Swallowing bionic electrical stimulation time sequence generation method based on multi-mode BCI and storage medium

Abstract

The invention provides a swallowing electrical stimulation time sequence generation method based on multi-mode BCI and bionic time sequence electrical stimulation, and the method comprises the steps: selecting a first swallowing electrical stimulation sequence template from a standard physiological swallowing muscle activation time sequence model based on the individual information of a patient, and obtaining a second swallowing electrical stimulation sequence template through a reinforcement learning algorithm; the swallowing intention of the patient is recognized through a multi-mode BCI, and according to the second swallowing electrical stimulation sequence template, an electrical stimulation instruction is sent to a plurality of target muscle groups according to a preset time sequence; and optimizing and updating parameters of the second swallowing electrical stimulation sequence template by using a reinforcement learning algorithm to obtain a third swallowing electrical stimulation sequence template. Through a personalized treatment scheme, real-time feedback and self-adaptive adjustment, accurate swallowing intention recognition and continuous optimization of the treatment effect, the problems existing in traditional swallowing rehabilitation treatment are solved, the treatment effect is remarkably improved, and the rehabilitation speed of a patient is remarkably increased.

Description

Technical Field

[0001] This invention relates to the field of brain-computer interface (BCI) and intelligent medical rehabilitation, specifically to a method for generating a swallowing bionic electrical stimulation timing sequence based on multimodal BCI. Background Technology

[0002] Many diseases can cause swallowing disorders, especially for patients with MND (Multiple Diseases of the Narrow-Range), where difficulty swallowing can lead to potentially fatal complications. Traditional swallowing rehabilitation methods have limited effectiveness, particularly neuromuscular electrical stimulation (NMES), which uses fixed parameters and patterns to stimulate peripheral muscles. Its core flaws are: disconnect from central nervous system intentions; stimulation is triggered by the device at set times, not by the patient's swallowing intent, thus failing to create closed-loop neural remodeling. The stimulation patterns are non-physiological, often employing synchronous stimulation, which contradicts the sequential and coordinated contraction of normal swallowing muscles, resulting in limited effectiveness and potentially exacerbating incoordination. Furthermore, the parameters are fixed and cannot be adaptively adjusted based on individual patient differences and disease progression.

[0003] While existing BCI technologies attempt to trigger stimulation with "thought," most only achieve a simple "on / off" function, failing to address the deeper issue of "how to stimulate more precisely and in accordance with physiological laws." This invention is based on a profound understanding of the physiological and pathological mechanisms of swallowing, utilizing AI technology to transform this knowledge into executable, personalized, and precise stimulation programs. Although existing BCI technologies attempt to trigger stimulation with "thought," most only achieve a simple "on / off" function, and their application paradigm remains limited to the "trigger-rehabilitation" model, which assumes that neural pathways can be reconstructed through repeated intention-stimulus pairing training. However, for patients with progressive and irreversible neurological diseases such as motor neuron disease (MND), their swallowing function cannot be truly restored, and traditional rehabilitation training models have limited effectiveness. These patients urgently need a device that can "replace" or "enhance" their damaged neural function, providing safe and effective swallowing assistance immediately when needed, without demanding permanent restoration of neural function. The starting point of this invention is based on this clinical need, aiming to develop a biomimetic "neural prosthesis" or "functional assistive device". Its core is to transform clinicians' knowledge of the physiological mechanism of swallowing into an executable, personalized, and precise stimulation program, providing the patient with a "perfect" biomimetic muscle contraction sequence when the patient initiates the intention, thereby compensating for the lost function. Summary of the Invention

[0004] This invention aims to address the problems of disconnect between stimulation patterns and physiological timing, and lack of personalization in existing technologies. It provides a method for generating timing sequences for swallowing biomimetic electrical stimulation based on multimodal BCI, characterized by comprising: A standard physiological swallowing muscle activation timing model was established based on the surface electromyography (sEMG) signals of swallowing-related muscles in healthy individuals. Based on individual patient information, a first swallowing electrical stimulation sequence template is selected from the standard physiological swallowing muscle activation timing model. The physiological signals of the patient using the first swallowing electrical stimulation sequence template are collected in real time, and the second swallowing electrical stimulation sequence template is obtained through a reinforcement learning algorithm; The patient's swallowing intention is identified through multimodal BCI, and electrical stimulation commands are sent to multiple target muscle groups according to a predetermined time sequence based on the second swallowing electrical stimulation sequence template. The patient's swallowing effect feedback data using the second swallowing electrical stimulation sequence template is collected, and the parameters of the second swallowing electrical stimulation sequence template are optimized and updated using a reinforcement learning algorithm to obtain a third swallowing electrical stimulation sequence template.

[0005] In a second aspect, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions for causing a computer to execute the above-described method for generating swallowing biomimetic electrical stimulation timing based on multimodal BCI.

[0006] A key innovation of this invention lies in its dual nature. For patients with recoverable function (such as those in the early stages of stroke): the system is a rehabilitation training system. By repeatedly providing precise and physiologically consistent "intention-feedback" loops, it promotes neural remodeling in the brain, with the ultimate goal of enabling patients to perform effective swallowing independently even after weaning off the device. For patients with irreversible functional impairment (such as MND), the system is a functional aid device. Its design aims to compensate for impaired swallowing function, much like eyeglasses correct vision. When in use, the device is triggered by "thought," immediately providing a sequence of biomimetic electrical stimulation to strengthen and take over the fine control of the throat muscles, thereby safely completing the swallowing action; when not in use, the stimulation stops, and the patient returns to their baseline state. Its core value lies in the "immediate effect" of functional improvement, rather than pursuing permanent functional rehabilitation, providing a new way for patients in the middle and late stages of MND and other conditions to maintain oral feeding and improve their quality of life.

