A percutaneous auricular vagus nerve stimulation system and method for treating dysphagia after stroke

Abstract

This invention relates to the field of rehabilitation medical equipment technology, and more particularly to a percutaneous auricular vagus nerve stimulation system and method for treating dysphagia after stroke. The system includes: a positioning module, which collects skin impedance distribution information in the concha region and determines the target physical coordinates based on the skin impedance distribution information; a titration module, which, based on the target physical coordinates, obtains the sensory threshold and pain threshold of the subject to electrical stimulation and calculates and generates an individualized intensity benchmark; and a signal module, which, after establishing the individualized intensity benchmark, simultaneously collects the submental muscle group signal and respiratory waveform of the subject and maps them to form a multimodal feature vector. In this invention, the output of the biomimetic frequency-modulated pulse is dynamically synchronized with the patient's spontaneous exhalation phase, effectively improving the physiological coordination of breathing and swallowing during the intervention process and reducing the risk of adverse reactions such as airway aspiration or coughing during electrical stimulation therapy.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation medical equipment technology, and in particular to a percutaneous vagus nerve stimulation system and method for treating dysphagia after stroke. Background Technology

[0002] In the rehabilitation treatment of dysphagia, neuromuscular electrical stimulation is a common intervention method. It usually involves applying periodic currents to the submental or laryngeal muscle groups to induce muscle contraction, thereby assisting in swallowing. However, the swallowing process itself has a natural coupling relationship with the respiratory rhythm, especially since it often occurs during the expiratory phase under normal physiological conditions. Therefore, how to balance the respiratory rhythm and the timing of swallowing triggering during electrical stimulation intervention has gradually become one of the key issues affecting the safety and effectiveness of treatment.

[0003] In existing technologies, most electrical stimulation devices use fixed frequency or output methods based on simple trigger signals, lacking dynamic perception and matching of the patient's real-time respiratory status. This may result in the electrical stimulation trigger time not being consistent with the respiratory phase, which in some cases can easily cause the swallowing and inhalation processes to overlap, thereby increasing the probability of choking or aspiration, and also affecting the comfort and stability of the overall intervention process. Summary of the Invention

[0004] To overcome the above deficiencies, the present invention provides a percutaneous vagus nerve stimulation system and method for treating dysphagia after stroke, aiming to improve the problem that most existing electrical stimulation devices use fixed frequency or output methods based on simple trigger signals.

[0005] In a first aspect, the present invention provides the following technical solution: a percutaneous vagus nerve stimulation system for treating dysphagia after stroke, comprising:

[0006] The positioning module collects skin impedance distribution information in the concha region and determines the target's physical coordinates based on the skin impedance distribution information.

[0007] The titration module, based on the target physical coordinates, obtains the sensory threshold and pain threshold of the test object to electrical stimulation, and calculates and generates an individualized intensity benchmark.

[0008] After establishing the individualized intensity benchmark, the signal module synchronously acquires the submental muscle group signals and respiratory waveforms of the subject under test, and maps them to form a multimodal feature vector;

[0009] The control module, based on the multimodal feature vector, performs swallowing action recognition and associates and encapsulates it with the target physical coordinates to generate trigger commands;

[0010] The verification module, based on the trigger command, performs current limiting processing through loop impedance monitoring and outputs pulse control parameters.

[0011] The execution module, based on the pulse control parameters, drives the electrodes to apply an electrical stimulation signal corresponding to the swallowing action and acquires swallowing physiological feedback data;

[0012] The optimization module calculates the evaluation index deviation value and generates parameter correction amount based on the swallowing physiological feedback data, and updates the individualized intensity benchmark based on the parameter correction amount.

[0013] Preferably, in the positioning module, determining the target physical coordinates based on the skin impedance distribution information specifically includes the following steps:

[0014] Construct a two-dimensional topological map of skin impedance in the concha region;

[0015] Extract the impedance valley points from the two-dimensional topology map of the skin impedance;

[0016] The impedance valley points are mapped to a preset coordinate system to generate target physical coordinates.

[0017] Preferably, in the titration module, the step of obtaining the sensory threshold and pain threshold of the test subject to electrical stimulation and calculating and generating an individualized intensity benchmark specifically includes the following steps:

[0018] A step current is applied to the skin position corresponding to the target physical coordinates, and the current value that induces a sensory response is recorded as the sensory threshold.

[0019] Continue applying step current and record the current value that induces pain response as the pain threshold;

[0020] The difference between the pain threshold and the sensation threshold is calculated, and the difference is multiplied by a preset adjustment coefficient and added to the sensation threshold to generate an individualized intensity benchmark.

[0021] Preferably, in the signal module, the synchronous acquisition of the submental muscle group signal and respiratory waveform of the subject under test, and the mapping to form a multimodal feature vector, specifically includes the following steps:

[0022] Extract the temporal envelope features and frequency energy features of the submental muscle group signal;

[0023] Extract the respiratory phase features of the respiratory waveform;

[0024] The time-domain envelope feature, the frequency-domain energy feature, and the breathing phase feature are matrix-concatenated to construct a multimodal feature vector.

[0025] Preferably, in the control module, the step of recognizing the swallowing action based on the multimodal feature vector and associating and encapsulating it with the target physical coordinates to generate a trigger command specifically includes the following steps:

[0026] The multimodal feature vectors are input into a preset classification model to determine the initiation time of the swallowing action;

[0027] Extract the timestamp corresponding to the startup time;

[0028] The timestamp and the target physical coordinates are encapsulated to generate a trigger command.

[0029] Preferably, in the verification module, the current limiting processing through loop impedance monitoring and the output pulse control parameters specifically include the following steps:

[0030] In response to the trigger command, the loop impedance value of the interface between the electrode and the skin is acquired;

[0031] Calculate the difference between the loop impedance value and the preset upper limit threshold impedance value;

[0032] The individualized intensity benchmark is attenuated and modulated based on the comparison difference to generate pulse control parameters.

[0033] Preferably, in the execution module, the driving electrode applies an electrical stimulation signal corresponding to the swallowing action, and the acquisition of swallowing physiological feedback data specifically includes the following steps:

[0034] The pulse control parameters are analyzed to generate a square wave pulse sequence;

[0035] The multimodal feature vector is analyzed to extract respiratory phase features. During the exhalation phase, the square wave pulse sequence is output to the skin position corresponding to the target physical coordinates.

[0036] The contraction duration and peak amplitude of the submental muscle group signals after stimulation were extracted to construct swallowing physiological feedback data.

[0037] Preferably, the step of outputting the square wave pulse sequence to the skin location corresponding to the target physical coordinates during the expiration phase of the respiratory phase specifically includes the following steps:

[0038] Extract discrete time sampling points characterized during the expiratory phase, calculate the instantaneous phase angle corresponding to the discrete time sampling points, and generate a phase angle sequence;

[0039] The phase angle sequence is input into a preset nonlinear mapping function to obtain a frequency modulation coefficient sequence corresponding to the discrete time sampling point;

[0040] Based on the frequency modulation coefficient sequence, the pulse period of the square wave pulse sequence is scaled periodically to reconstruct and generate a biomimetic frequency modulation pulse sequence.

[0041] The biomimetic frequency-modulated pulse sequence is output to the skin position corresponding to the target physical coordinates.

