Self-adaptive neuromuscular regulation and control device based on myoelectricity feedback

Abstract

The invention discloses a self-adaptive neuromuscular regulation and control device based on myoelectricity feedback. The self-adaptive neuromuscular regulation and control device comprises a myoelectricity sensor attached to the skin surface of a target muscle group. The establishing module is used for establishing an electromyographic signal matrix and respectively establishing a space weight matrix for describing a muscle cooperative activation mode and an activation matrix for describing muscle time sequence activation intensity; and iterating the space weight matrix and the activation matrix to obtain the space weight matrix and the activation matrix after iteration. The time sequence compensation module is used for extracting a target muscle collaborative activation curve from the iterated activation matrix; matching the activation curve with a health template, and determining an activation peak point according to a matching result; and according to the activation peak point, determining whether an activation time difference anomaly exists, and when the activation time difference anomaly exists, performing electrical stimulation compensation based on a time sequence according to the activation time difference anomaly parameters in the next period. The electrical stimulation can be effectively corrected according to the effect, and the neuromuscular regulation and control effect is further improved.

Description

Technical Field

[0001] The present invention relates to the field of myoelectric technology, and in particular to an adaptive neuromuscular control device based on myoelectric feedback. Background Art

[0002] In recent years, with the rapid development of biomedical engineering, electrical stimulation technology has been widely used in clinical rehabilitation therapy. However, current electrical stimulation devices present numerous challenges in treatment. For one thing, most existing devices lack self-adjustment capabilities, and most utilize fixed stimulation parameters, making it difficult to accommodate individual patient differences. Different combinations of parameters such as current intensity, frequency, and waveform can lead to significant variations in therapeutic effects. Furthermore, patient differences in muscle atrophy, fatigue resistance, and sensitivity to electrical stimulation also significantly impact the stability of treatment outcomes, often resulting in insufficient or excessive stimulation. While some electrical stimulation devices, such as functional electrical stimulators, offer self-adjustment capabilities, most only adjust parameters based on a single phase angle or gait phase, with fewer devices utilizing the target muscle's electromyographic signal for stimulation adjustment. Because the reference signal source and the target muscle's response are not directly aligned, this can lead to incorrect stimulation parameters or delayed feedback.

[0003] In the prior art, the electrical stimulation signal at the next moment can be adaptively adjusted based on the response of the human muscles to the electrical stimulation signal at one moment, thereby realizing supervisory control of the electrical stimulation signal. In the process of realizing the present invention, the inventors discovered the following technical problems: Since the closed-loop control is performed based on the electromyographic signal and the motion signal, the electromyographic signal and the motion signal actually contain individualized characteristic information. Although the prior art takes into account the advantages of muscle synergy, it does not take into account the changes in muscle synergy with electrical stimulation and the impact on muscle fatigue, making the electrical stimulation effect not accurate and safe, thereby affecting the effectiveness of electrical stimulation. Summary of the Invention

[0004] The embodiment of the present invention provides an adaptive neuromuscular control device based on electromyographic feedback to solve the technical problem in the prior art that neuromuscular control cannot be effectively corrected according to real-time effects.

[0005] In view of this, an embodiment of the present invention provides an adaptive neuromuscular control device based on electromyographic feedback, comprising:

[0006] Myoelectric sensors are attached to the skin surface of the target muscle group and are used to collect multi-channel myoelectric signals of the target muscle group;

[0007] An establishment module is used to establish an electromyographic signal matrix based on the collected multi-channel electromyographic signals, and respectively establish a spatial weight matrix for describing the muscle synergistic activation pattern and an activation matrix for describing the muscle temporal activation intensity; iterate the spatial weight matrix and the activation matrix to obtain an iterated spatial weight matrix and activation matrix, so that the difference between the product of the spatial weight matrix and the activation matrix and the electromyographic signal matrix is ​​less than a preset difference threshold;

[0008] The timing compensation module is used to extract the activation curve of the target muscle synergy from the iterated activation matrix; match the activation curve with a preset healthy template and determine the activation peak point based on the matching result; determine whether there is an activation timing anomaly based on the matching point corresponding to the activation peak point; if there is an activation timing anomaly, perform timing compensation in the next cycle based on the activation timing anomaly parameters. Further, the timing compensation module includes:

[0009] A fusion unit is used to determine the rising edge point and the falling edge point using the activation peak point, calculate the delay deviation of the activation peak point, the rising edge point, and the falling edge point relative to the healthy template, and fuse the delay deviation results to form a standard delay deviation;

[0010] Determine whether to activate the advanced state or the delayed state by using the standard delay deviation;

[0011] Accordingly, in the activation hysteresis state, the stimulation time is calculated based on the standard delay deviation in the next cycle, and the electrical stimulation is triggered at the stimulation time, t stim_new =t event -min(|Δt|,30ms)+α, α=20ms safety margin

[0012] In the activation advance state, the stimulation time is calculated based on the standard delay deviation, and the electrical stimulation is triggered at the stimulation time, t stim_new =t event +|Δt|.