[0007] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart of a swallowing biomimetic electrical stimulation timing generation method based on multimodal BCI provided in this disclosure embodiment.

[0010] Figure 2 A flowchart illustrating the method for constructing a physiological swallowing muscle activation timing model provided in this embodiment of the disclosure.

[0011] Figure 3 A flowchart illustrating a method for obtaining the optimal personalized stimulus scheme using a reinforcement learning framework, as provided in this embodiment of the disclosure.

[0012] Figure 4 The flowchart illustrates the closed-loop feedback and adaptive learning method for personalized time-series templates of final optimization parameters provided in this embodiment of the disclosure.

[0013] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. Detailed Implementation

[0014] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0015] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0016] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0017] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0018] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0019] For detailed implementation procedures, please refer to the appendix. Figure 1 This invention provides a method for generating swallowing biomimetic electrical stimulation timing based on multimodal BCI, comprising the following steps: Step S1: Collect surface electromyographic signals of the main swallowing muscles in healthy individuals when swallowing foods of different shapes, and construct a physiological swallowing muscle activation time sequence model.

[0020] Surface electromyography (sEMG) signals of major muscles, including the mylohyoid, anterior belly of the digastric, geniohyoid, thyrohyoid, and thyroarytenoid muscles, were collected from healthy subjects during the swallowing of foods of different shapes (liquid, paste, and solid). Algorithms were used to analyze the onset time, peak time, duration, and temporal relationship of each muscle activation, forming a standard physiological swallowing muscle activation timing model. Specifically: Healthy participants aged 18-60 were strictly selected according to stratification requirements. Eligible participants were divided into three groups: 18-30 years old (youth group), 31-45 years old (middle-aged group), and 46-60 years old (middle-aged and elderly group). This ensured the template covered the differences in muscle function across different age groups (e.g., muscle activation speed in the middle-aged and elderly group is approximately 10%-15% slower than in the youth group), minimizing interference from individual differences on the template. For detailed implementation methods, please refer to [link to implementation details]. Figure 2 The following is combined with Figure 2 Explain in detail how to implement step S1.

[0021] Step S110: Collect sEMG signals of the mylohyoid muscle, anterior belly of the digastric muscle, geniohyoid muscle, thyrohyoid muscle and thyroarytenoid muscle, as well as collect laryngeal acceleration and swallowing sound signals.

[0022] A medical-grade 8-channel sEMG acquisition instrument (such as the Noraxon MyoSystem 1400) was selected to acquire the sEMG signals. The sampling frequency was set to 2000Hz, and Ag / AgCl surface electrodes were used. The sEMG electrodes were precisely positioned, attached, and fixed to the mylohyoid muscle, the anterior belly of the digastric muscle, the geniohyoid muscle, the thyrohyoid muscle, and the thyroarytenoid muscle.

[0023] A miniature triaxial accelerometer and a directional medical microphone were used to collect laryngeal acceleration and swallowing sound signals. The miniature triaxial accelerometer extracted the y-axis waveform of the laryngeal acceleration signal to define the swallowing cycle, obtaining the moment when the y-axis signal at the starting point (t1) rises from the baseline, the moment when the y-axis signal at the peak point (T0) reaches its maximum value, and the moment when the y-axis signal at the ending point (t2) falls back to the baseline.

[0024] From the continuously acquired sEMG and swallowing sound signals, the signal segment from t1-50ms to t2+50ms (with a reserved transition area) is extracted, excluding the resting signal between two swallows. The extracted swallowing sound signals are examined; if they contain the complete "three phases" (oral phase 0.1-0.3s / 500-800Hz, pharyngeal phase 0.3-0.8s / 800-1200Hz, esophageal phase 0.8-1.2s / 200-500Hz) and the total duration is 1.0-1.5s, it is considered a valid swallow; if it contains only a single phase or the duration is abnormal, the corresponding sEMG signal is simultaneously removed. Step S120: Preprocess the EMG signal to extract the smooth sEMG envelope.

[0025] A fourth-order Butterworth bandpass filter (passband 10-500Hz) is used to remove low-frequency noise (electrodermal signals) below 10Hz and high-frequency noise (electronic interference) above 500Hz, while retaining the effective electromyographic components in the 20-400Hz range; a fourth-order low-pass filter (cutoff frequency 50Hz) is used to filter high-frequency interference (such as neck muscle tremors) other than ≤10Hz laryngeal movement signals; and a fourth-order bandpass filter (passband 200-2000Hz) is used to remove respiratory sounds below 200Hz and environmental noise above 2000Hz, highlighting the main swallowing sound energy frequency band of 300-1800Hz. The db4 wavelet basis was used to decompose the sEMG signal into five levels (frequency ranges: 250-500Hz, 125-250Hz, 62.5-125Hz, 31.25-62.5Hz, and 15.625-31.25Hz). The db4 wavelet can be used to decompose the signal into different frequency components and extract information within specific frequency ranges through reconstruction. Due to its tight support and orthogonality, the db4 wavelet performs excellently in removing noise from the signal while preserving its main features. The Birgé-Massart threshold criterion was used to denoise the high-frequency layer (250-500Hz), improving the signal-to-noise ratio by 15%-20%. The decomposed sEMG signal was then subjected to full-wave rectification and smoothed using a 50ms moving average window to obtain a smoothed sEMG envelope. The Birgé-Massart threshold criterion is a method for threshold selection in wavelet transform, primarily used for signal denoising. This method can effectively remove noise while preserving the main features of the signal by adaptively selecting the threshold.

