises a human body contact portion, a conductive portion, and a support portion. The human body contact portion may be a metallic material, and in a preferred embodiment, it may be a platinum-iridium alloy. The conductive portion may be a metallic material, and in a preferred embodiment, it may be in the form of a metal wire, such as a platinum-iridium alloy or a nickel-cobalt alloy. The support portion may be a polymer material, and in a preferred embodiment, it may be silicone or epoxy resin.

[0066] The number of channels in the first electrode is flexible, ranging from 2 to 1024. However, increasing the number of channels leads to increased size and decreased stability. Increased electrode size results in increased trauma to the patient. If the number of channels is increased without increasing the size, the parts in contact with the human body and the conductive parts need to be miniaturized, which inevitably leads to decreased stability and makes them more prone to damage and breakage.

[0067] In a preferred embodiment, the number of channels in the signal acquisition unit ranges from 8 to 64. Too few channels contain insufficient information, while too many channels present the aforementioned problems. 8 to 64 channels can achieve a balance between information content, size, and stability.

[0068] Furthermore, the overall control unit detects the bioelectrical signals acquired by the signal acquisition unit, and the bioelectrical signals can be electroencephalograms (EEGs) or electromyograms (EMGs). In a preferred embodiment, the bioelectrical signals can be EEGs.

[0069] The overall control unit includes at least a control strategy model and a controller. The input to the overall control unit is a bioelectrical signal, and the output is stimulation parameters.

[0070] Each input / output operation of the overall control unit should be less than a specific computation time. Excessive computation time results in a noticeable delay in intention-stimulus response, reducing the patient's sense of control over their limbs. In a preferred embodiment, this computation time is ≤150ms. Within this timeframe, the patient's perception of the delay is minimal.

[0071] In some embodiments, the overall control unit may optionally include a preprocessor and a decision-maker.

[0072] In some embodiments, optionally, the control strategy model includes at least a first sub-model and a second sub-model. For example... Figure 2 The diagram shows the workflow of the overall control unit, where the dashed lines represent optional components.

[0073] The preprocessor is responsible for the initial processing of the raw time-domain data. Preprocessing is sometimes necessary because EEG signals are often non-stationary and time-varying, and typically contain a lot of noise, such as power line interference and eye movement artifacts. Therefore, preprocessing is needed to filter out these interferences. In addition, different frequency bands of EEG (such as alpha and beta waves) often correspond to different physiological states. Preprocessing the data to extract frequency band features helps improve model performance. Furthermore, raw EEG data is generally high-dimensional, such as 64 channels × 5000Hz sampling. Directly inputting high-dimensional data into the model can easily lead to the curse of dimensionality, because models often involve vector operations, and the computational load increases exponentially with increasing dimensionality.

[0074] In some embodiments, the preprocessor may perform processing methods such as Fourier transform, fast Fourier transform, wavelet transform, short-time Fourier transform, Hilbert-Huang transform, principal component analysis, blind source separation, and wavelet packet transform. These methods are easily implemented in host computer or embedded software using C or C++.

[0075] The input to the control strategy model is the bioelectric signal, or a preprocessed bioelectric signal. The output of the control strategy model is the patient's motor intention and whether or not it includes motor intensity. Motor intention is a classification result; different motor intentions correspond to different muscle activities and different neural modulation stimuli. Different motor intensities correspond to the degree to which the muscles should be activated under that intention. Generally speaking, the more complex and elaborate the motor intention, the closer it is to the physiological movements of a normal person, while the simpler and more singular the motor intention, the closer it is to the result of the individual activity of a certain muscle. Generally speaking, the stronger the motor intention, the higher the proportion of activated motor neurons (MNs) and the greater the muscle fiber contraction force. The weaker the motor intention, the lower the proportion of activated motor neurons (MNs) and the less muscle fiber contraction force.