[0013] Furthermore, the timing compensation module further includes:

[0014] The template updating unit is used to modify the original healthy template using the real-time activation matrix to form a new healthy template when the standard delay deviation is greater than the preset delay deviation threshold within a preset number of consecutive cycles.

[0015] Furthermore, the device further comprises:

[0016] Six-axis IMU sensor, attached to the key parts of the lower limb movement, is used to collect three-axis acceleration and three-axis angular velocity signals;

[0017] The gait stage determination module is used to determine the current gait stage based on the three-axis acceleration and three-axis angular velocity signals.

[0018] Furthermore, the device further comprises:

[0019] a determination module, configured to determine maximum and minimum muscle activation levels according to muscle activation conditions during gait phases, and determine a reasonable interval of muscle activation based on the maximum and minimum muscle activation levels; and

[0020] Determine the maximum stimulation amplitude and stimulation pulse width without causing pain;

[0021] The stimulation signal control calculation module is used to adjust the amplitude and pulse width of the electrical stimulation signal according to the position of the muscle contraction force in a reasonable range and the fatigue degree of the muscle.

[0022] Furthermore, the device further comprises:

[0023] A real-time feature calculation module is used to calculate the time domain feature root mean square value, the frequency domain feature average power frequency and the fatigue feature value according to the real-time electromyographic signal;

[0024] Accordingly, the stimulation signal control calculation module includes:

[0025] The fatigue value activation calculation unit is used to map the fatigue characteristic value to the proportional parameters of the amplitude and pulse width using a sign function.

[0026] Furthermore, the fatigue value activation calculation unit is implemented in the following manner: The scale parameter is Accordingly, the stimulation signal control calculation module includes:

[0027] The amplitude control calculation unit is implemented in the following way:

[0028]

[0029] The pulse width control calculation unit is implemented in the following way:

[0030]

[0031] in,

[0032] α is the suppression parameter.

[0033] Furthermore, the real-time feature calculation module includes:

[0034] Fatigue characteristic value calculation unit, including:

[0035] The myoelectric sample entropy value calculation subunit is used to calculate the myoelectric sample entropy value using the myoelectric signal;

[0036] The electromyographic integral value calculation subunit is used to calculate the electromyographic integral value using the electromyographic signal. Furthermore, the device further includes:

[0037] Aligning the spatial weight matrix with the healthy template, calculating the similarity between the spatial weight matrix and the healthy template, determining whether there is abnormal synergy based on the similarity, and suspending electrical stimulation when the synergy similarity is lower than a preset similarity threshold;

[0038] In the case of abnormal coordination, the muscle corresponding to the weight abnormality is determined by comparing the columns of the spatial weight matrix with the corresponding columns of the healthy template;

[0039] When the muscle weight is too high, the stimulation adjustment parameter is calculated according to the difference between the muscle corresponding column in the spatial weight matrix and the muscle corresponding column in the healthy template, and the muscle is electrically stimulated according to the stimulation adjustment parameter at the next stimulation moment;

[0040] When the muscle weight is low, the stimulation adjustment parameter is calculated according to the muscle corresponding column in the health template, and the muscle is electrically stimulated according to the stimulation adjustment parameter at the next stimulation moment.

[0041] The embodiment of the present invention provides an adaptive neuromuscular control device based on electromyographic feedback. The device is attached to the skin surface of the target muscle group through an electromyographic sensor and is used to collect multi-channel electromyographic signals of the target muscle group. A module is used to establish an electromyographic signal matrix based on the collected multi-channel electromyographic signals, and respectively establish a spatial weight matrix for describing the muscle synergistic activation pattern and an activation matrix for describing the muscle temporal activation intensity. The spatial weight matrix and the activation matrix are iterated to obtain the iterated spatial weight matrix and activation matrix, so that the difference between the product of the spatial weight matrix and the activation matrix and the electromyographic signal matrix is ​​less than a preset difference threshold. A timing compensation module is used to extract the activation curve of the target muscle synergy from the iterated activation matrix. The activation curve is matched with a preset healthy template and the activation peak point is determined based on the matching result. The presence of activation timing anomaly is determined based on the matching point corresponding to the activation peak point. If activation timing anomaly exists, timing compensation is performed in the next cycle based on the activation timing anomaly parameter. By decomposing the electromyographic signal matrix into a spatial weight matrix and an activation matrix, and using iteration, the activation matrix can reflect the temporal activation intensity. Based on the matching results of the iterative activation matrix and the healthy template, the system searches for peak points and determines whether there are activation timing anomalies based on these peak points. If there are any, timing compensation is performed in the next cycle. This can address the misalignment of muscle activation timing caused by electrical stimulation of muscle coordination and improve the synchronization between electrical stimulation and gait phase. This allows for effective correction of electrical stimulation based on the desired effect, further enhancing the effectiveness of electrical stimulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0043] Figure 1 Schematic diagram of the hardware structure of the adaptive neuromuscular control device based on myoelectric feedback provided by an embodiment of the present invention;