[0042] Preferably, in the optimization module, the step of calculating the deviation value of the evaluation index and generating the parameter correction amount, and updating the individualized intensity benchmark based on the parameter correction amount, specifically includes the following steps:

[0043] The difference between the swallowing physiological feedback data and the preset baseline feature data is calculated as the evaluation index deviation value;

[0044] The deviation value of the evaluation index is input into the proportional-integral-derivative controller to obtain the parameter correction amount;

[0045] The parameter correction amount is algebraically superimposed with the individualized strength benchmark to obtain the updated individualized strength benchmark.

[0046] Secondly, the present invention provides the following technical solution: a percutaneous vagus nerve stimulation method for treating dysphagia after stroke, comprising the following steps:

[0047] S1. Collect skin impedance distribution information in the concha region, and determine the target physical coordinates based on the skin impedance distribution information;

[0048] S2. Based on the target physical coordinates, obtain the sensory threshold and pain threshold of the test object to electrical stimulation, and calculate and generate an individualized intensity benchmark.

[0049] S3. After establishing the individualized intensity benchmark, simultaneously collect the submental muscle group signals and respiratory waveforms of the subject to be tested, and map them to form a multimodal feature vector;

[0050] S4. Based on the multimodal feature vector, perform swallowing action recognition and associate it with the target physical coordinates to generate a trigger command;

[0051] S5. Based on the trigger command, current limiting is performed through loop impedance monitoring, and pulse control parameters are output.

[0052] S6. Based on the pulse control parameters, apply an electrical stimulation signal corresponding to the swallowing action to the driving electrode to obtain swallowing physiological feedback data.

[0053] S7. Based on the swallowing physiological feedback data, calculate the evaluation index deviation value and generate parameter correction amount, and update the individualized intensity benchmark based on the parameter correction amount.

[0054] The present invention has the following beneficial effects:

[0055] 1. In this invention, the output of the bionic frequency-modulated pulse is dynamically synchronized with the patient's spontaneous exhalation phase, which effectively improves the physiological coordination of breathing and swallowing during the intervention process and reduces the risk of adverse reactions such as airway aspiration or coughing during electrical stimulation therapy.

[0056] 2. In this invention, the evaluation deviation is calculated by continuously quantifying the actual contraction response of the target muscle group, and the individualized stimulation intensity benchmark of subsequent cycles is finely adjusted accordingly. This avoids insufficient or excessive stimulation due to muscle fatigue or tolerance in long-term applications, and maintains the stability of long-term treatment effects.

[0057] 3. In this invention, the initial safety parameter benchmark is automatically established by utilizing multimodal features and mapped to generate specific electrode array output routes. This reduces the reliance on manual experience-based debugging during the initial calibration stage of the equipment, improves the accuracy of target muscle spatial gating, and shortens the preparation time before clinical treatment. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a percutaneous vagus nerve stimulation system for treating dysphagia after stroke, as proposed in this invention.

[0059] Figure 2 This is a flowchart of a percutaneous vagus nerve stimulation method for treating dysphagia after stroke, as proposed in this invention. Detailed Implementation

[0060] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1:

[0062] In a first embodiment of the present invention, the present invention provides a percutaneous vagus nerve stimulation system for treating dysphagia after stroke, such as... Figure 1 As shown, it includes:

[0063] The positioning module collects skin impedance distribution information in the concha region and determines the target's physical coordinates based on the skin impedance distribution information;

[0064] Furthermore, in the positioning module, determining the target's physical coordinates based on skin impedance distribution information specifically includes the following steps:

[0065] Construct a two-dimensional topological map of skin impedance in the concha region;

[0066] Extract impedance valley points from the two-dimensional topological map of skin impedance;

[0067] The impedance valley point is mapped to a preset coordinate system to generate the target physical coordinates.

[0068] Specifically, the positioning module is located by covering the surface of the concha region. A high-density microelectrode array is used to inject high-frequency microcurrents with a frequency range of 10kHz to 100kHz and an amplitude range of 1μA to 100μA into the skin. A four-electrode measurement method is employed to simultaneously acquire the voltage matrix at the corresponding interface, thus eliminating interference from the electrode's own contact impedance. In the four-electrode measurement method, the two outer electrodes are responsible for injecting the excitation current, and the two inner electrodes are responsible for acquiring the induced voltage. The formula for calculating the skin impedance amplitude of a single measurement node is as follows:

[0069] ;

[0070] in the formula This indicates that the geometric center point of a single four-electrode measurement assembly is used as the equivalent coordinate. The amplitude of the skin impedance at the location, in ohms, is the magnitude of the complex impedance; This represents the differential voltage amplitude acquired by the two inner sensing electrodes relative to this equivalent coordinate, in volts. This indicates the amplitude of the excitation current injected into the two outer electrodes, in amperes. and These represent the equivalent row and column numbers of a single measurement combination in the two-dimensional Cartesian coordinate system of the electrode array grid, respectively, with values ​​ranging from 1 to... With 1 to .

[0071] In the two-dimensional Cartesian coordinate system of the electrode array grid, for discrete Skin impedance data points Spatial smoothing and expansion are performed using a bicubic spline interpolation algorithm, with natural boundary conditions applied. Specifically, the second derivative of the interpolation function is set to zero at the edge of the electrode array to generate a continuous skin impedance distribution function. The grid resampling resolution was set to 0.1 mm to construct a two-dimensional topological map of skin impedance in the concha region within the grid plane.

[0072] After acquiring a continuous two-dimensional topological map of skin impedance, the system extracts impedance valley points by solving for the spatial local minima of the skin impedance distribution function. The system calculates the spatial derivative of the topological map, with the following conditions:

[0073] ;

[0074] in the formula Represents the skin impedance distribution function In grid coordinates The first-order spatial gradient vector at a point indicates that the point is an extremum when the vector is zero. Represents the skin impedance distribution function In grid coordinates The result of the Laplace operator action at that point, i.e. The calculated value is used to confirm that the extreme point is a local minimum point when the value is greater than zero, which is the candidate set of impedance valley points.

[0075] When multiple candidate impedance valleys exist in the solution, the system extracts the anatomical prior region of the auricular branch of the vagus nerve mapped onto the electrode array grid plane as a mask. The system extracts all candidate impedance valleys falling within this mask region and selects the point with the absolute minimum impedance amplitude as the unique target impedance valley. Its horizontal and vertical coordinates in the two-dimensional Cartesian coordinate system of the electrode array grid are denoted as follows: and If there are no candidate valley points within the mask area, the sampling point with the absolute minimum impedance amplitude within that area is directly taken as the target impedance valley point.

[0076] To convert the grid coordinates into physical space coordinates that the actuator can directly drive, the system establishes a preset coordinate system with the midpoint of the line connecting the root of the helix foot and the lower edge of the antitragus as the origin, and uses a spatial affine transformation for mapping, as shown in the formula:

[0077] ;

[0078] in the formula for The column vector represents the target physical coordinates that are ultimately output to the control mechanism; for A column vector is mathematically constructed from the numerical values ​​obtained in the preceding steps. As the first row element, the value The transpose matrix formed by the elements of the second row; To characterize the spatial rotation and scaling mapping from the two-dimensional Cartesian coordinate system of the electrode array grid to a preset coordinate system. Transformation matrix; To characterize the relative offset between the origins of two coordinate systems Translate column vectors. Matrix With column vectors All calibrations are determined through a pre-shipment calibration process. Specifically, at least four calibration reference points with known physical locations are selected within the effective coverage area of ​​the microelectrode array. The coordinate matrix of each reference point in the electrode array grid coordinate system and its known physical coordinate matrix in the preset coordinate system are substituted into the least squares method for solution.