[0026] Step S130: Based on the smooth sEMG envelope, perform muscle activation time series analysis and inter-muscle time series relationship analysis. Based on the sEMG envelope and the peak point (T0) of the laryngeal acceleration signal on the y-axis, three key parameters were extracted from the mylohyoid muscle, the anterior belly of the digastric muscle, the geniohyoid muscle, the thyrohyoid muscle, and the thyroarytenoid muscle: (1) Activation start time (tstart): the first time the envelope exceeds the activation start time. For example, if the envelope exceeds the threshold 80ms before T0, then... (2) Peak time ( ):exist The envelope line fell back to Within a given time period, find the global maximum value of the envelope (calculate the average of 5 adjacent data points, excluding local small peaks); the corresponding time is [the value of the envelope]. (3) Duration ( ):from The envelope line fell back to its last point. Time difference, for example: =T0-80ms, fallback time=T0+120ms, then =200ms. Wherein, σ represents the standard deviation of the sEMG signal in healthy subjects at rest for 10 seconds. At rest, human muscles do not actively contract, but muscle fibers still exhibit minute random electrical activity (referred to as "resting electromyographic noise"). Simultaneously, interference from skin conductance and device current noise may be introduced during signal acquisition. A larger σ value indicates a wider range of fluctuations in these random noises; a smaller σ value indicates a more stable resting signal and less noise interference.

[0027] Normalize the sEMG envelopes of any two groups of target muscles (mean to zero, standard deviation to 1) to eliminate bias caused by differences in signal amplitude (e.g., the sEMG amplitude of the anterior belly of the digastric muscle is larger than that of the thyroarytenoid muscle). Calculate the cross-correlation function of the envelopes of the two muscle groups. ,in The normalized sEMG envelope is given, where T represents the duration of swallowing. For time delay, the peak value of cross-correlation corresponds to This refers to the activation delay. For example, the activation delay of the mylohyoid muscle and the anterior belly of the digastric muscle. =20ms, indicating that the mylohyoid muscle is activated 20ms earlier than the anterior belly of the digastric muscle. Step S140: Generate a standard timing template based on muscle activation timing analysis and intermuscular timing relationship analysis.

[0028] The K-means clustering algorithm (K=3, corresponding to three food traits) was used to analyze the time-series parameters of all healthy subjects. , , , Perform clustering. Output three clustering results, corresponding to the swallowing timing parameter sets for liquid, pasty, and solid foods, respectively. Calculate the mean of the timing parameters for all samples in each cluster. with standard deviation If any parameter of a sample deviates (like If the duration of the abnormal delay is extended to 500ms, it is identified as an abnormal sample and removed.

[0029] Ultimately, three standard "physiological swallowing muscle activation timing models" (Template liquid; Template paste; Template solid) were generated, forming a normal human swallowing timing template library. Taking Template paste as an example, the following physiological swallowing muscle activation timing model was generated.

[0030]

[0031] Step S2: Based on the patient's pathological information and physiological signals, select the closest initial timing template from the normal person swallowing timing template library, and obtain the optimal personalized stimulation scheme through the reinforcement learning framework.

[0032] Acquire patient physiological information, including laryngeal elevation amplitude, swallowing sounds, and upper esophageal sphincter (UES) opening degree. Convert the physiological information into quantitative values ​​to obtain a patient quantitative index vector. For example, the degree of laryngeal elevation. Level 1 =0.25, Level 2 =0.5, Level 3 =0.75, Level 4 =1.0.

[0033] The weighted Euclidean distance model is used to select the closest initial time sequence template from the normal human swallowing time sequence template library.

[0034] Based on patient quantitative index vector The differences between the template values ​​and the corresponding indicator values ​​of healthy individuals in the standard template library are calculated, and the template with the smallest difference is selected as the initial template. The weighted Euclidean distance model is used to calculate the difference value; the specific formula is as follows: ; in: The weighted Euclidean distance between the patient and the k-th standard template. Let i be the weight of the i-th indicator. The i-th quantitative indicator of the patient, The ideal value (1.0) is the i-th index of the k-th template.

[0035] Physiological signals are collected in real time using an initial time-series template, and the optimal personalized stimulation scheme is obtained through a reinforcement learning framework.

[0036] Following step S110, sEMG signals from the mylohyoid, anterior digastric, geniohyoid, thyrohyoid, and thyroarytenoid muscles were acquired, along with laryngeal acceleration and swallowing sound signals. The acquired raw signals were preprocessed (same as the filtering and denoising in step S1002) to extract key features as input for reinforcement learning. Key features of the raw signals included at least the target muscle sEMG, non-target muscle sEMG, laryngeal acceleration, swallowing sounds, and clinical assessment.