[0076] It should be noted that the output of the control strategy model, in addition to classification, can also include intensity values, and the output of intensity values ​​is important. This is because a patient's motor goal is related to a function, such as extending the knee from a relaxed state by 90° of flexion to 45°. Knee extension is the intention of movement, and extending to 45° is the intensity value. The control strategy model outputs intensity values, which, through the strategy of the controller in the overall control unit described in this invention, enable the patient to complete a quantitative motor goal without the aid of other devices or modules. Compared to using a feedback module to update the stimulus parameters output by the controller, this method is more direct and the system is simpler.

[0077] In some embodiments, the movement intention is a specific action, such as stepping with the left leg, stepping with the right leg, or standing.

[0078] In some embodiments, the movement is intended to be the movement of a specific joint, such as hip flexion, hip extension, hip adduction, hip abduction, knee flexion, knee extension, ankle flexion, and ankle extension.

[0079] In some embodiments, the movement is intended to be the activity of specific muscles, such as iliopsoas contraction, quadriceps contraction, biceps femoris contraction, tibialis anterior contraction, soleus contraction, and gastrocnemius contraction.

[0080] In some embodiments, the intensity value of the motion intensity may be associated with the probability of a specific motion intention.

[0081] In some embodiments, the movement intention can be a set of category and intensity. For example, a specific action is combined with the intensity of that action. Taking stepping with the left leg as an example, when there is a step in front, the height of stepping with the left leg needs to be greater than when there is flat ground in front. In a preferred embodiment, the intensity is represented by a normalized value, such as a normalized value of 0-1 or 1-100.

[0082] In some embodiments, the movement intention can be a combination of multiple categories, such as hip flexion and knee flexion. This is because different groups of neurons in different parts of the brain correspond to different limb segments, and walking often involves the coordinated movement of multiple joints.

[0083] Preferably, the overall control unit is responsible for outputting all current motor intentions by one or a group of models, or by multiple or more groups of models. Because the patient's limbs are bilateral, the complete motor intention when performing a certain action is also bilateral, including multiple limb segments. Therefore, the overall control unit also needs to consider this. Thus, the motor intention needs to be bilateral and include at least one limb segment from each side.

[0084] In some embodiments, the control policy model can be a machine learning model, such as a classifier. In other embodiments, the control policy model can be a deep learning model, such as a neural network.

[0085] In some embodiments, the control strategy model comprises different sub-models. Different sub-models focus on different signal characteristics, and their combined use results in higher decoding accuracy. These sub-models can also be referred to as hierarchical structures or other ways that express the same meaning, and all such expressions without departing from the spirit of the invention are included within the scope of protection of this invention.

[0086] In some embodiments, the control strategy model includes at least a first sub-model and a second sub-model, configured to focus on temporal and spatial features, respectively. The way human EEG accomplishes different motor intentions is primarily characterized by spatiotemporal features, meaning it simultaneously incorporates both temporal and spatial characteristics. Spatial features are mainly reflected in localization; for example, different neuronal groups in different parts of the major motor cortex control the movement of different limb segments. Spatial features are mainly reflected in the different signals received on different channels of the signal acquisition section, because different channels correspond to different neuronal groups. Temporal features are mainly reflected in the dynamic evolution of movement; the characteristics of a movement change from its inception to its maintenance to its termination.

[0087] Therefore, constructing a first sub-model and a second sub-model, focusing on spatial and temporal features respectively, can improve the robustness and interpretability of intent decoding. Its advantage lies in decoupling the spatiotemporal characteristics of the original data; the spatial model focuses on the relationships between channels, while the temporal module focuses on dynamic evolution, avoiding feature confusion during mixed learning of single models. Furthermore, it has strong optimizability and scalability, allowing for individual optimization and expansion of specific modules. Additionally, staged training can reduce the risk of overfitting with small sample sizes. Figure 3As shown, for the same data, constructing the first sub-model and the second sub-model, compared with using the first sub-model and the second sub-model alone, improves the performance and stability. This is reflected in the reduced dispersion of accuracy for different patients or different data segments.