[0079] By converting individualized discrete impedance characteristics into a continuous mathematical surface, and generating target physical coordinates bound to the anatomical structure through affine mapping.

[0080] The titration module, based on the target physical coordinates, obtains the sensory threshold and pain threshold of the test object to electrical stimulation, and calculates and generates an individualized intensity benchmark.

[0081] Furthermore, in the titration module, obtaining the sensory threshold and pain threshold of the test subject to electrical stimulation, and calculating and generating an individualized intensity benchmark specifically includes the following steps:

[0082] A step current is applied to the skin position corresponding to the target physical coordinates, and the current value that induces a sensory response is recorded as the sensory threshold.

[0083] Continue applying step current and record the current value that induces pain response as the pain threshold;

[0084] The difference between the pain threshold and the sensory threshold is calculated, multiplied by a preset adjustment coefficient, and added to the sensory threshold to generate an individualized intensity benchmark.

[0085] Specifically, the titration module acquires the target physical coordinates in the form of a two-dimensional column vector output by the positioning module. ,in and The units are all millimeters, corresponding to the physical displacement components of the positioning module along the horizontal and vertical axes in a preset two-dimensional coordinate system. This preset two-dimensional coordinate system has the lowest point of the concha as its origin, and extends horizontally to the right as... The positive direction of the axis is vertically upward. The positive axis. The system drives the stimulation electrode to the corresponding real skin position via a micro-electronic control mechanism. Before performing the fitting action, the micro-electronic control mechanism first calls the pre-stored normal vector field data of the concha surface. This data is calculated offline during the factory calibration stage by performing a full-field scan of the standard ear mold using a 3D structured light scanner, and is stored in the storage unit in the form of a lookup table. The 3D coordinate system used in the lookup table has the same origin and the same... shaft and Based on the axial direction, the direction extending outward along the reference plane of the concha is defined as... In the positive direction of the axis, in the preset two-dimensional coordinate system shaft and The axis represents the projection component of this three-dimensional coordinate system onto the reference plane of the concha cavity. The system uses... In and To retrieve the unit normal vector at the corresponding position in the lookup table using a two-dimensional index. ,in , , The unit normal vector is located along the above three-dimensional coordinate system. axis, axis, The dimensionless components in the axial direction satisfy the following conditions: The micro-electromechanical system drives the stimulation electrode to translate towards the skin surface along the normal vector direction, while simultaneously reading the output value of the piezoresistive force sensor integrated at the electrode end in real time. When the contact force value enters the preset range of 0.5 Newtons to 1.5 Newtons and remains stable for more than 200 milliseconds, the system determines that the electrode has reliably adhered, terminates the translational movement, and locks the current pose.

[0086] After the electrodes are in place, the system plays a standardized voice prompt to the subject before applying the step current, informing them to press the handheld interactive trigger immediately when they first feel any slight stimulation in their ear, and then press the trigger again when the stimulation increases to the point of being unbearable. The system clearly states that the trigger does not need to be released between the two presses. The system uses the falling edge of each press, i.e. the moment the finger first presses, as the effective trigger timestamp.

[0087] After notification, the system applies a stepping current consisting of symmetrical biphase square waves to the contact position. The duration of both the positive and negative phase pulses is set to the same value between 200 and 500 microseconds, the phase interval between the positive and negative phases is set to 50 microseconds, the pulse frequency is set to between 20 and 25 Hz, and the initial peak current amplitude is set to 0 mA. The system increases the peak current amplitude in a stepwise manner with a constant step size of 0.1 mA and a fixed step period of 1 second. That is, the peak current amplitude changes once after each complete 1-second step period, and remains constant within a single step period.

[0088] During the incrementing process, the system monitors the level signal of the interactive trigger in real time. When the object under test presses the trigger for the first time, the system captures the falling edge timestamp of the button action, extracts the constant peak current amplitude corresponding to the step cycle in which the timestamp is located, and records it as the sensing threshold. The system then continues to increment with the original parameters and step size. When the subject presses the trigger for the second time, the system captures the second falling edge timestamp, extracts the constant peak current amplitude corresponding to the step cycle, and records it as the pain threshold. If the peak current amplitude has increased to the system's built-in 10 mA absolute safety limit before the second button press, the system will forcibly stop the current output and record 10 mA as the pain threshold. .

[0089] After the system completes dual-threshold data acquisition, it calls a preset algebraic mapping algorithm to generate an individualized intensity benchmark. The calculation formula is as follows:

[0090] ;

[0091] in the formula This represents the final generated individualized intensity benchmark, which is the peak current amplitude in milliamperes, and is used by subsequent stimulation output modules as the default driving peak current amplitude. The sensory threshold recorded in the preceding steps is the peak current amplitude corresponding to the first stimulus perception of the test object, in milliamperes; The pain threshold recorded in the preceding steps is the peak current amplitude corresponding to when the subject reaches the maximum tolerable boundary, in milliamperes. This represents the preset adjustment coefficient, which is a dimensionless empirical constant. Its value is a globally fixed value that is uniformly applicable to all test subjects. Before leaving the factory, it is solidified and written into the storage unit by the R&D personnel based on clinical data statistics. The value range is set to 0.4 to 0.6, which is used to define the individualized intensity benchmark in the middle and upper part of the interval between the sensory threshold and the pain threshold.

[0092] By converting individual differences in physiological sensitivity into quantifiable peak current values, an individual-calibrated intensity reference benchmark is provided for subsequent stimulus output.

[0093] The signal module, after establishing an individualized intensity benchmark, simultaneously acquires signals from the submental muscle group and respiratory waveforms of the subject under test, and maps them to form a multimodal feature vector;

[0094] Furthermore, in the signal module, the simultaneous acquisition of signals from the submental muscle group and respiratory waveforms of the subject under test, and the mapping to form a multimodal feature vector, specifically includes the following steps:

[0095] Extracting temporal envelope features and frequency domain energy features from submental muscle group signals;

[0096] Extract the respiratory phase features of the respiratory waveform;

[0097] The time-domain envelope features, frequency-domain energy features, and respiratory phase features are matrix-concatenated to construct a multimodal feature vector.

[0098] Specifically, the system uses the individualized intensity benchmark generated by the titration module as the basis for subsequent control parameters. The micro-electromechanical system continuously outputs a biphasic square wave electrical stimulation sequence to the target physical coordinates based on the peak current amplitude corresponding to this individualized intensity benchmark. During the continuous output of electrical stimulation, the signal module simultaneously activates a multi-channel physiological signal acquisition front-end for real-time monitoring. The system uses a surface electromyography (EMG) sensor attached to the submental triangle region between the lower edge of the mandible and the hyoid bone of the test subject to acquire raw EMG signals at a sampling rate of 1000 Hz. The system then obtains the submental muscle group signals after interference suppression by using a bandpass filter with a cutoff frequency of 20 Hz to 400 Hz and a 50 Hz or 60 Hz notch filter that is automatically switched based on the local device time zone configuration. Simultaneously, the system acquires raw respiratory signals at a sampling rate of 100 Hz by using a piezoelectric breathing chest band attached to the test subject at the level of the lower sternal angle, i.e., at the height of the diaphragm. The system then uses a low-pass filter with a cutoff frequency of 1 Hz to filter out high-frequency motion artifacts and obtain a smooth respiratory waveform. To ensure strict alignment of heterogeneous data, the two acquisition front-ends share the same high-precision hardware clock source for timestamp recording.