[0044] Figure 2 Schematic diagram of the electrical stimulation process of the adaptive neuromuscular control device based on electromyographic feedback provided by an embodiment of the present invention;

[0045] Figure 3 This is a system structure diagram of a gated recurrent network in an adaptive neuromuscular control device based on electromyographic feedback provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0047] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0048] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0049] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0050] Figure 1 Schematic diagram of the hardware structure of the adaptive neuromuscular control device based on myoelectric feedback provided by an embodiment of the present invention; Figure 2 Schematic diagram of the electrical stimulation process of the adaptive neuromuscular control device based on myoelectric feedback provided by an embodiment of the present invention. Figure 1 and Figure 2 The adaptive neuromuscular control device based on electromyographic feedback comprises:

[0051] Myoelectric sensors are attached to the skin surface of the target muscle group and are used to collect multi-channel myoelectric signals of the target muscle group;

[0052] An establishment module is used to establish an electromyographic signal matrix based on the collected multi-channel electromyographic signals, and respectively establish a spatial weight matrix for describing the muscle synergistic activation pattern and an activation matrix for describing the muscle temporal activation intensity; iterate the spatial weight matrix and the activation matrix to obtain an iterated spatial weight matrix and activation matrix, so that the difference between the product of the spatial weight matrix and the activation matrix and the electromyographic signal matrix is ​​less than a preset difference threshold;

[0053] The timing compensation module is used to extract the activation curve of the target muscle synergy from the iterated activation matrix; match the activation curve with a preset healthy template and determine the activation peak point based on the matching result; determine whether there is an activation timing anomaly based on the matching point corresponding to the activation peak point, and if there is an activation timing anomaly, perform timing compensation in the next cycle according to the activation timing anomaly parameters.

[0054] In this embodiment, multiple electromyographic sensors can be used to collect electromyographic signals, and then a multi-stage amplification circuit is used, combined with an adaptive power frequency notch filter and bandpass filtering, to significantly improve the signal-to-noise ratio and achieve high-precision signal acquisition. In addition, the gain can be dynamically adjusted to avoid signal saturation or distortion. Each channel corresponds to an electromyographic sensor, and multi-channel electromyographic signals are collected. For example, a Myoware electromyographic sensor can be used, which supports dual-lead differential input and a bandwidth range of 10 to 500 Hz. It is attached to the skin surface of the target muscle group to collect multi-channel raw electromyographic signals of the target muscle group. The collected raw electromyographic signal is first amplified, then de-powered, and then bandpass filtered. Finally, the electromyographic signal is rectified and operationally amplified to obtain the envelope of the electromyographic signal. Furthermore, a two-stage amplifier circuit is used to amplify the signal, and its gain is adjustable to amplify the uV level signal into the mV level electromyographic signal. A 50Hz notch circuit is selected to filter out the power frequency interference in the electromyographic signal. According to the characteristics of the electromyographic signal, a 20-450Hz 4th-order Butterworth filter is used to perform band-pass filtering on the electromyographic signal.

[0055] Optionally, a window can be set according to a gait cycle, and the matrix D∈R can be constructed from the multi-channel EMG data within the window (200ms, 50% overlap). 4×N (N is the number of sampling points, 4 channels), the matrix D is a 4-channel electromyographic signal matrix, each row corresponds to a muscle channel. And construct the decomposition matrix D≈W·H, where W and H are two non-negative matrices, W∈R 4×k is the spatial weight matrix, H∈R K×N is the activation coefficient matrix, k is the number of collaborative modules (take k=2), and the NPU matrix multiplication instruction (KPU) is used to parallelize the iterative process and accelerate the processing process. For example, it can include:

[0056] Initialize the non-negative matrix W and the non-negative matrix H, and set the upper limit of the number of iterations and the error threshold. According to the settings, the solution goal is to minimize the error between the two matrices, that is,

[0057]

[0058] Where V′=W·H

[0059] According to the iterative formula and The matrices W and H are iterated, and the iteration ends when the number of iterations reaches an upper limit or the error is less than a threshold. After the iteration is complete, the activation matrix H is obtained, which reflects the temporal activation strength of the collaborative module. During healthy gait, H exhibits periodic fluctuations, synchronized with the gait cycle phase.