[0037] A Deep Deterministic Policy Gradient (DDPG) approach is used to optimize the optimal personalized stimulation scheme. Key features of the original signal acquired using an initial temporal template are used as input signals, and the output is the optimized electrical stimulation parameters for the personalized stimulation scheme. These optimized electrical stimulation parameters include the stimulation time difference for a specific muscle. ,strength ,frequency The adjustment value.

[0038] In the Deep Deterministic Policy Gradient (DDPG) algorithm, the Actor network and the Critic network are two core components that work together to optimize the policy. The Actor network is a policy network that directly maps a state `s` to a deterministic action `a`. Its goal is to learn an optimal policy. The goal is to maximize the cumulative reward by choosing action 'a' given a state 's'. In DDPG, the Actor network generates actions. It outputs an action 'a' based on the current state 's', which is used to interact with the environment to obtain rewards and the next state. The Critic network is a value network that evaluates the value Q(s,a) of taking action 'a' given state 's'. Its goal is to learn an accurate Q-function to evaluate the quality of the current policy. The Critic network provides feedback to the Actor network. By evaluating the value of each state-action pair under the current policy, it helps the Actor network better adjust its policy.

[0039] Both the Actor and Critic networks in the deep deterministic policy gradient model employ a 3-layer fully connected neural network (FC), using ReLU (in the hidden layers) and specific output activation functions (ensuring the parameters are within a valid range). The learning rate for both the Actor and Critic networks is 0.0001. The experience replay pool has a capacity of 10000 and is used to store State-Action-Reward-NextState samples. The target network's soft update rate is... =0.001. Noise was explored using the Ornstein-Uhlenbeck procedure.

[0040] For details on how to obtain the optimal personalized stimulus scheme using a reinforcement learning framework, please refer to [link to relevant documentation]. Figure 3 Specifically, it includes the following steps: Step S231: Load the initial template parameters as the current parameters. , ,in The current stimulation time difference is the time difference between the start time of electrical stimulation and the start time of spontaneous activation of the patient's swallowing muscles, with an initial value of [value missing]. =0ms means that electrical stimulation and spontaneous muscle activation are initiated synchronously. This ensures that electrical stimulation can precisely match the natural contraction rhythm of the muscle, avoiding premature stimulation that could cause muscle disorder, or delayed stimulation that could fail to assist contraction. The current intensity of the electrical stimulation applied to the muscle is the magnitude of the electrical current, with an initial value of [value missing]. =3mA. By activating the target muscle with an appropriate current intensity, 3mA will not cause discomfort to the patient (such as soreness) due to excessive strength, and can also generate a weak signal to help monitor the muscle response, laying the foundation for subsequent adjustments. The current stimulation frequency is the pulse firing frequency of the electrical stimulation, and the initial value is... =50Hz is close to the optimal contraction frequency of swallowing muscles, which can generate continuous contraction force while avoiding muscle fatigue, and assist in the smooth swallowing of food boluses. At the same time, the Actor / Critic network weights are initialized (He normal initialization) and an experience replay pool is created.

[0041] Step S232: Perform Episode loop training. Obtain real-time signal characteristics of the patient ( The current parameters, together with the current parameters, form the State vector. State vector S t It consists of two parts: one is the patient's real-time physiological signal characteristics, specifically including To target muscle activation intensity, The ratio of laryngeal elevation to the standard value; C is the cough marker. The first is to score the food bolus residue. The second is the current electrical stimulation parameters ( , This is used by the Actor network to determine the current situation and output adjustment actions. The Actor network, based on... Output parameter adjustment amount The actual adjustment action is obtained by superimposing the exploration noise. The adjusted parameters were applied to electrical stimulation, and the patient's swallowing performance was recorded, resulting in a reward. Next state ;Will Stored in the experience replay pool; when the number of samples in the experience replay pool is ≥256, 256 samples are randomly selected to form the pool. Calculate the loss of the Critic network. The loss function of the Critic network is the mean squared error loss, which is used to evaluate the difference between the current Q-value and the target Q-value. Where M=256, This represents the Q-value output by the current Critic network. The Q-value of the target Critic network output, discount factor =0.9, The action output by the target Actor network. This represents the reward obtained at time step i. The Adam optimizer is used for backpropagation optimization to update the Critic network (minimize...). (and the Actor network (maximizing expected reward through policy gradient). Update the target Actor network.) With the target Critic network To ensure training stability, repeat the above steps until 50 episodes are completed. If the average reward of 5 consecutive episodes is ≥8 (close to the optimal reward of 10), training can be terminated early. Extract the 3 sets of parameters with the highest rewards from the last 10 episodes, calculate the average value as the final optimized parameter personalized time series template.

[0042] Step S233: Finally, generate and verify the personalized timing template for optimized parameters.

[0043] In the personalized template construction stage, the input data includes two core parts: the timing parameters of the initial template. The activation initiation time covers 5 target muscles (k=1-5 corresponding to the mylohyoid muscle, anterior belly of the digastric muscle, etc.). Peak time With duration These parameters originate from two sources: first, the standard templates matched in the early stages (such as Template #2 for muscular paste); and second, the optimized electrical stimulation parameters, including the stimulation time difference for each muscle. (Individual adjustments may be necessary due to differences in muscle function; for example, the thyrohyoid muscle requires stimulation 5ms earlier, and the thyroarytenoid muscle requires stimulation 3ms later.) Stimulation intensity With frequency (Different muscles have different parameters, such as the anterior abdominal strength of the digastric muscle at 3.5 mA and the strength of the thyroarytenoid muscle at 2.8 mA, to suit the physiological characteristics of each muscle).