[0099] After the system starts and fills the initial 200-millisecond data buffer, it segments the preprocessed submental muscle group signal using a fixed-length sliding window to extract temporal envelope features. The sliding window duration is set to 200 milliseconds, and the window step size is set to 100 milliseconds, resulting in a continuous feature output frame rate of 10 frames per second. The temporal envelope features are calculated using the root mean square algorithm, with the following formula:

[0100] ;

[0101] in the formula Indicates the first Temporal envelope characteristics of submental muscle group signals in one analysis window, in microvolts; This represents the total number of sampling points contained within a single window, and the value is always 200. A positive integer, representing the index of the discrete-time sampling point within a single analysis window, with a value ranging from 1 to... ; Indicates the first The first window The actual amplitude value of the electromyographic signal corresponding to each discrete sampling point, in microvolts; It is a positive integer representing the sequence frame index of the sliding analysis window on the global time axis.

[0102] The system synchronously extracts frequency domain energy features within a sliding window. To meet the radix-2 operation requirements of the Fast Fourier Transform (FFT), the system pads the last 200 sampling points in the window with zeros to 256 points before performing a Discrete Fast Fourier Transform (DFT), generating a spectral distribution with a frequency resolution of approximately 3.9 Hz. The frequency domain energy features are obtained by integrating and summing the power spectral density within the effective physiological frequency band, as shown in the formula:

[0103] ;

[0104] in the formula Indicates the first The frequency domain energy characteristics of each analysis window, in microvolts squared; A natural number representing the node index of a discrete frequency sequence; Indicates the first The time-series signal of the submental muscle group within each window is padded with zeros to 256 points and then subjected to a Discrete Fast Fourier Transform. At the frequency index node... The output is a complex frequency domain value, which contains a real part and an imaginary part. The absolute value sign indicates the modulus of the complex frequency domain value. and These represent the lower and upper limits of the discrete frequency node index corresponding to the target integral frequency band, respectively, and their corresponding physical frequencies are mapped to 20 Hz and 400 Hz in the effective electromyographic activity range, respectively.

[0105] For synchronously acquired respiratory waveforms, the system extracts synchronous data segments perfectly aligned with the electromyography window based on hardware timestamps, and eliminates baseline drift by subtracting the mean of the DC bias. The system employs a fast Hilbert algorithm based on frequency-domain fast Fourier transform and inverse transform to construct an analytical signal, and removes edge effect errors by discarding 10% of the data at each end of the overlapping sliding window, thereby solving for the accurate instantaneous phase characteristics. The formula is as follows:

[0106] ;

[0107] in the formula Indicates the relationship with the first Instantaneous phase characteristics of respiratory waveforms with perfectly aligned timestamps of each electromyography analysis window, in radians, with a numerical distribution range from negative to positive pi, representing the complete respiratory cycle from inspiration to expiration. The aligned respiratory waveform time series represents the time series at the 1st... The actual physical amplitude scalar value after removing the mean at each frame index time; Indicates that for the first After globally performing a frequency domain Hilbert transform on the entire respiratory waveform time series at time n to generate an orthogonal time series, the corresponding time series at time n is extracted from this orthogonal time series. The scalar value of the orthogonal component at each frame index time.

[0108] Because the physical dimensions and numerical scales of the above three features differ significantly, the system employs an extreme value normalization algorithm before feature stitching. Based on historical calibration benchmarks, it linearly maps the time-domain envelope features and frequency-domain energy features to a dimensionless interval of 0 to 1, and maps the breathing phase features by dividing by pi to a dimensionless interval of -1 to +1. Subsequently, the system uses time-series frame indexing... To match the benchmark, the three normalized features at the same time point are subjected to high-dimensional mapping and matrix concatenation to construct a unified multimodal feature vector, the formula of which is:

[0109] ;

[0110] in the formula Indicates the first Output of time series frames Multimodal feature vectors; This represents the dimensionless feature of the normalized time-domain envelope. This represents the dimensionless characteristic of the frequency domain energy after normalization. This indicates the dimensionless characteristics of the normalized respiratory phase; the upper right corner... The symbol indicates that the transpose operation is performed on the matrix in this row, converting it into a strict column vector form to meet the matrix operation dimension requirements of the subsequent control algorithm for the input variables of the system state space.

[0111] By transforming multi-source heterogeneous electromyographic activity characteristics and respiratory rhythm states into time-aligned and dimensionlessly uniform mathematical vectors, a standardized underlying data foundation is laid for subsequent dynamic analysis of the synergistic relationship between swallowing actions and respiratory cycles.

[0112] The control module, based on multimodal feature vectors, recognizes swallowing actions and associates them with the target's physical coordinates to generate trigger commands.

[0113] Furthermore, in the control module, based on multimodal feature vectors, swallowing action recognition is performed and associated with the target physical coordinates for encapsulation, generating trigger commands. This specifically includes the following steps:

[0114] Input the multimodal feature vector into the preset classification model to determine the initiation time of the swallowing action;

[0115] Extract the timestamp corresponding to the startup time;

[0116] The timestamp and target physical coordinates are encapsulated to generate a trigger command.

[0117] Specifically, the control module continuously receives normalized multimodal feature vectors output by the signal module in time-series frames. The control module has a pre-installed logistic regression classification model. The weight parameters of this model are trained offline using a maximum likelihood estimation algorithm on a manually labeled multimodal dataset containing historical swallowing actions and resting states. The sample labels in the training dataset are defined according to the following rules: feature vector frames collected during periods confirmed by swallowing imaging to have actual pharyngeal swallowing actions are labeled as positive samples with a label value of 1; feature vector frames collected during periods when the subject is breathing calmly and has no swallowing intention are labeled as negative samples with a label value of 0. The system inputs the three-dimensional multimodal feature vector of each frame into the classification model frame by frame, calculating the predicted probability that the subject is in a swallowing state for the current time-series frame. The prediction formula is as follows:

[0118] ;

[0119] in the formula Indicates the first The probability prediction value of a time series frame corresponding to the actual swallowing action of the subject under test is a dimensionless scalar with a value range between 0 and 1. This represents the 3x1 multimodal feature vector input from the preceding steps; This represents the weight column vector of the classification model, where the three elements correspond to the feature weight coefficients of the three dimensionless features in the multimodal feature vector: time domain envelope, frequency domain energy, and respiratory phase. This represents the transpose of the column vector containing the weights. This represents the bias term constant of the classification model; This represents the linear projection scalar value of the current multimodal feature vector onto the hyperplane of the classification model. The larger its absolute value, the higher the classification confidence of the current frame.