[0060] In order to solve the problem of misaligned muscle activation timing, such as the lag of the tibialis anterior muscle in the swing phase, and improve the synchronization between electrical stimulation and gait phase, the stimulation scheme was timing-coordinatedly corrected based on the H matrix.

[0061] First, extract the target muscle synergy module in the real-time H matrix, for example, module 2 corresponds to the activation curve H of ankle dorsiflexion 实时 Then, the healthy template H is aligned by the dynamic time warping (DTW) algorithm 模板 , locate the activation peak point t peak . Using H matrix and health template H 模板 The activation peak point t in peak It is determined whether an activation timing abnormality exists, and if so, timing compensation is performed in a next cycle according to an activation timing abnormality parameter.

[0062] Optionally, the timing compensation module includes:

[0063] A fusion unit is used to determine the rising edge point and the falling edge point using the activation peak point, calculate the delay deviation of the activation peak point, the rising edge point, and the falling edge point relative to the healthy template, and fuse the delay deviation results to form a standard delay deviation;

[0064] Determine whether to activate the advanced state or the delayed state by using the standard delay deviation;

[0065] Accordingly, in the activation hysteresis state, the stimulation time is calculated based on the standard delay deviation in the next cycle, and the electrical stimulation is triggered at the stimulation time, t stim_new =t event -min(|Δt|,30ms)+α, α=20ms safety margin;

[0066] In the activation advance state, the stimulation time is calculated based on the standard delay deviation, and the electrical stimulation is triggered at the stimulation time, tstim_new =t event +|Δt|.

[0067] For example, the activation peak point t peak , determine the three-point interval:

[0068] Rising edge point t L : The moment when the activation intensity first reaches the peak value m% (the initial stage of muscle activation)

[0069] Peak point t peak : Moment of maximum activation intensity (point of maximum motor unit recruitment)

[0070] Falling edge point t R : The moment when the activation intensity drops back to m% of the peak value (the starting point of activation decay)

[0071] Calculate the timing deviation for three time points:

[0072] Δt rise =t L_real -t L_template

[0073] Δt peak =t peak_real -t peak_template

[0074] Δt fall =t R_real -t R_template

[0075] Δt=α·Δt L +β·Δt peak +γ·Δt R

[0076] α, β, and γ are preset parameters. Optionally, α, β, and γ are set to 0.3, 0.5, and 0.2, respectively.

[0077] Accordingly, the stimulation timing of the next cycle can be adjusted according to the timing deviation:

[0078] When there is a lag, lag compensation is performed and the target muscle stimulation is triggered in advance in the next cycle. The formula is t stim_new =t event -min(|Δt|,30ms)+α, α=20ms safety margin. When there is a lead, perform overtime inhibition and delay the stimulus triggering timing: t stim_new =

[0079] t event +|Δt|.

[0080] Furthermore, the timing compensation module may also include: a template update unit, which is used to modify the original healthy template using the real-time activation matrix to form a new healthy template when the standard delay deviation is greater than the preset delay deviation threshold within a preset number of consecutive cycles. Using the above method, the healthy template can be continuously adjusted according to personalized requirements to make the activation timing adjustment more personalized. For example, if |Δt| is <15ms for three consecutive cycles, the healthy template is updated: and ω are weight parameters, exemplarily, and ω are 0.85 and 0.15 respectively.

[0081] This embodiment provides an electromyographic sensor attached to the skin surface of a target muscle group to collect multi-channel electromyographic signals from the target muscle group; a module is used to establish an electromyographic signal matrix based on the collected multi-channel electromyographic signals, and a spatial weight matrix for describing the muscle synergistic activation pattern and an activation matrix for describing the muscle temporal activation intensity are respectively established; the spatial weight matrix and the activation matrix are iterated to obtain the iterated spatial weight matrix and activation matrix, so that the difference between the product of the spatial weight matrix and the activation matrix and the electromyographic signal matrix is ​​less than a preset difference threshold; a timing compensation module is used to extract the activation curve of the target muscle synergy from the iterated activation matrix; the activation curve is matched with a preset healthy template, and the activation peak point is determined based on the matching result; the presence of activation timing anomaly is determined based on the matching point corresponding to the activation peak point. If activation timing anomaly exists, timing compensation is performed in the next cycle based on the activation timing anomaly parameter. By decomposing the electromyographic signal matrix into a spatial weight matrix and an activation matrix, and using iteration, the activation matrix can be composed to reflect the temporal activation intensity. Based on the matching results of the iterative activation matrix and the healthy template, the system searches for peak points and determines whether there are activation timing anomalies based on these peak points. If there are any, timing compensation is performed in the next cycle. This can address the misalignment of muscle activation timing caused by electrical stimulation of muscle coordination and improve the synchronization between electrical stimulation and gait phase. This allows for effective correction of electrical stimulation based on the desired effect, further enhancing the effectiveness of electrical stimulation.