[0088] In some embodiments, the first sub-model is configured to primarily focus on spatial features, and can use a convolutional network as shown in the figure, a convolutional neural network with added spatial attention mechanism, or an adaptive spatial filter.

[0089] In some embodiments, the second sub-model is configured to focus primarily on temporal features and may use technologies such as Transformer, Temporal Convolutional Network (TCN), LSTM, or Gated Recurrent Unit (GRU).

[0090] In some embodiments, a preprocessor can be used as part of a first sub-model and / or a second sub-model. A preprocessor that is biased towards spatial resolution or temporal resolution can be used to output its output as the input of the model, together forming the first sub-model and / or the second sub-model. For example, if the first sub-model is configured to focus primarily on spatial features, a combination of wavelet transform and convolutional neural network can be used.

[0091] Preferably, a decision unit can be included after the model. When using a model with strong feature extraction capabilities, a decision unit can be added. When using a model with strong feature extraction capabilities, the model can be configured to primarily extract features, and the decision unit can be configured to generate classification results. The advantages of this approach are: strong modularity, which facilitates the maintenance of the overall control system; enhanced feature visualization and transparency, and improved interpretability.

[0092] In some embodiments, the decision-maker can be a model.

[0093] In some embodiments, to save computational resources, the decision-maker can be a model with lower computational cost, such as SVM, RF, LR, or LDA.

[0094] The input to the controller is the motion intention and whether or not the motion intensity is present, as output by the control strategy model; the output of the controller is the stimulus parameters.

[0095] Stimulation parameters include waveform, channel combination, channel polarity, frequency, pulse width, and amplitude; stimulation parameters are combinations of these.

[0096] In some embodiments, the controller directly outputs the stimulation parameters.

[0097] In some embodiments, stimulation parameters corresponding to different motor intentions are pre-stored as stimulation strategies, whereby the stimulation strategies represent the selection of stimulation parameters. The stimulation parameters and strategies are configured to correspond to a stimulation target. The stimulation target is configured to correspond to the motor intention. For example, given stimulation strategy 1 and stimulation strategy 2, where stimulation strategy 1 drives left leg flexion and stimulation strategy 2 drives right leg flexion, when the motor intention output is "right leg flexion," the stimulation parameter output by the controller should be stimulation strategy 2.

[0098] In some embodiments, motion intensity is configured primarily by at least one of frequency, pulse width, and amplitude, i.e., the motion intensity is correlated with these parameters. This correlation should be within a range, because too low a stimulus is ineffective, while too high a stimulus may cause patient discomfort or lead to dangerously vigorous movements. Specifically, it has a minimum threshold and a maximum threshold, and the range is between the minimum and maximum thresholds.

[0099] In some embodiments, the maximum threshold is the value of one of the stimulation parameters: frequency, pulse width, and amplitude.

[0100] In some embodiments, the maximum threshold is the value of at least two of the stimulation parameters frequency, pulse width, and amplitude.

[0101] In some embodiments, the maximum threshold is the sum of at least two of the stimulation parameters frequency, pulse width, and amplitude.

[0102] In some embodiments, the maximum threshold is the product of at least two of the stimulation parameters frequency, pulse width, and amplitude.

[0103] In some embodiments, a minimum threshold is configured as the effective threshold for movement, meaning that stimulation parameters below this threshold cannot elicit movement. This can be determined by observing the motor response or by electromyography (EMG).

[0104] In some embodiments, the minimum threshold is configured as a threshold for the movement to reach a certain intensity, such as a muscle contracting to a certain length, a joint reaching a certain torque, a limb moving to a certain position, a limb resisting a certain resistance, or an electromyography reaching a certain intensity.

[0105] In some embodiments, this association is linear.

[0106] In some embodiments, this association can be achieved through a model, such as using a deep learning model.