[0044] Template parameter calculation needs to be completed in two categories. Personalized activation timing parameters are obtained by superimposing the initial timing and the optimization time difference, i.e., the muscle activation start time. Peak time = + And duration The parameters need to be dynamically adjusted based on the patient's actual sEMG signal. If the patient's muscle activation duration is 20ms longer than the standard template, the corresponding parameter should be increased by 20ms. The electrical stimulation synchronization parameters need to be precisely matched with the muscle activation sequence, and the stimulation start time should be set to... 2ms (starting 2ms in advance to ensure synchronization with spontaneous muscle activation), the stimulation duration is directly used. This ensures that electrical stimulation and muscle contraction work in tandem throughout the entire process.

[0045] In the template validity verification process, , , An electrical stimulation system was used to instruct patients to swallow food of a texture (e.g., a paste) that matched a template. Physiological signals and clinical effects were collected from five swallowing attempts. The designed efficacy evaluation indicators included: safety, choking incidence = (number of choking attempts / 5) × 100% (≤20%); effectiveness, average bolus residue = (sum of residue scores from 5 attempts) / 5 (≤1 point); coordination, number of non-target muscle misactivations (≤1 per swallow), and peak amplitude ratio of laryngeal acceleration. (Requirement ≥0.6, an improvement of ≥20% from the initial value). If the target is not met, for example, the coughing incidence rate is 40%, backtrack the reinforcement learning process, check the reward function weights, increase the coughing penalty coefficient to 4, retrain for 10 episodes, fine-tune the parameters, and verify again.

[0046] Step S3: Trigger the execution of the final optimized parameter personalized temporal template biomimetic sequence through multimodal intent.

[0047] The patient issues swallowing commands through motor imagery (MI), steady-state visual evoked potentials (SSVEP), and other possible cortical EEG signals. Upon recognition by the BCI system, stimulation is not immediate; instead, the system invokes a personalized timing template with the patient's final optimized parameters. Based on this personalized timing template, the control module sends commands to multiple electrical stimulation units with millisecond-level precision, sequentially stimulating the target muscles to perfectly reproduce the coordinated movements of normal swallowing. The specific steps are as follows: Raw EEG data is acquired using a multimodal EEG cap, which may include steady-state visual evoked potentials (SSVEPs), motor imagery (MI), and other scalable EEG signals. The acquired EEG signals are preprocessed to remove noise and artifacts, improving signal quality. Features helpful in identifying user intent are extracted from the preprocessed signals. Intent recognition algorithms are applied to analyze the extracted features and determine the validity of the user's intent. If the intent is valid, the system invokes the patient's final optimized parameters and a personalized timing template to generate biomimetic timing instructions. These biomimetic timing instructions, generated based on the matching template, guide the output of electrical stimulation. Electrical pulses are output through a multi-channel electrical stimulator according to the biomimetic timing instructions. These pulses are delivered to the target muscle groups via electrodes. The electrical pulses stimulate the target muscle groups, producing the expected physiological response, such as swallowing.

[0048] Step S4: Perform closed-loop feedback and adaptive learning on the personalized time series template of the final optimized parameters.

[0049] After each stimulation, the system collects therapeutic data on the swallowing sensation using sensors (such as the aforementioned accelerometer). This data is then fed into an AI optimization engine and compared to the expected goals. If the effect is not optimal (e.g., significant residue remains), the engine will automatically fine-tune the timing or parameters of the next stimulation, achieving increasingly precise personalized treatment with each use. See also Figure 4 The closed-loop feedback and adaptive learning of the personalized time-series template for the final optimized parameters specifically includes the following steps: Step S400: Swallowing data is collected from patients using the personalized timing template with the final optimized parameters. During the data collection phase, acceleration, swallowing sounds, and sEMG signals are recorded simultaneously, and timestamps are marked using TTLs synchronization triggers (accurate to 0.05ms) to ensure time alignment of different data types. Within 10 seconds after collection, medical staff input the bolus residue score (0-4 points, 0 points for no residue, 4 points for extremely large residue) and swallowing completion time (the duration from the start of stimulation to the end of the swallowing sound, in ms) through the system interface. If data collection from a certain sensor fails (e.g., acceleration signal interruption), the system automatically uses backup data (e.g., estimating the swallowing completion time using the duration of the sEMG signal) to avoid feedback interruption due to data loss.

[0050] Step S410: The collected swallowing data is denoised and quantized to obtain a standardized therapeutic effect feature vector.

[0051] A fourth-order low-pass filter (cutoff frequency 50Hz) is used to remove high-frequency noise, and standard gravitational acceleration is used ( ) Calibrate peak amplitude (ensuring comparable amplitude values ​​for different patients and acquisition scenarios). Distinguish swallowing sounds from environmental noises using short-time energy analysis (50ms window size); energy exceeding baseline... ( Segments with a noise level exceeding the standard deviation of ambient noise were defined as valid swallowing sounds, and the total duration of valid segments was calculated as the "swallowing sound duration." Noise was removed using the wavelet denoising algorithm (db4 wavelet basis, 5-level decomposition) from step one. The root mean square (RMS) of the target muscle's sEMG signal during the stimulation period was calculated as an activation intensity indicator. The sEMG signals of non-target muscles exceeding the standard deviation of ambient noise were statistically analyzed. ( The number of times (remaining at rest standard deviation) is used as an indicator of false activation.