[0120] To filter out isolated physiological artifacts and accurately pinpoint the true swallowing initiation moment, the system performs continuous state monitoring logic on the output probability sequence. The system sets a swallowing determination threshold of 0.8, a fixed parameter determined during the R&D phase using 10-fold cross-validation on the aforementioned labeled dataset with Youden's exponent as the optimization criterion, balancing recognition sensitivity and specificity. When the predicted probability value of each of three consecutive time-series frames (a 300-millisecond observation window) is greater than or equal to this threshold, the system determines that a valid swallowing action has been detected. The system then identifies the first time-series frame in these three consecutive frames that meets the threshold condition as the physical initiation frame of the swallowing action. To ensure time reference consistency, the system calls a globally shared high-precision hardware clock register, identical to the preceding signal acquisition front-end, to extract the absolute system timestamp corresponding to the time the physical initiation frame was acquired.

[0121] After successfully extracting the absolute system timestamp, the control module calls the positioning module to initially generate and store the two-dimensional target physical coordinates in the system's global shared memory. The system performs memory-level data alignment between the extracted timestamp and the target physical coordinates, and constructs a trigger instruction column vector according to the following encapsulation rules:

[0122] ;

[0123] in the formula This represents the three-by-one dimension trigger instruction column vector ultimately generated by the control module, which serves as the data communication payload sent to the stimulus execution bus. This represents the absolute system timestamp scalar value corresponding to the physical initiation frame of the preceding swallowing action, in milliseconds, and occupies the first 32-bit unsigned integer field of the payload data frame. and These represent the scalar values ​​of the physical displacement components of the target physical coordinates retrieved from shared memory along the horizontal and vertical axes of a preset two-dimensional coordinate system, respectively. The units are both in millimeters, and they occupy the second and third 16-bit signed integer fields of the load data frame; the upper right corner... The symbol represents the transpose of the three row-arranged elements into a column vector format for encapsulation; the downstream stimulus execution unit parses the data sequentially according to the field order. As the absolute time reference for stimulus triggering, and This serves as the target displacement input for the stimulation head positioning mechanism.

[0124] This enabled real-time recognition of the swallowing intention from multiple physiological signals and generated a comprehensive driving command that included a time trigger reference and a spatial target location.

[0125] The verification module, based on the trigger command, performs current limiting processing through loop impedance monitoring and outputs pulse control parameters.

[0126] Furthermore, in the verification module, current limiting is performed through loop impedance monitoring, and the output pulse control parameters specifically include the following steps:

[0127] In response to a trigger command, the loop impedance value of the interface between the electrode and the skin is acquired;

[0128] Calculate the difference between the loop impedance value and the preset upper limit threshold impedance value;

[0129] The individualized intensity benchmark is attenuated and modulated based on the comparison difference to generate pulse control parameters.

[0130] Specifically, the verification module continuously monitors the 3×1-dimensional trigger command column vector issued by the control module. After parsing the absolute system timestamp and the target physical displacement component, the system activates the hardware-level impedance detection loop before the trigger time specified by the timestamp arrives. To avoid prematurely inducing unexpected muscle contraction, the system sends a painless detection pulse below the human sensory nerve irritation threshold to the electrode contact corresponding to the current target physical displacement component through a micro-electronic control mechanism. The pulse amplitude is set to 1 mA and the pulse width is set to 100 microseconds. At the constant current flat-top period of the positive phase waveform output by the detection pulse, i.e., at the 50th microsecond, the system synchronously triggers the analog-to-digital converter to sample the instantaneous voltage feedback signal applied to the electrode-skin contact interface, and divides this instantaneous voltage value by the 1 mA detection constant current amplitude to calculate the actual loop impedance value of the electrode-skin contact interface.

[0131] After obtaining the loop impedance value, the system calculates the difference between it and the preset impedance upper limit threshold stored in the safety configuration register. The calculation formula is as follows:

[0132] ;

[0133] in the formula This represents the difference between the loop impedance value and the preset upper limit threshold impedance, expressed in ohms. This represents the actual circuit impedance value calculated by sampling the voltage during the flat-top period of the probe pulse, in ohms. This indicates the system's preset upper limit impedance threshold, which is set here to 2000 ohms to meet the general burn prevention safety standards for medical electrical equipment. The unit is ohms.

[0134] After obtaining the comparison difference, the system reads the individualized strength benchmark from the global shared memory. This value is determined by the upstream titration step during the initialization phase before patient use through an incremental current titration process: the titration step increases the output current stepwise from 0 mA in 0.5 mA increments, simultaneously acquiring the root mean square amplitude of the electromyography signal on the patient's laryngeal surface. When this root mean square amplitude first exceeds three times the standard deviation of the resting baseline mean, the current output current value is multiplied by a safety reduction factor of 0.8 to serve as the individualized intensity benchmark, with an effective range of 1 mA to 20 mA. This value is written to global shared memory after titration and remains valid throughout the entire treatment cycle. The system calls this individualized intensity benchmark and performs nonlinear attenuation modulation based on the comparison difference. When the comparison difference is less than or equal to 0, it indicates that the current contact impedance is within the absolutely safe range, and the system does not attenuate; when the comparison difference is greater than 0, it indicates a risk of skin burns due to excessive local current density caused by electrode loosening, and the system initiates exponential attenuation limiting, the modulation formula of which is:

[0135] ;

[0136] in the formula This indicates the amplitude of the final output limited pulse current after attenuation modulation, in milliamperes. This represents the individualized strength baseline scalar value retrieved from the system's global shared memory, with a valid range of 1 mA to 20 mA, and the unit is mA; Represents an exponential function with the natural constant as its base; This represents the system's preset attenuation coefficient scalar, which is set to 0.005 and is the reciprocal of ohms. The engineering basis for this value is that when the impedance exceeds the limit of 200 ohms, the output current will rapidly decrease to about 37% of the reference value, thus balancing the sensitivity to prevent burns and the continuity of weak stimulation. This represents the maximum value between 0 and the comparison difference. When the comparison difference is less than or equal to 0, the function outputs 0, making the overall calculation result of the exponent term 1, ensuring that the current reference remains unchanged and is not tampered with.

[0137] After calculating the amplitude of the limited pulse current, the system encapsulates it with preset fixed stimulation therapy parameters into a matrix to generate the final pulse control parameters sent to the hardware execution unit. The formula is as follows:

[0138] ;

[0139] in the formula This represents a 3×1-dimensional column vector of pulse control parameters output to the underlying hardware analog-to-digital converter circuit. This indicates the amplitude of the limiting pulse current calculated previously, in milliamperes; This represents the system's preset electrical stimulation pulse width scalar value, which is set to 200 microseconds here based on the effective depolarization time of the swallowing muscle group; This indicates the system's preset electrical stimulation pulse frequency scalar value, which is set to 80 Hz here based on the physiological characteristics of inducing smooth tetanic contractions; the upper right corner... The symbol indicates that the three elements arranged in a row are transposed into a strict column vector format. After encapsulation, the system packages the pulse control parameters together with the original trigger instruction into a low-level driver data frame.

[0140] Closed-loop monitoring and abnormal amplitude limiting of electrode contact status were achieved before the stimulation pulse was delivered, effectively preventing local burns caused by electrode deterioration.

[0141] The execution module, based on pulse control parameters, drives the electrodes to apply electrical stimulation signals corresponding to the swallowing action and acquires swallowing physiological feedback data.

[0142] Furthermore, in the execution module, the driving electrodes apply electrical stimulation signals corresponding to the swallowing action, and the acquisition of swallowing physiological feedback data specifically includes the following steps:

[0143] Analyze the pulse control parameters and generate a square wave pulse sequence;

[0144] The respiratory phase features are extracted by parsing the multimodal feature vectors. During the expiratory phase, the square wave pulse sequence is output to the skin position corresponding to the target physical coordinates.