[0107] Preferably, the controller has an anti-disturbance module. Sometimes, transient jumps or disturbances may occur due to interference, noise, or transient neural excitation. Sometimes these are caused by the patient's inattention or distraction. However, in summary, these jumps or disturbances do not reflect the patient's continuous, focused, and genuine intention to move. In such cases, the controller should be able to handle these jumps or disturbances. This improves safety and the continuity of the patient's movements. Furthermore, it reduces the frequency of switching of the device's stimulus output, extending the device's lifespan. Figure 4 As shown, the controller output is displayed with and without the anti-disturbance module, reflecting the patient's different movement intentions during continuous walking. It can be seen that the output is more stable, more robust, and closer to the true value after adding the anti-disturbance module. The horizontal axis represents time.

[0108] In some embodiments, the disturbance rejection module can be based on a filtering algorithm. For example, it can be based on a Bayesian filtered Hidden Markov Model (HMM), assuming that the probability of "maintaining the current state" in the state transition probability matrix is ​​much higher than that of "switching states".

[0109] In some embodiments, the anti-disturbance module can be based on a weighted rule approach, which can have a weighted state that accumulates as the intent of the model output changes. Only when the weighted state has accumulated to a certain point is the current output intent changed. At this point, the weighted state begins to accumulate again.

[0110] In some embodiments, this weighted state is linearly cumulative, with a cumulative amount X. The current output intent is changed only when the number of changed intent outputs accumulates to X; otherwise, it remains unchanged.

[0111] In some embodiments, this weighted state is non-linearly cumulative, having a cumulative amount X. It also has at least coefficients A and B, and at least one point of change N. When the number of changing intention outputs is less than N, the cumulative amount of a single output is A. When the number of changing intention outputs is greater than N, the cumulative amount of a single output is B. The current output intention is changed only when the number of changing intention outputs accumulates to X; otherwise, it remains unchanged.

[0112] Preferably, the controller has a safety module. Sometimes, interference, noise, transient neural excitation, lack of concentration, inattentiveness, or model misjudgment can lead to unsafe stimulus commands. The safety module can prevent this. For example, during a walking task, the controller should not output the intention to flex both the left and right legs simultaneously.

[0113] In some embodiments, the device unit may include external mechanical devices, with or without electrical stimulation. In this case, the controller output is the corresponding joint angle or motor torque. This method of associating joint angles or motor torque with target movements has been widely used in the field of exoskeletons. However, in the early stages of rehabilitation for some patients, or during the use of exoskeletons by some patients with very severe injuries, or because some patients have muscle atrophy, electrical stimulation alone may not be sufficient to complete the movements corresponding to specific motor intentions.

[0114] Optionally, the overall control unit includes a feedback component. This feedback component consists of sensors and a calculator. The calculator and controller can be separate or integrated. Sometimes, the commands output by the controller, after execution by the device unit, may not perfectly align with the motion target. In such cases, the feedback component should correct this. Correction is the core function of the feedback component, intended for further correction, but not mandatory. The target of the correction is the command output by the controller. When executing a specific motion intention, a motion target can be preset, such as a support force reaching a certain number of N, or a certain percentage of body weight; for example, the hip joint reaching a certain angle. If the motion result deviates from this target, the corresponding controller output command is corrected. The calculator is responsible for calculating the deviation value or percentage between the sensor measurements and the motion target.

[0115] In some embodiments, the feedback component sensor may be optical, such as a camera or an infrared motion capture system.

[0116] In some embodiments, the feedback component sensor may be a kinematic sensor, such as an accelerometer, angular velocity meter, gyroscope, or a combination thereof.

[0117] In some embodiments, the feedback component sensor may be an electromyography (EMG) sensor.

[0118] In some embodiments, the correction involves a cycle, which can be N target movements. Preferably, N is between 3 and 20. For patients with severely impaired motor function, even if this system restores some motor function, the results of repeating the same movement multiple times may not be consistent due to nerve damage, muscle atrophy, etc. Therefore, correction should not be based on the result of a single movement.