[0082] In a preferred embodiment of this embodiment, the device may further include: a six-axis IMU sensor attached to key areas of lower limb movement for collecting triaxial acceleration and triaxial angular velocity signals; and a gait phase determination module for determining the current gait phase based on the triaxial acceleration and triaxial angular velocity signals. A six-axis IMU sensor can be attached to key areas of the patient's foot, such as the dorsum of the foot and calf, using the SPI protocol to collect triaxial acceleration and triaxial angular velocity signals. The raw IMU data is then dynamically normalized and filtered using a sliding window mean to suppress high-frequency noise. A Kalman filter is then used to suppress IMU noise. The baseline value is automatically calibrated based on the patient's resting state to eliminate the gravity acceleration component, enabling more accurate determination of the gait cycle and providing reliable gait information for electrical stimulation control. For example, heel strike (HS) and toe-off (TO) events can be detected based on the foot's vertical acceleration and pitch angular velocity, and the current gait phase (stance phase or swing phase) can be determined based on these gait events. The stance phase is the time window from the HS event to the TO event (about 60% of the gait cycle), and the swing phase is the time window from the TO event to the next HS event (about 40% of the gait cycle).

[0083] Furthermore, the device may also include: a determination module for determining the maximum and minimum muscle activation levels according to the muscle activation status of the gait stage, and determining a reasonable range of muscle activation based on the maximum and minimum muscle activation levels; and determining the maximum stimulation amplitude and stimulation pulse width without causing pain; a stimulation signal regulation and calculation module for obtaining the amplitude and pulse width of the electrical stimulation signal after regulation based on the position of the muscle contraction force in the reasonable range and the fatigue level of the muscle according to the currently calculated fatigue level.

[0084] Exemplarily, the patient's personalized stimulation parameters may be recorded before electrical stimulation begins, specifically including:

[0085] The surface electromyographic signals in the state of relaxation (rest) and maximum voluntary contraction (MVC) are recorded as RMS r and RMS m , RMS r Represents the baseline electrical activity when the muscle is not activated, RMS m Calibrate the patient's personalized maximum muscle strength level.

[0086] At the same time, record and measure the maximum stimulation amplitude and stimulation pulse width when the patient does not feel pain, and record it as A max and W max , to prevent excessive current from damaging muscles;

[0087] Record the minimum stimulation amplitude and pulse width that can cause muscle contraction in the patient, and record it as A min and W min, ensure that the muscles can be activated normally and avoid ineffective stimulation.

[0088] Using the above method, the muscle activation status can be determined to determine the maximum and minimum muscle activation levels, and based on the maximum and minimum muscle activation levels, the reasonable interval of muscle activation can be determined, as well as the maximum stimulation amplitude and stimulation pulse width without causing pain, and the minimum stimulation amplitude and pulse width with contraction function.

[0089] Accordingly, the device further comprises: a real-time feature calculation module for calculating the time domain feature root mean square value, the frequency domain feature average power frequency and the fatigue feature value according to the real-time electromyographic signal;

[0090] Accordingly, the stimulation signal control calculation module includes:

[0091] The fatigue value activation calculation unit is used to map the fatigue characteristic value to the proportional parameters of the amplitude and pulse width using a sign function.

[0092] Optionally, real-time features of the electromyographic signal can be calculated, including the time domain feature root mean square value (RMS), the frequency domain feature mean power frequency (MPF), the nonlinear feature sample entropy (SampEn), and the electromyographic integral value iEMG.

[0093] The RMS value of the EMG signal is often used to assess the strength of muscle contraction:

[0094]

[0095] Where N is the amount of EMG data in the window, x i is the i-th data of the signal, which is used to represent the average power of the collected signal.

[0096] The MPF feature of the electromyographic signal refers to the average value of the power frequency distribution of the electromyographic signal generated during muscle contraction, which can be used to assess the strength and fatigue of muscle contraction:

[0097]

[0098] f represents the frequency of the surface electromyographic signal, and PSD represents the power spectrum of the surface electromyographic signal.

[0099] The muscle fatigue characteristic Q value can be used to detect the fatigue state of the muscle. The muscle fatigue characteristic Q value is represented by the myoelectric sample entropy value and the myoelectric integral value.

[0100]

[0101] u s and u i are the EMG sample entropy and EMG integral values ​​at the stimulation site, respectively.

[0102] Stimulation parameters (pulse width and amplitude) and maximum pulse width W max and maximum amplitude A max Proportional, proportional adjustment, for example, when RMS reaches the middle value, the stimulation parameter is 50% of the maximum value, the larger the electromyographic value, the larger the stimulation parameter of the next cycle, to promote the gradual increase of muscle strength in the affected muscle, to achieve "weak muscle strength is enhanced, strong muscle strength is maintained" adaptive treatment, to avoid insufficient stimulation or overload caused by fixed parameters.