[0052] The noise-reduced data is converted into measurable quantitative indicators, including the improvement rate of laryngeal elevation. Swallowing efficiency index Muscle synergy index Safety indicators .in, , This represents the peak value that has been raised this time. This represents the peak elevation from the previous stimulus. , The time to complete swallowing. For residual score. , This represents the number of times non-target muscles were misactivated. The target muscle mass. =1-0.5 , This is the result of the cough test.

[0053] To eliminate differences in the dimensions of different features, Min-Max normalization is used to map all features to the [0,1] interval: ,in These are the original eigenvalues. , These are the historical minimum and maximum values ​​for this feature (based on statistics from the patient's previous 10 treatments). The generated standardized feature vector... .

[0054] Calculate the difference between the target standardized eigenvector value and the actual standardized eigenvector value. = Obtain the difference feature vector = A positive deviation indicates that the feature has not reached the target value, while a negative deviation indicates that it exceeds the target value (no adjustment is needed).

[0055] Step S420: Based on the difference feature vector, perform closed-loop feedback and adaptive learning on the personalized time-series template of the final optimized parameters.

[0056] Compared to the reinforcement learning in step S2, incremental reinforcement learning (IRL) can continuously update the model using therapeutic data after each stimulus, avoiding parameter obsolescence due to changes in patient function and achieving real-time adaptation. Incremental reinforcement learning (IRL) is a reinforcement learning method that allows an agent to progressively update its policy while continuously receiving new information.

[0057] The input to incremental reinforcement learning (IncrementalRL) is the current efficacy bias vector. Current stimulus parameters , I and F are divided into the timing, intensity, and frequency of 5 muscles; the output is the parameter adjustment amount. Temporal difference learning (TD-Learning) is employed, through... Update the action value function, where the learning rate α = 0.1 and the discount factor γ = 0.8. The specific process is as follows: Calculate the Pearson correlation coefficients between each stimulation parameter and therapeutic characteristics. Parameters with an absolute correlation coefficient |r| ≥ 0.3 are selected as sensitive parameters, such as the I-correlation coefficient of the thyrohyoid muscle. If r = 0.45, then I is the sensitive parameter.

[0058] Based on TD-Learning, parameter adjustment is performed by inputting the current state S (deviation vector ΔF + current parameter). Determine the range of motion based on the therapeutic effect level (e.g., good therapeutic effect corresponds to small-range adjustment of motion); select the motion that maximizes Q(S,A). (i.e., the optimal parameter adjustment amount); calculate the adjusted parameters. And verify the validity of the parameters (e.g.) The current must be within the range of 1-10mA; otherwise, the boundary value will be used.

[0059] Step S430, based on the adjusted parameters The parameter set is updated for the final optimized parameter personalized time series template.

[0060] Based on optimized parameters When updating the personalized timing template parameter set, first decompose The stimulation time difference Δt, intensity I, and frequency F of the five target muscles are mapped to the required timing adjustments and electrical stimulation parameters for the template. The current baseline timing parameters of the template are called, and a new activation onset / peak time is calculated based on "baseline value + optimized offset". The activation duration is calibrated using the patient's real-time sEMG signal to form a new timing parameter set. The stimulation onset time is calculated according to the rule of "activation onset time advanced by 2ms", and the stimulation duration is determined using the new duration, integrated into an electrical stimulation synchronization parameter table. Subsequently, the validity of the parameters is verified, therapeutic data is associated and labeled, the updated template is stored and archived, and the calling interface is updated to achieve dynamic template adaptation to the patient's current swallowing function.

[0061] This invention utilizes a standard physiological swallowing muscle activation timing model based on healthy individuals, combined with individual patient information, to select a suitable initial swallowing electrical stimulation sequence template. This method can provide a customized treatment plan for each patient, improving treatment outcomes.

[0062] The system acquires physiological signals from patients using an initial swallowing electrical stimulation sequence template in real time, and then obtains an optimized swallowing electrical stimulation sequence template through a reinforcement learning algorithm. This real-time feedback and adaptive adjustment mechanism can dynamically adjust the electrical stimulation parameters according to the patient's immediate response, ensuring the synchronicity and effectiveness of the treatment.

[0063] By using multimodal BCI technology to identify a patient's swallowing intention, electrical stimulation commands can be triggered more accurately and promptly. This not only improves the synchronicity of treatment but also enhances patient engagement and treatment effectiveness.

[0064] By collecting swallowing feedback data from patients using an optimized swallowing electrical stimulation sequence template, and further optimizing the electrical stimulation parameters using reinforcement learning algorithms, continuous optimization of treatment effects can be achieved. This method allows for dynamic adjustment of the treatment plan based on the patient's recovery progress, ensuring long-term treatment effectiveness.

[0065] The aforementioned techniques can significantly improve the speed of swallowing function recovery, reduce rehabilitation time, and thus improve patients' quality of life. At the same time, this personalized, real-time feedback, and adaptive treatment plan can enhance patients' confidence in treatment and promote the rehabilitation process.