[0145] The contraction duration and peak amplitude of the submental muscle group signals after stimulation were extracted to construct swallowing physiological feedback data.

[0146] Furthermore, during the expiratory phase, when the respiratory phase is characterized by exhalation, outputting the square wave pulse sequence to the skin location corresponding to the target physical coordinates specifically includes the following steps:

[0147] Extract discrete time sampling points characterized during the expiratory phase, calculate the instantaneous phase angles corresponding to the discrete time sampling points, and generate a phase angle sequence;

[0148] The phase angle sequence is input into a preset nonlinear mapping function to obtain the frequency modulation coefficient sequence corresponding to the discrete time sampling points;

[0149] Based on the frequency modulation coefficient sequence, the pulse period of the square wave pulse sequence is scaled periodically to reconstruct and generate a biomimetic frequency modulation pulse sequence.

[0150] The biomimetic frequency-modulated pulse sequence is output to the skin location corresponding to the target physical coordinates.

[0151] Specifically, the execution module receives the underlying driver data frame packaged in the previous step and parses out the 3x1 dimensional pulse control parameters and the target physical displacement components. Based on the amplitude of the limiting pulse current and the width of the electrical stimulation pulse in the pulse control parameters, combined with the constant reference pulse period calculated from the electrical stimulation pulse frequency, the system generates an initial equiperiodic symmetrical biphasic square wave pulse sequence in the underlying hardware driver. Simultaneously, the system maps the target physical displacement components to the specific row and column physical coordinates of the hardware relay gating matrix according to a preset spatial scale, establishing a dedicated output routing channel to the corresponding electrode contact.

[0152] To ensure physiological coordination between electrical stimulation and the patient's own respiratory rhythm, the system analyzes the multimodal feature vectors continuously received by the control module in parallel, extracting the thoracic impedance signal sequence as the respiratory phase feature. The system performs symmetrical mirror extension at both ends of this feature sequence to eliminate boundary distortion effects caused by the finite sequence length. Subsequently, it performs a discrete Hilbert transform, and calculates the four-quadrant arctangent with quadrant compensation by combining the signs of the imaginary and real parts, thereby obtaining the instantaneous phase angles corresponding to discrete-time sampling points and generating a phase angle sequence. The calculation formula is as follows:

[0153] ;

[0154] in the formula Indicates the first The instantaneous phase angle scalar of each discrete-time sampling point takes a value ranging from negative pi to positive pi, and the unit is radians; The index number of the sampling point in the discrete-time series is a positive integer; Represents the first eigenvector in a multimodal feature vector. The real component of the thoracic impedance amplitude at each sampling point is a scalar quantity with the physical dimension of ohms. express The imaginary component scalar obtained after the discrete Hilbert transform has the physical dimension of ohms. This represents the bivariate four-quadrant arctangent function. This function automatically compensates for quadrant differences by independently identifying the positive and negative signs of the imaginary and real components, thus accurately mapping the complete respiratory cycle. The interval between 0 and positive pi represents the impedance decrease period caused by thoracic volume recoil, physiologically corresponding precisely to the expiratory phase. The system extracts discrete-time sampling points falling within this interval as the time enable window for safe stimulus delivery.

[0155] After extracting discrete-time sampling points representing the expiratory phase, the system inputs the aforementioned phase angle sequence into a preset nonlinear mapping function to obtain the frequency modulation coefficient sequence corresponding to each discrete-time sampling point. The mapping formula is as follows:

[0156] ;

[0157] in the formula Indicates the first The frequency modulation coefficient corresponding to each discrete-time sampling point is a scalar, dimensionless; The index number of the sampling point in the discrete-time series is a positive integer; This indicates that the first term calculated by the aforementioned formula is... The instantaneous phase angle scalar of a discrete-time sampling point, in radians; This represents the system's preset modulation depth coefficient, which is set to 0.25 here. Its physiological basis is to ensure that the pulse frequency is continuously increased by a maximum of 25% to closely match the actual firing acceleration characteristics of motor neurons when swallowing reaches its peak during the pharyngeal phase. This represents the sine function.

[0158] Based on the generated frequency modulation coefficient sequence, the system dynamically scales the pulse period of the initial square wave pulse sequence period by period to reconstruct a biomimetic frequency-modulated pulse sequence. The period scaling formula is as follows:

[0159] ;

[0160] in the formula Indicates the number after dynamic scaling. A scalar measure of the instantaneous duration of each pulse cycle, in seconds; The index number of the sampling point in the discrete-time series is a positive integer; This indicates that the first term derived from the aforementioned formula is... The frequency modulation coefficient corresponding to each discrete-time sampling point is a scalar, dimensionless; This represents a scalar value, in Hertz, representing the reference value of the electrical stimulation pulse frequency derived from the pulse control parameters. Through the above modulation, the system forms a smooth peak in the stimulation frequency during mid-exhalation. The system then outputs a biomimetic frequency-modulated pulse sequence, including the dynamic cycle duration, the amplitude of the original limited pulse current, and the width of the electrical stimulation pulse, to the skin surface corresponding to the target physical coordinates through the established output routing channel to apply stimulation.

[0161] After the biomimetic frequency-modulated pulse sequence output ends, the system acquires surface electromyographic (EMG) signals of the submental muscle group via an independent data acquisition channel for a duration of 2000 milliseconds. This duration covers the average physiological time of 1500 milliseconds for a single complete pharyngeal swallowing action, with added safety redundancy. The system performs full-wave rectification on the EMG signal and extracts a smooth signal envelope using a low-pass filter with a cutoff frequency of 5 Hz. The system calls upon the mean and standard deviation of the resting baseline, pre-recorded during the initial device calibration phase in global shared memory, to calculate the time difference between the start and end points where the envelope amplitude continuously exceeds the mean of the resting baseline plus three times the standard deviation, as the contraction duration; simultaneously, it extracts the maximum absolute value of the envelope within this time period as the peak amplitude. The system matrixes the extraction results to construct swallowing physiological feedback data, the encapsulation formula of which is:

[0162] ;

[0163] in the formula This represents a 2x1 dimensional column vector of swallowing physiological feedback data constructed by the system. This represents a scalar value indicating the duration of contraction of the extracted submental muscle group, in milliseconds. This represents the peak amplitude scalar value of the extracted submental muscle group signal, in microvolts; the upper right corner... The symbol indicates that the inline elements are transposed into a column vector format.

[0164] This enabled the biomimetic dynamic modulation output of the stimulation frequency according to the phase of spontaneous exhalation, and accurately quantified the real biomechanical response characteristics of the target muscle group to electrical stimulation.

[0165] The optimization module calculates the deviation value of the evaluation index and generates the parameter correction amount based on the swallowing physiological feedback data, and updates the individualized intensity benchmark based on the parameter correction amount.

[0166] Furthermore, in the optimization module, the deviation value of the evaluation index is calculated and the parameter correction amount is generated. The update of the individualized intensity benchmark based on the parameter correction amount specifically includes the following steps:

[0167] The difference between the swallowing physiological feedback data and the preset baseline characteristic data is calculated as the evaluation index deviation value;

[0168] The deviation value of the evaluation index is input into the proportional-integral-derivative controller to obtain the parameter correction amount;

[0169] The parameter correction amount is algebraically superimposed with the individualized strength benchmark to obtain the updated individualized strength benchmark.