[0103] The fatigue value activation calculation unit is implemented in the following manner:

[0104] The scale parameter is Accordingly, the stimulation signal control calculation module includes:

[0105] The amplitude control calculation unit is implemented in the following way:

[0106]

[0107] The pulse width control calculation unit is implemented in the following way:

[0108]

[0109] in,

[0110] α is the suppression parameter.

[0111] When RMS(k) is greater than RMS m In order to avoid muscle discomfort caused by excessive changes in stimulation parameters, the stimulation parameters are adjusted to a fixed maximum stimulation parameter; when RMS(k) is less than RMS r When the stimulation parameters are adjusted to the minimum stimulation parameters, the muscles are prompted to contract.

[0112] In addition, in order to stop electrical stimulation after fatigue is detected and implement fatigue emergency braking, the sign function is used, and its expression is as follows:

[0113]

[0114] Use sign to construct a non-negative activation function

[0115]

[0116] Using this formula prevents the stimulus parameters from having amplitudes when hour,

[0117] The expression of the mapping model is as follows:

[0118]

[0119] Among them, γ MPF It is fatigue suppression in the MPF dimension, and MPF reflects the change in the spectrum distribution of the electromyographic signal.

[0120] The relationship between MPF ​​and fatigue: When muscles fatigue, high-frequency electromyographic components decrease and the proportion of low-frequency components increases, resulting in a decrease in MPF. MPF(0) is the initial MPF value at rest and serves as an initial reference. When MPF ​​decreases by more than 20% (i.e., MPF(k) / MPF(0) < 0.8), it indicates that the muscle has entered the early stage of fatigue.

[0121]

[0122] Where α is the suppression coefficient, which is optional and can be set to 0.2. The suppression coefficient γ_MPF can be used to proportionally reduce the stimulation intensity (for example, when MPF ​​drops to 70% of the initial value, γ_MPF = 0.94), delaying the progression of fatigue worsening, extending the effective treatment time, and improving the treatment effect.

[0123] Furthermore, the real-time feature calculation module includes: a fatigue feature value calculation unit, including: an electromyographic sample entropy value calculation subunit, used to calculate the electromyographic sample entropy value using the electromyographic signal; and an electromyographic integral value calculation subunit, used to calculate the electromyographic integral value using the electromyographic signal.

[0124] Sample entropy describes the complexity of a time series. A smaller value indicates a less complex time series and a lower probability of a new sequence. It is calculated as follows: Compared to the relaxed state at the beginning of stimulation, the rate of muscle pattern change decreases in the fatigued state, the activity of motor units tends to be more consistent, and the random component decreases, resulting in a lower EMG sample entropy.

[0125] In this embodiment, the entropy of myoelectric samples is calculated as follows: given an EMG signal sequence {x(i), i = 1, 2, 3 ..., N} of length N, starting from any point, m consecutive data are randomly selected to construct a set of m-dimensional vectors X m (i), denoted as X m (i)=[x(i),x(i+1),…x(i+m-1)], where i, 2,…N-m+1.

[0126] Define vector X m (i) and X m The distance d between (j) is the maximum absolute value of the difference between the corresponding elements of the vector, that is,

[0127]

[0128] Given a threshold r, define To satisfy {d(X m (i),X m The average value of the number of j where (j))≤r,j≠i}.

[0129] Increase m by 1 and define it according to the above steps To satisfy {d(X m+1 (i),X m+1 The average value of the number of j where (j))≤r,j≠r}.

[0130] The sample entropy of the EMG signal is defined as:

[0131]

[0132] In the above calculation process, m is the dimension of the reconstructed phase space, called the embedding dimension. In this study, m = 2 is selected; r is an arbitrarily given distance, called the similarity tolerance. Experience has shown that r = (0.1~0.25)Std, where Std represents the standard deviation of the data, and r = 0.15Std.

[0133] IEMG refers to the total amount of discharge of motor units involved in muscle activity within a certain period of time, and its amplitude usually increases with the degree of fatigue.

[0134]

[0135] Among them, t1 is the start time of collecting EMG signals, t2 is the end time of collecting EMG signals, and x(t) is the size of EMG signals.