[0119] In some embodiments, the feedback component may be a PID controller.

[0120] In this technical field, the specific control parameters and objectives for feedback control in this type of application are often limited or lack detailed description. In some embodiments, the feedback component can be based on the following formula:

[0121] u(t+1)=u(t)+μ⋅e(t)⋅x(t); where u(t) is the control parameter; μ is the learning rate, used to control the parameter update speed to avoid being too drastic; and x(t) is the feature vector.

[0122] The present invention can be controlled in the above manner because the control parameters listed in the embodiments are related to the deviation value of the motion target. This is feasible, and this embodiment has been confirmed in a clinical pre-experiment of closed-loop epidural electrical stimulation based on electroencephalography. Figure 5 As shown, the motion target is the unilateral hip joint angle, the control parameter is the amplitude, and the feedback device sensor is an infrared motion capture sensor. Taking this example, by changing the amplitude, the hip joint angle can be changed accordingly to achieve the preset value. Figure 6 The figures shown represent the stimulus amplitude and hip joint angle during the control process, as well as the stick figure modeling of the lower limb posture based on the stimulus amplitude and sensor feedback during the control process.

[0123] In some embodiments, the control parameter is the value of one of the stimulation parameters: frequency, pulse width, and amplitude.

[0124] In some embodiments, the control parameters are the values ​​of at least two of the stimulation parameters frequency, pulse width, and amplitude.

[0125] In some embodiments, the control parameter is the sum of at least two of the stimulation parameters frequency, pulse width, and amplitude.

[0126] In some embodiments, the control parameter is the product of at least two of the stimulation parameters frequency, pulse width, and amplitude.

[0127] In some embodiments, the control parameters include the force or torque of the power unit.

[0128] In some embodiments, x(t) is the deviation or deviation ratio between the motion result and the motion target, calculated by the feedback component calculator.

[0129] Using feedback components to control gait parameters relies on pre-set targets, rather than the patient's own intentions and needs. For example, when controlling stride height using feedback components, a target stride height needs to be set, such as 15cm, 20cm, or other specific values. When the patient walks, sometimes on flat ground and sometimes crossing steps, if the set value is too high, it becomes too strenuous on flat ground, resulting in unnecessary additional power consumption for the device. If the set value is too low, it becomes impossible to cross steps. However, the implantable brain-spinal interface system described in this invention controls gait parameters based on the patient's bioelectrical signals, that is, based on the patient's own intentions, thus avoiding this problem. This is because the patient knows at what time and with what intensity they need to move during exercise.

[0130] Using the patient's bioelectrical signals to control intention and intensity is physiological and consistent with that of a healthy person. Since both intention and intensity are initiated by the patient's own bioelectrical signals, the patient needs to generate different intentions and intensities to control their limbs to produce different types and intensities of activity. In contrast, controlling motor parameters through external feedback devices involves pre-set parameters and is not physiological. This difference leads to the implantable brain-spinal joint interface system of this invention being more conducive to the remodeling of the patient's brain-spinal cord-muscle neural circuits, particularly to the remodeling of the patient's ability to control the intensity of muscle activity. For example, in our clinical pre-experiment with closed-loop EEG epidural stimulation, patient number 001 was able to achieve voluntary control of step height within 3 weeks. Furthermore, in our clinical pre-experiment with closed-loop EEG epidural stimulation, as the recovery time increased, the stimulation amplitude for muscle activation continuously decreased, indicating that the patient's own neural signals increased, thus reducing the required stimulation amplitude.

[0131] It should be noted that the stimulus amplitude is measured in terms of current, so the decrease in the required stimulus amplitude is not due to a change in impedance over time. Figure 7 As shown, the data for three patients in the clinical pre-trial data of closed-loop epidural electrical stimulation (EEG) are presented. The horizontal axis represents the normalized rehabilitation time from day 1 to day 1, and the vertical axis represents the stimulation amplitude corresponding to the normalized motor threshold (other stimulation parameters remain unchanged), with the stimulation amplitude on day 1 of rehabilitation as the initial value. It can be seen that the required stimulation amplitude decreases with rehabilitation time, which demonstrates that the patients' ability to control their lower limb muscles independently is increasing.