[0170] Specifically, the optimization module continuously receives a 2x1 dimension swallowing physiological feedback data column vector uploaded by the execution module after the end of the previous electrical stimulation output cycle. To quantify the effectiveness of the current electrical stimulation inducing the target muscle response, the system reads preset baseline feature data from global read-only memory. This data is manually entered into the system by clinicians during the initial treatment planning phase based on the patient's swallowing imaging assessment results, and specifically includes two scalars: target contraction duration and target peak amplitude. The system calculates the relative deviations between the two actual extracted indicators in the current swallowing physiological feedback data and the corresponding preset baseline indicators, and obtains a single evaluation indicator deviation value through weighted fusion calculation. The calculation formula is as follows:

[0171] ;

[0172] In the formula, Indicates the first The dimensionless evaluation index deviation value scalar calculated from each treatment cycle; Represents a positive integer representing the ordinal number of the current treatment cycle; This represents a scalar value, in milliseconds, representing the duration of the submental muscle contraction parsed from the preceding data stream. This represents the system's preset target contraction duration scalar value. It is a real number that is strictly greater than zero to prevent triggering a division-by-zero exception, and the unit is milliseconds. This represents the peak amplitude scalar value of the submental muscle group signal parsed from the preceding data stream, in microvolts. This represents the system's preset target peak amplitude scalar, a real number that is strictly greater than zero, and its unit is microvolts; and The dimensionless weighting coefficients representing the contraction duration and peak amplitude are set to 0.4 and 0.6 respectively, based on the inventors' statistical conclusions from their electromyographic response calibration experiments on patients with swallowing disorders. When the deviation value is greater than zero, it indicates that the current muscle response is weak and has not reached the expected target; when the deviation value is less than zero, it indicates that the current muscle contraction response is over-recruited; when the deviation value is equal to zero, it indicates that the muscle response is just within the target range and the system does not make incremental intervention.

[0173] After obtaining the deviation value of the evaluation index, the optimization module inputs it into a preset incremental proportional-integral-derivative controller. The system uses an incremental structure instead of a positional structure because the incremental algorithm only outputs the difference in the change of the control quantity, naturally possessing anti-integral saturation characteristics, which can avoid the danger of sudden current changes caused by state switching during medical electrical stimulation. This controller combines the deviation evolution trend between the current cycle and the previous two historical cycles to output parameter correction amounts for adjusting the stimulation intensity. For boundary cases where historical deviation values ​​are missing in the first and second cycles after treatment begins, the system automatically initializes the missing historical deviation values ​​to zero to ensure a smooth start-up of the calculation. The calculation formula is:

[0174] ;

[0175] In the formula, Indicates the first The parameter correction scalar value output by the controller for each treatment cycle, in milliamperes; Indicates the current number The dimensionless evaluation index deviation value scalar for each cycle; and These represent the dimensionless evaluation index deviation values ​​between the previous and previous two cycles in the global memory cache, respectively. This represents the proportional gain coefficient of the controller, set to 0.5, with units of milliamperes. This represents the integral gain coefficient scalar, set to 0.15, with units in milliamperes; The differential gain coefficient scalar is set to 0.05, with units in milliamperes. The physical dimension configuration of the above three gain coefficient scalars is used to convert the dimensionless input deviation into a current control dimension, and the specific values ​​are all empirical safety boundary values ​​tuned in multiple rounds of clinical electrical stimulation trials based on the critical proportionality method.

[0176] After obtaining the parameter correction amount, the system calls the original individualized intensity benchmark currently stored in the global shared memory, and algebraically superimposes the parameter correction amount with the original individualized intensity benchmark to complete the closed-loop update of the stimulus parameters. The update formula is as follows:

[0177] ;

[0178] In the formula, This represents the updated individualized intensity baseline scalar value available for use in the next treatment cycle, in milliamperes. This represents the original individualized strength baseline scalar used in the current cycle and initially written to the global shared memory by the upstream step, in milliamperes; This represents the parameter correction scalar calculated in this cycle, in milliamperes; constants 1 and 20 represent the lower and upper bound constant scalars of the system's hard limiting circuit, respectively, both in milliamperes. and These represent the functions for maximizing and minimizing the values, respectively. These two functions, combined with a limiting constant, constitute a hard limiting circuit. The lower limit of the safe range (1 mA) and the upper limit (20 mA) are strictly determined according to the specific safety standards for nerve and muscle stimulators in medical electrical equipment, ensuring that the superimposed current is limited within the physiologically tolerable range. After the electrical stimulation pulse output of the current treatment cycle has completely ended, the system immediately overwrites the updated value to the original address in the global shared memory, so that this updated value only applies to the next completely new trigger cycle.

[0179] This step enables closed-loop adaptive fine-tuning of electrical stimulation intensity based on the actual muscle response, maintaining the stability of swallowing intervention effects during long-term treatment.

[0180] Example 2:

[0181] Most electrical stimulation devices use fixed-frequency or simple trigger signal output methods, lacking dynamic sensing and matching of the patient's real-time respiratory status. This can lead to a mismatch between the electrical stimulation trigger time and the respiratory phase, potentially causing overlap of swallowing and inhalation processes in some cases, increasing the probability of choking or aspiration, and also affecting the comfort and stability of the overall intervention process. To address these issues, this invention provides a percutaneous vagus nerve stimulation method for treating post-stroke dysphagia, such as... Figure 2 As shown. Includes the following steps:

[0182] S1. Collect skin impedance distribution information in the concha region and determine the target physical coordinates based on the skin impedance distribution information;

[0183] S2. Based on the target physical coordinates, obtain the sensory threshold and pain threshold of the test object to electrical stimulation, and calculate and generate an individualized intensity benchmark.

[0184] S3. After establishing the individualized intensity benchmark, the submental muscle group signals and respiratory waveforms of the subject to be tested are collected simultaneously and mapped to form a multimodal feature vector;

[0185] S4. Based on multimodal feature vectors, swallowing action recognition is performed and associated with the target physical coordinates to generate trigger commands;

[0186] S5. Based on the trigger command, current limiting is performed through loop impedance monitoring, and pulse control parameters are output.

[0187] S6. Based on pulse control parameters, drive the electrodes to apply electrical stimulation signals corresponding to swallowing actions and obtain swallowing physiological feedback data.

[0188] S7. Based on swallowing physiological feedback data, calculate the deviation value of the evaluation index and generate the parameter correction amount, and update the individualized intensity benchmark based on the parameter correction amount.