[0066] This swallowing electrical stimulation timing generation method based on multimodal BCI and biomimetic temporal electrical stimulation solves the problems existing in traditional swallowing rehabilitation therapy through personalized treatment plans, real-time feedback and adaptive adjustment, accurate swallowing intention recognition and continuous optimization of treatment effects. It significantly improves the treatment effect and the patient's recovery speed and has important clinical application value.

[0067] The following example illustrates this with an ALS patient whose main symptom is weakened swallowing muscle function during the oropharyngeal phase.

[0068] Based on the patient's VFSS assessment results, a weighted Euclidean distance model was used to calculate the matching degree. The patient's quantitative index vector was: X = [0.5 (laryngospasm level 2), 0.75 (UES open level 3), 0.8 (choking 20%), 0.6 (Kubotia level 3)], and the weight vector was W = [0.4, 0.3, 0.2, 0.1]. The template with the smallest weighted distance (d = 0.31) to "Template #2 (pureed food)" in step S1 was selected as the initial template, and the initial time series parameters were... as follows:

[0069] The initial electrical stimulation parameters were determined as follows: =3mA, safe starting strength, suitable for the muscle fatigue characteristics of ALS patients; =40Hz, lower than the standard 50Hz, to avoid over-excitation of muscles; =[0,0,0,0,0]. The core settings for reinforcement learning include real-time patient signal characteristics (root mean square RMS of sEMG for M1-M5, laryngeal elevation amplitude ratio). =0.5, Coughing Marker C, Residual Score ) + Current parameter ( Parameter adjustment amount =±3ms, =±0.3mA, ±5Hz), set reward function To increase the penalty weight for coughing and reduce the risk of aspiration in ALS patients, the training process was iterated for 50 episodes, eventually converging to the optimal parameters. .

[0070] pass , = + Adjusting the duration based on the patient's sEMG signal The final parameters are as follows:

[0071] The patient issues swallowing commands through motor imagery (MI), steady-state visual evoked potentials (SSVEP), and other possible cortical EEG signals. Upon recognition by the BCI system, it does not immediately stimulate the muscles but instead invokes a personalized timing template with the patient's final optimized parameters. Based on this personalized timing template, the control module sends commands to multiple electrical stimulation units with millisecond-level precision, sequentially stimulating the target muscles and perfectly replicating the coordinated movements of normal swallowing.

[0072] After several iterations of closed-loop feedback and adaptive learning on the personalized time-series template for the final optimized parameters, the optimized parameters are as follows:

[0073] The system found the optimal stimulation parameters for the patient, achieving the best auxiliary effect.

[0074] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

[0075] A computer device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0076] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory, causing the computer device to perform all or part of the steps of the learning outcome prediction method based on learning behavior data mining of the foregoing embodiments of this disclosure.

[0077] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0078] like Figure 5 This is a schematic diagram of a computer device provided for an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the computer device in the embodiments of the present disclosure. Figure 5 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0079] like Figure 5 As shown, a computer device may include a processor (such as a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) or programs loaded from storage devices into random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer device. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0080] Typically, the following devices can be connected to the I / O interface: input devices, such as sensors or visual information acquisition devices; output devices, such as displays; storage devices, such as magnetic tapes or hard drives; and communication devices. Communication devices allow the computer device to communicate wirelessly or wiredly with other devices (such as edge computing devices) to exchange data. Although a computer device with various devices is illustrated, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0081] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the learning outcome prediction method based on learning behavior data mining according to embodiments of this disclosure are performed.

[0082] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0083] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the learning outcome prediction method based on learning behavior data mining according to the foregoing embodiments of the present disclosure are performed.

[0084] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0085] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0086] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0087] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0088] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0089] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0090] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0091] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0092] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for generating a swallowing biomimetic electrical stimulation time sequence based on multimodal BCI, characterized in that, include: A standard physiological swallowing muscle activation timing model was established based on the surface electromyography (sEMG) signals of swallowing-related muscles in healthy individuals. Based on individual patient information, a first swallowing electrical stimulation sequence template is selected from the standard physiological swallowing muscle activation timing model. The physiological signals of the patient using the first swallowing electrical stimulation sequence template are collected in real time, and the second swallowing electrical stimulation sequence template is obtained through a reinforcement learning algorithm; The patient's swallowing intention is identified through multimodal BCI, and electrical stimulation commands are sent to multiple target muscle groups according to a predetermined time sequence based on the second swallowing electrical stimulation sequence template. The patient's swallowing effect feedback data using the second swallowing electrical stimulation sequence template is collected, and the parameters of the second swallowing electrical stimulation sequence template are optimized and updated using a reinforcement learning algorithm to obtain a third swallowing electrical stimulation sequence template.

2. The method as described in claim 1, characterized in that, The establishment of a standard physiological swallowing muscle activation time sequence model based on surface electromyography (sEMG) signals of swallowing-related muscles in healthy individuals specifically includes: sEMG signals of the mylohyoid muscle, anterior belly of the digastric muscle, geniohyoid muscle, thyrohyoid muscle, and thyroarytenoid muscle were collected, as well as laryngeal acceleration and swallowing sound signals. The surface electromyography (sEMG) signal is preprocessed to extract a smooth sEMG envelope. Based on the smooth sEMG envelope, laryngeal acceleration, and swallowing sound signals, muscle activation timing analysis and intermuscular timing relationships are obtained. A standard physiological swallowing muscle activation timing model was established based on the muscle activation timing analysis and intermuscular timing relationship analysis.