[0136] In a preferred implementation of this embodiment, the device may also include: a pause module, used to align the spatial weight matrix with the healthy template, calculate the approximation between the spatial weight matrix and the healthy template, determine whether there is abnormal coordination based on the approximation, and suspend electrical stimulation when the coordination similarity is lower than a preset similarity threshold; a comparison module, used to compare the columns of the spatial weight matrix with the corresponding columns of the healthy template to determine the muscles corresponding to the weight abnormality when there is abnormal coordination; a first electrical stimulation adjustment unit, used to calculate the stimulation adjustment parameters based on the difference between the muscle corresponding columns in the spatial weight matrix and the muscle corresponding columns in the healthy template when the muscle weight is too high, and electrically stimulate the muscles according to the stimulation adjustment parameters at the next stimulation moment; a second electrical stimulation adjustment unit, used to calculate the stimulation adjustment parameters based on the muscle corresponding columns in the healthy template when the muscle weight is too low, and electrically stimulate the muscles according to the stimulation adjustment parameters at the next stimulation moment.

[0137] Linear interpolation can be used to align the data with the healthy template library. The spatial weight matrix of the healthy template library is then compared to calculate the cosine similarity S. A similarity of S < 0.7 indicates abnormal synergy. Furthermore, when the synergy similarity is too low, for example, S < 0.3, an emergency mode is triggered, pausing stimulation.

[0138]

[0139] When a muscle is detected to have a high weight, the stimulation parameters are modified according to the indicators of the healthy template library to reduce its amplitude. This is achieved in the following way:

[0140] A new (k+1)=A(k+1)·(1-α·|W 患,i -W 健,i |);

[0141] For example, the patient's rectus femoris weight W 患 =0.8(Healthy template W 健 =0.3), indicating that the rectus femoris was overactivated, which may be due to the compensation caused by the weakness of the antagonist muscle. Therefore, the stimulation intensity was reduced to 76% of the initial stimulation parameters (A new =A×(1-0.3×0.5)=0.85A), suppressing abnormal coordination. Where α is a parameter.

[0142] When it is detected that the weight of a muscle is too low, the stimulation parameters are modified according to the indicators of the healthy template library to enhance its amplitude. For example, this can be achieved in the following way:

[0143] A 补偿 (k+1)=A(k+1)·(1+β·W 健,i )

[0144] If the tibialis anterior muscle weight W in the healthy template 健 =0.9, and the patient weight W 患 =0.4, then the stimulus intensity after compensation is increased to A 补偿=1.18A, strengthen ankle dorsiflexion coordination, and promote the reconstruction of correct gait patterns. In this embodiment, an improved recurrent neural network can also be used to perform time series analysis on electromyographic signals to achieve intelligent judgment of fatigue thresholds. The system can use long short-term memory networks (LSTM), temporal convolutional networks (TCN) and gated recurrent units (GRU). This system preferably uses gated recurrent units as the core prediction module. Compared with LSTM, it improves computing efficiency by 20%-30% by simplifying the gate structure, making it more suitable for deployment in low-power embedded device environments. The system input features include a three-dimensional vector of time-domain integrated electromyographic values ​​(iEMG), sample entropy, and feature Q values. The input tensor dimension is defined as [batch, seq_length, features], where batch represents the data batch size, seq_length represents the time series window length, and features corresponds to the feature dimension. In a preferred embodiment, batch = 2, seq_length = 20, and features = 3 are set to form a three-order tensor input structure of [2, 20, 3].

[0145] Figure 3 This is a system structure diagram of the gated recurrent network in the adaptive neuromuscular control device based on electromyographic feedback provided by an embodiment of the present invention, which is the internal structure of the GRU unit with a time step of t. and Represents matrix multiplication and matrix addition operations respectively. The GRU network includes a reset gate (r t ) and update gate (z t ) Double gating mechanism, the update gate determines which information to discard and which information to add, and the reset gate determines how much of the network state at the previous moment is written to capture short-term memory. Represents the candidate hidden layer and updates the currently retained information. Its mathematical model is expressed as:

[0146] r t =σ(W r x t +U r h t-1 )

[0147] z t =σ(W z x t +U z h t-1 )

[0148]

[0149] Where W r ,W z ,W,U Z and U is the trainable weight matrix, h t-1Represents the information of the previous time step t-1, x t Represents the input vector at the tth time step, σ is the Sigmoid function, and ⊙ represents the Hadamard product operation. The network output y t+1 The fatigue prediction Q value corresponding to the next time series is recorded as Q(k+1).

[0150] Fatigue determination uses a dual-insurance mechanism: When Q(k+1) / Q(0) < 0.5, the system automatically terminates output after the current electrical stimulation cycle. Specifically, if the real-time acquired Q(k) value reaches the threshold before the predicted value, the system immediately initiates an emergency stop. This dual-judgment strategy achieves a 10ms-level real-time response through edge computing devices, effectively preventing muscle damage caused by prediction delays.