[0189] Specifically, the system first collects skin impedance distribution information in the concha region using a miniature probe array, and then uses a minimum value optimization algorithm to locate the point of lowest impedance as the target physical coordinate. Next, based on this coordinate, a test pulse is applied to extract the sensory and pain thresholds of the subject, and an individualized intensity benchmark with a safety buffer is calculated and stored in global memory. After establishing the benchmark, the system simultaneously collects submental electromyography and thoracic respiratory impedance signals, which are preprocessed, spliced, and mapped into a multimodal feature vector sequence. Then, a decoding model is used to analyze the feature sequence to identify the initiation of swallowing, and the identification result is associated with the target coordinate and encapsulated to generate a trigger command. Upon receiving the command, the system monitors the loop impedance in real time, performs over-limit peak clipping and current limiting, combines these to generate pulse control parameters, and maps them... A dedicated hardware output route corresponding to the physical coordinates is established. To achieve physiological coordination, the system extracts the respiratory impedance sequence from the multimodal features, obtains the instantaneous phase angle through signal transformation to lock the expiratory phase window, and dynamically modulates the pulse frequency to generate a biomimetic frequency-modulated pulse for output. After stimulation, the electromyographic envelope is collected to extract the contraction duration and peak amplitude as swallowing physiological feedback data. Finally, the system calculates the weighted relative deviation between the actual feedback data and the preset target baseline, inputs it into an incremental proportional-integral-derivative controller to obtain the parameter correction amount, algebraically superimposes the correction amount with the original intensity benchmark, and performs hard limiting processing based on the physiological safety threshold. Then, it immediately overwrites and updates the individualized intensity benchmark for the next cycle, thereby completing the closed-loop adaptive fine-tuning of the stimulation parameters.

[0190] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, characterized in that, include: The positioning module collects skin impedance distribution information in the concha region and determines the target's physical coordinates based on the skin impedance distribution information. The titration module, based on the target physical coordinates, obtains the sensory threshold and pain threshold of the test object to electrical stimulation, and calculates and generates an individualized intensity benchmark. After establishing the individualized intensity benchmark, the signal module synchronously acquires the submental muscle group signals and respiratory waveforms of the subject under test, and maps them to form a multimodal feature vector; The control module, based on the multimodal feature vector, performs swallowing action recognition and associates and encapsulates it with the target physical coordinates to generate trigger commands; The verification module, based on the trigger command, performs current limiting processing through loop impedance monitoring and outputs pulse control parameters. The execution module, based on the pulse control parameters, drives the electrodes to apply an electrical stimulation signal corresponding to the swallowing action and acquires swallowing physiological feedback data; The optimization module calculates the evaluation index deviation value and generates parameter correction amount based on the swallowing physiological feedback data, and updates the individualized intensity benchmark based on the parameter correction amount.

2. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 1, is characterized in that, In the positioning module, determining the target's physical coordinates based on the skin impedance distribution information specifically includes the following steps: Construct a two-dimensional topological map of skin impedance in the concha region; Extract the impedance valley points from the two-dimensional topology map of the skin impedance; The impedance valley points are mapped to a preset coordinate system to generate target physical coordinates.

3. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 1, is characterized in that... In the titration module, the process of obtaining the sensory threshold and pain threshold of the test subject to electrical stimulation and calculating and generating an individualized intensity benchmark specifically includes the following steps: A step current is applied to the skin position corresponding to the target physical coordinates, and the current value that induces a sensory response is recorded as the sensory threshold. Continue applying step current and record the current value that induces pain response as the pain threshold; The difference between the pain threshold and the sensation threshold is calculated, and the difference is multiplied by a preset adjustment coefficient and added to the sensation threshold to generate an individualized intensity benchmark.

4. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 1, is characterized in that... In the signal module, the synchronous acquisition of the submental muscle group signal and respiratory waveform of the subject under test, and the mapping to form a multimodal feature vector, specifically includes the following steps: Extract the temporal envelope features and frequency energy features of the submental muscle group signal; Extract the respiratory phase features of the respiratory waveform; The time-domain envelope feature, the frequency-domain energy feature, and the breathing phase feature are matrix-concatenated to construct a multimodal feature vector.

5. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 1, characterized in that, In the control module, the process of recognizing swallowing actions based on the multimodal feature vectors, associating and encapsulating them with the target's physical coordinates, and generating trigger commands specifically includes the following steps: The multimodal feature vectors are input into a preset classification model to determine the initiation time of the swallowing action; Extract the timestamp corresponding to the startup time; The timestamp and the target physical coordinates are encapsulated to generate a trigger command.

6. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 1, is characterized in that... In the verification module, the current limiting process through loop impedance monitoring and the output pulse control parameters specifically include the following steps: In response to the trigger command, the loop impedance value of the interface between the electrode and the skin is acquired; Calculate the difference between the loop impedance value and the preset upper limit threshold impedance value; The individualized intensity benchmark is attenuated and modulated based on the comparison difference to generate pulse control parameters.

7. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 1, is characterized in that... In the execution module, the driving electrode applies an electrical stimulation signal corresponding to the swallowing action, and the acquisition of swallowing physiological feedback data specifically includes the following steps: The pulse control parameters are analyzed to generate a square wave pulse sequence; The multimodal feature vector is analyzed to extract respiratory phase features. During the exhalation phase, the square wave pulse sequence is output to the skin position corresponding to the target physical coordinates. The contraction duration and peak amplitude of the submental muscle group signals after stimulation were extracted to construct swallowing physiological feedback data.

8. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 7, is characterized in that... The step of outputting the square wave pulse sequence to the skin location corresponding to the target physical coordinates during the expiration phase of the respiratory phase specifically includes the following steps: Extract discrete time sampling points characterized during the expiratory phase, calculate the instantaneous phase angle corresponding to the discrete time sampling points, and generate a phase angle sequence; The phase angle sequence is input into a preset nonlinear mapping function to obtain a frequency modulation coefficient sequence corresponding to the discrete time sampling point; Based on the frequency modulation coefficient sequence, the pulse period of the square wave pulse sequence is scaled periodically to reconstruct and generate a biomimetic frequency modulation pulse sequence. The biomimetic frequency-modulated pulse sequence is output to the skin position corresponding to the target physical coordinates.

9. A transcutaneous vagus nerve stimulation system for treating dysphagia after stroke, as described in claim 1, characterized in that, In the optimization module, the calculation of the evaluation index deviation value and the generation of parameter correction amount, and the updating of the individualized intensity benchmark based on the parameter correction amount specifically include the following steps: The difference between the swallowing physiological feedback data and the preset baseline feature data is calculated as the evaluation index deviation value; The deviation value of the evaluation index is input into the proportional-integral-derivative controller to obtain the parameter correction amount; The parameter correction amount is algebraically superimposed with the individualized strength benchmark to obtain the updated individualized strength benchmark.

10. A percutaneous vagus nerve stimulation method for treating dysphagia after stroke, characterized in that, A percutaneous vagus nerve stimulation system for treating post-stroke dysphagia as described in any one of claims 1-9 comprises the following steps: S1. Collect skin impedance distribution information in the concha region, and determine the target physical coordinates based on the skin impedance distribution information; S2. Based on the target physical coordinates, obtain the sensory threshold and pain threshold of the test object to electrical stimulation, and calculate and generate an individualized intensity benchmark. S3. After establishing the individualized intensity benchmark, simultaneously collect the submental muscle group signals and respiratory waveforms of the subject to be tested, and map them to form a multimodal feature vector; S4. Based on the multimodal feature vector, perform swallowing action recognition and associate it with the target physical coordinates to generate a trigger command; S5. Based on the trigger command, current limiting is performed through loop impedance monitoring, and pulse control parameters are output. S6. Based on the pulse control parameters, apply an electrical stimulation signal corresponding to the swallowing action to the driving electrode to obtain swallowing physiological feedback data. S7. Based on the swallowing physiological feedback data, calculate the evaluation index deviation value and generate parameter correction amount, and update the individualized intensity benchmark based on the parameter correction amount.

Images