[0132] Furthermore, compared to simple epidural stimulation or closed-loop EEG epidural stimulation without intensity control, the closed-loop EEG epidural stimulation protocol of this invention exhibits better muscle response. (See below.) Figure 8 As shown, taking the left vastus lateralis muscle as an example, the diagrams depict the resting state, the attempted movement state without electrical stimulation, the state of closed-loop EEG epidural stimulation without intensity control, and the state of closed-loop EEG epidural stimulation with intensity control. It can be observed that patients with closed-loop EEG epidural stimulation with intensity control exhibit greater actual muscle strength compared to those without intensity control, even with the same stimulation amplitude. Furthermore, compared to epidural stimulation alone, patients with closed-loop EEG epidural stimulation also exhibit greater actual muscle strength, even with the same stimulation amplitude.

[0133] Furthermore, based on preliminary clinical trials of closed-loop epidural stimulation, in patients with ASIA score A and AISA score B, treatment and rehabilitation using the implantable brain-spinal interface system of this invention can restore rhythmic control of lower limb walking function. (See below.) Figure 9As shown, from top to bottom, the patient's electroencephalogram (EEG) signals, true intention, electromyography (EMG) of important muscles, and movement of both lower limbs are displayed, and these contents are synchronized in time.

[0134] Furthermore, the device unit is configured to accept commands output by the controller and generate a target motion result through electrical stimulation and / or a power device.

[0135] Specifically, the device unit has a signal processing module, a computing module, a power supply module, and a communication module, and also includes at least one of a stimulus output module and a power module.

[0136] In a preferred embodiment, the signal processing module, computing module, power supply module, communication module, and stimulation output module are integrated into a single hardware device implanted subcutaneously in the patient. If a power module is included, it is located outside the patient's body, and the communication module is responsible for communication between them.

[0137] The signal processing module is configured to receive the bioelectric analog signal from the signal acquisition section, sample and amplify it, and convert it into a digital signal. This can be achieved using some general-purpose chips and peripheral circuits, such as the ADS 1299. In a preferred embodiment, the noise of the signal processing module is ≤5uV, and the sampling rate is ≥500Hz.

[0138] The computing module is configured to run the overall workflow control section. The computing module can be a computer, a server, or a computing chip and its peripheral circuitry. This can be implemented using digital signal processing (DSP) chips in conjunction with embedded software developed in C or C++.

[0139] In a preferred embodiment, the computation time for the computing module to perform one input-output calculation is ≤150ms. This can be achieved using a digital signal processing chip with strong processing power, low power consumption, and low size, such as the Ezairo 8300.

[0140] The stimulus output module is configured to output stimulus parameters configured by the controller. Stimulus parameter output technology is relatively mature in this field and has been implemented in products such as DBS and SCS. The stimulus output module includes a boost section, a waveform generation section, and a switch array. The stimulus output module can share the switch array with the signal processing module, or it can be separate.

[0141] In a preferred embodiment, the number of stimulation output channels should be ≥16, the frequency should be adjustable from 5-10KHz, the amplitude should be adjustable from 0-30mA, and the pulse width should be adjustable from 20-1000uS.

[0142] The power module can be a pneumatic, hydraulic, or motor-controlled prosthesis, exoskeleton, or mechanical limb.

[0143] The power supply module is configured to supply power to the device. When the device unit includes a stimulation module, preferably, the received power should be greater than 100mW. When the device unit does not include a stimulation module, preferably, the received power should be greater than 10mW.

[0144] Preferably, the power supply module may contain a battery so that it can still operate in the event of an interruption in energy transmission, thus ensuring the safety of the equipment.