[0151] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. An adaptive neuromuscular control device based on electromyographic feedback, characterized in that: include: Myoelectric sensors are attached to the skin surface of the target muscle group and are used to collect multi-channel myoelectric signals of the target muscle group; Establishing a module for establishing an electromyographic signal matrix based on the collected multi-channel electromyographic signals, and establishing a spatial weight matrix for describing the muscle synergistic activation pattern and an activation matrix for describing the muscle temporal activation intensity; Iterating the spatial weight matrix and the activation matrix to obtain an iterated spatial weight matrix and an activation matrix, so that a difference between a product of the spatial weight matrix and the activation matrix and the electromyographic signal matrix is ​​less than a preset difference threshold; A timing compensation module is used to extract the activation curve of the target muscle synergy from the iterated activation matrix; Matching the activation curve with a preset healthy template, and determining the activation peak point according to the matching result; It is determined whether there is an activation timing abnormality according to the matching point corresponding to the activation peak point. If there is an activation timing abnormality, timing compensation is performed in the next cycle according to the activation timing abnormality parameter.

2. The device according to claim 1, characterized in that The timing compensation module includes: A fusion unit is used to determine the rising edge point and the falling edge point using the activation peak point, calculate the delay deviation of the activation peak point, the rising edge point, and the falling edge point relative to the healthy template, and fuse the delay deviation results to form a standard delay deviation; Determine whether to activate the advanced state or the delayed state by using the standard delay deviation; Accordingly, in the activation hysteresis state, the stimulation time is calculated based on the standard delay deviation in the next cycle, and the electrical stimulation is triggered at the stimulation time, t stim_new =t event -min(|Δt|,30ms)+α, α = 20ms safety margin; In the activation advance state, the stimulation time is calculated based on the standard delay deviation, and the electrical stimulation is triggered at the stimulation time, t stim_new =t event +|Δt|.

3. The device according to claim 2, characterized in that The timing compensation module further includes: The template updating unit is used to modify the original healthy template using the real-time activation matrix to form a new healthy template when the standard delay deviation is greater than the preset delay deviation threshold within a preset number of consecutive cycles.

4. The device according to claim 1, characterized in that The device further comprises: Six-axis IMU sensor, attached to the key parts of the lower limb movement, is used to collect three-axis acceleration and three-axis angular velocity signals; The gait stage determination module is used to determine the current gait stage based on the three-axis acceleration and three-axis angular velocity signals.

5. The device according to claim 1, characterized in that The device further comprises: a determination module, configured to determine maximum and minimum muscle activation levels according to muscle activation conditions during gait phases, and determine a reasonable interval of muscle activation based on the maximum and minimum muscle activation levels; and Determine the maximum stimulation amplitude and stimulation pulse width without causing pain; The stimulation signal control calculation module is used to adjust the amplitude and pulse width of the electrical stimulation signal according to the position of the muscle contraction force in a reasonable range and the fatigue degree of the muscle.

6. The device according to claim 5, characterized in that The device further comprises: A real-time feature calculation module is used to calculate the time domain feature root mean square value, the frequency domain feature average power frequency and the fatigue feature value according to the real-time electromyographic signal; Accordingly, the stimulation signal control calculation module includes: The fatigue value activation calculation unit is used to map the fatigue characteristic value to the proportional parameters of the amplitude and pulse width using a sign function.

7. The device according to claim 6, characterized in that The fatigue value activation calculation unit is implemented in the following manner: The scale parameter is Accordingly, the stimulation signal control calculation module includes: The amplitude control calculation unit is implemented in the following way: The pulse width control calculation unit is implemented in the following way: in, α is the suppression parameter.

8. The device according to claim 6, characterized in that The real-time feature calculation module includes: Fatigue characteristic value calculation unit, including: The myoelectric sample entropy value calculation subunit is used to calculate the myoelectric sample entropy value using the myoelectric signal; The myoelectric integral value calculation subunit is used to calculate the myoelectric integral value using the myoelectric signal.

9. The device according to claim 5, characterized in that The device further comprises: a pause module, configured to align the spatial weight matrix with the healthy template, calculate the similarity between the spatial weight matrix and the healthy template, determine whether there is abnormal synergy based on the similarity, and pause electrical stimulation when the synergy similarity is lower than a preset similarity threshold; A comparison module is used to compare the columns of the spatial weight matrix with the corresponding columns of the healthy template when there is abnormal coordination, and determine the muscle corresponding to the weight abnormality; A first electrical stimulation adjustment unit is configured to calculate a stimulation adjustment parameter according to a difference between a muscle corresponding column in a spatial weight matrix and a muscle corresponding column in a healthy template when a muscle weight is too high, and to electrically stimulate the muscle according to the stimulation adjustment parameter at the next stimulation moment; The second electrical stimulation adjustment unit is used to calculate the stimulation adjustment parameters according to the muscle corresponding column in the health template when the muscle weight is low, and to electrically stimulate the muscle according to the stimulation adjustment parameters at the next stimulation moment.

Images