[0145] The communication module is configured to send information to the device unit via other devices, or to send data from the device unit to other devices. Preferably, the data transmission bandwidth of the communication module is ≥20Kb / s. This can be achieved by using a communication chip based on BLE, NFC, or a proprietary protocol, such as a Qualcomm BLE chip.

[0146] Furthermore, the stimulation coupling unit can be a second electrode. Preferably, the number of channels in the second electrode should be ≥4.

[0147] In some embodiments, the second electrode is configured to be placed around the human spinal cord.

[0148] In a preferred embodiment, the second electrode is disposed in the epidural space of the human spinal cord.

[0149] In a preferred embodiment, the second electrode is disposed in the lumbosacral segment of the human spinal cord.

[0150] In some embodiments, the second electrode is positioned in the human brain, such as in the brain parenchyma, or outside or under the dura mater.

[0151] In a preferred embodiment, the second electrode is disposed in a motor-related area of ​​the human brain, such as the primary motor area, the motor sensory area, the auxiliary motor area, or the sensory area.

[0152] In some embodiments, when the second electrode is positioned in a motor-related region of the human brain, it can be shared with the first electrode of the signal acquisition unit. Each channel of the second electrode can perform either stimulation or acquisition functions, which can be achieved by a switch array on the PCBA of the device unit.

[0153] In some embodiments, the second electrode is a sheet electrode.

[0154] In some embodiments, the second electrode is a columnar electrode.

[0155] In some embodiments, the second electrode is a macroelectrode, wherein the major and minor axes (diameter for circles, side length for rectangles) of each contact point in contact with the human body are ≥0.5 mm. The fabrication method of the macroelectrode has been repeatedly mentioned in SCS, DBS products and their disclosed prior art, and will not be repeated here.

[0156] In some embodiments, the second electrode is a microelectrode, wherein the major and minor dimensions (diameter for circles, side length for rectangles) of each contact point in contact with the human body are ≤100µm. The microelectrode can be implemented using MEMS processes, such as depositing metal materials on a PI film, etching circuitry using laser / plasma / electrochemical methods, and then encapsulating it with a PI film.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. An implantable brain-spinal interface system, characterized in that, include: A signal acquisition unit is placed at a specific location to acquire the patient's bioelectrical signals. The signal acquisition unit is a single-channel or multi-channel first electrode. The overall control unit detects the bioelectric signals acquired by the signal acquisition unit and runs the control strategy in real time, and outputs stimulation parameters to the neuromodulation stimulation coupling unit in real time through the device unit; the overall control unit includes at least a control strategy model and a controller, the controller outputs stimulation parameters based on the patient's movement intention and whether or not the movement intensity is included as input according to the control strategy model output; the stimulation parameters include the number of channels, polarity, waveform, frequency, amplitude, pulse width and their combinations. The device unit is configured to receive commands output by a controller and generate target motion results through electrical stimulation and / or a power device; the device unit includes a signal processing module, a computing module, a power supply module, and a communication module, and further includes at least one of a stimulation output module and a power module; the stimulation output module is configured to output stimulation parameters configured by the controller. A neuromodulation stimulation coupling unit is configured as a second electrode to release the stimulation parameters.

2. The implantable brain-spinal interface system according to claim 1, characterized in that: The first electrode consists of a human body contact part, a conductive part, and a support part.

3. The implantable brain-spinal interface system according to claim 1, characterized in that: The location methods for the specific location include empirical location or inspection location.

4. The implantable brain-spinal interface system according to claim 3, characterized in that: When the bioelectric signal originates from the examination location, the specific location is selected at the part where the signal response and target movement are most correlated.

5. The implantable brain-spinal interface system according to claim 1, characterized in that: The signal acquisition unit is placed in the brain parenchyma.

6. The implantable brain-spinal interface system according to claim 2, characterized in that: The material of the human body contact part is platinum-iridium alloy, the material of the conductive part is platinum-iridium alloy or nickel-cobalt alloy, and the material of the support part is silicone or epoxy resin.

7. The implantable brain-spinal interface system according to claim 1, characterized in that: The number of channels is 2-1024.

8. The implantable brain-spinal interface system according to claim 1, characterized in that: The number of channels is 8-64.

9. The implantable brain-spinal interface system according to claim 1, characterized in that: The calculation time for each input and output of the overall control unit is ≤150ms.

10. The implantable brain-spinal interface system according to claim 1, characterized in that: The overall control unit also includes a preprocessor responsible for the initial processing of the raw time-domain data and a decision unit configured to generate classification results.

11. The implantable brain-spinal interface system according to claim 1, characterized in that: The movement intention refers to a specific action, the activity of a specific joint, and / or the activity of a specific muscle.

12. The implantable brain-spinal interface system according to claim 1, characterized in that: The overall control unit is responsible for outputting all current motion intentions by one or a group of models, or by multiple or more groups of models.

13. The implantable brain-spinal interface system according to claim 1, characterized in that: The control strategy model includes at least a first sub-model and a second sub-model; the first sub-model is configured to focus on spatial features, and the second sub-model is configured to focus on temporal features, wherein the spatial features are reflected in positioning, and the temporal features are reflected in the dynamic evolution of motion.

14. The implantable brain-spinal interface system according to claim 13, characterized in that: The first sub-model uses a graph convolutional network model, a convolutional neural network model with spatial attention mechanism, or an adaptive spatial filter model; the second sub-model uses a Transformer network model, a temporal convolutional network model, an LSTM model, or a gated recurrent unit model.

15. The implantable brain-spinal interface system according to claim 1, characterized in that: The controller directly outputs stimulation parameters or stimulation strategies, where the stimulation strategy is the selection of stimulation parameters; the stimulation parameters and stimulation strategies are configured to correspond to stimulation targets, and the stimulation targets are configured to correspond to motor intentions.

16. The implantable brain-spinal interface system according to claim 1, characterized in that: The exercise intensity is configured by at least one of frequency, pulse width, and amplitude. The control parameter space has a minimum threshold and a maximum threshold. The maximum threshold is the value of one of frequency, pulse width, and amplitude, or the value of at least two of them, or the sum of at least two of them, or the product of at least two of them. The minimum threshold is determined by observing the exercise response or by electromyography.

17. The implantable brain-spinal interface system according to claim 1, characterized in that: The controller includes an anti-disturbance module, which is based on either a filtering algorithm or a weighted rule.

18. The implantable brain-spinal interface system according to claim 1, characterized in that: The controller has a security module.

19. The implantable brain-spinal interface system according to claim 1, characterized in that: The overall control unit has a feedback component, which consists of a sensor and a calculator.

20. The implantable brain-spinal interface system according to claim 1, characterized in that: The signal processing module, computing module, power supply module, communication module, and stimulation output module are integrated into a single hardware device and implanted subcutaneously in the patient; the power module is located outside the patient's body, and the communication module is connected to the power module.

21. The implantable brain-spinal interface system according to claim 1, characterized in that: The second electrode has ≥4 channels.

22. The implantable brain-spinal interface system according to claim 1, characterized in that: The second electrode is positioned around the spinal cord, in the epidural space, or in the lumbosacral segment of the spinal cord.

23. The implantable brain-spinal interface system according to claim 1, characterized in that: The second electrode is positioned in the brain parenchyma, outside the dura mater, under the dura mater, or in a motor-related area of ​​the brain.

24. The implantable brain-spinal interface system according to claim 23, characterized in that: When the second electrode is positioned in a motor-related area of ​​the brain, the second electrode is shared with the first electrode.

25. The implantable brain-spinal interface system according to claim 1, characterized in that: The second electrode is a sheet electrode, a columnar electrode, a macro electrode, or a micro electrode.

Images