3. The method as described in claim 2, characterized in that, The process of obtaining muscle activation timing analysis and intermuscular timing relationships based on the smoothed sEMG envelope, laryngeal acceleration, and swallowing sound signals specifically includes: Based on the smooth sEMG envelope and the peak point of the laryngeal acceleration signal on the y-axis, the activation start time, peak time, and duration of the mylohyoid muscle, the anterior belly of the digastric muscle, the geniohyoid muscle, the thyrohyoid muscle, and the thyroarytenoid muscle are extracted. The activation start time, peak time, and duration constitute the muscle activation timing analysis. The smoothed sEMG envelopes of any two groups of swallowing-related muscles were normalized, and the cross-correlation function of the normalized sEMG envelopes of the two groups of muscles was calculated. Obtain the temporal relationship between the muscles; wherein The normalized muscle sEMG envelope is shown, where T represents the duration of swallowing. For time delay.

4. The method as described in claim 1, characterized in that, The selection of the first swallowing electrical stimulation sequence template from the standard physiological swallowing muscle activation timing model based on individual patient information specifically includes: Obtain the patient's physiological information, wherein the physiological information includes at least one or more of the following: laryngeal elevation amplitude, swallowing sounds, and upper esophageal sphincter (UES) opening degree; Based on the patient's physiological information, a first swallowing electrical stimulation sequence template was selected from the standard physiological swallowing muscle activation timing model using a weighted Euclidean distance model.

5. The method as described in claim 4, characterized in that, The real-time acquisition of physiological signals from the patient using the first swallowing electrical stimulation sequence template, and the acquisition of the second swallowing electrical stimulation sequence template through a reinforcement learning algorithm, specifically includes: The patient's first physiological signal was acquired using the first swallowing electrical stimulation sequence template. The first physiological signal included the sEMG signal of the swallowing-related muscles, as well as laryngeal acceleration and swallowing sound signals. The first physiological signal is preprocessed to extract key features of the first physiological signal; The first swallowing electrical stimulation sequence template is optimized using a deep deterministic strategy gradient. The key features of the first physiological signal are used as input signals, and the output is the first electrical stimulation parameters after optimization of the first swallowing electrical stimulation sequence template. The second swallowing electrical stimulation sequence template is obtained based on the first electrical stimulation parameters.

6. The method as described in claim 5, characterized in that, The optimization of the first swallowing electrical stimulation sequence template using a deep deterministic gradient strategy, taking key features of the first physiological signal as input signals and outputting optimized first electrical stimulation parameters for the first swallowing electrical stimulation sequence template, specifically includes: Load the first swallowing electrical stimulation sequence template parameters as the current parameters , ,in For the current stimulus time difference, Initial value The current intensity represents the magnitude of the electrical current applied to the muscle during the current stimulation. As the initial value, The current stimulation frequency, Initial value; Constructing the State vector The State vector S t Includes target muscle activation intensity The ratio of the degree of laryngeal elevation to the standard value Coughing marker C, food bolus residue score and current electrical stimulation parameters , ; Perform loop training of the Actor network to obtain at least 256 samples, and calculate the Critic network loss based on at least 256 samples; The Adam optimizer is used for backpropagation optimization to update the Critic network and the Actor network respectively; The three sets of parameters with the highest rewards are extracted, and their average value is calculated as the update parameter of the first swallowing electrical stimulation sequence template, thereby obtaining the second swallowing electrical stimulation sequence template.

7. The method as described in claim 1, characterized in that, The step of recognizing the patient's swallowing intention through multimodal BCI and sending electrical stimulation commands to multiple target muscle groups according to a predetermined time sequence based on the second swallowing electrical stimulation sequence template specifically includes: The patient issued swallowing instructions based on cortical EEG signals obtained through motor imagery or steady-state visual evoked potentials. The BCI system recognizes the swallowing command and invokes the patient's second swallowing electrical stimulation sequence template. According to the second swallowing electrical stimulation sequence template, instructions are sent to the electrical stimulation units of multiple channels to sequentially stimulate the swallowing-related muscles.

8. The method as described in claim 1, characterized in that, The process of collecting swallowing effect feedback data from the patient using the second swallowing electrical stimulation sequence template, and optimizing and updating the parameters of the second swallowing electrical stimulation sequence template using a reinforcement learning algorithm to obtain the third swallowing electrical stimulation sequence template specifically includes: Swallowing data was acquired from the patient using the second swallowing electrical stimulation sequence template to obtain a second physiological signal; The second physiological signal is preprocessed to extract a standardized therapeutic effect feature vector; the difference between the target standardized feature vector value and the actual standardized feature vector value is calculated to obtain the difference standardized feature vector. Based on the difference-standardized feature vector, the third swallowing electrical stimulation sequence template is obtained by closed-loop feedback and adaptive learning on the second swallowing electrical stimulation sequence template.

9. The method as described in claim 8, characterized in that, The step of obtaining the third swallowing electrical stimulation sequence template by performing closed-loop feedback and adaptive learning on the second swallowing electrical stimulation sequence template based on the difference-standardized feature vector specifically includes: Calculate the Pearson correlation coefficient between the stimulation parameters of the swallowing-related muscles and the therapeutic characteristics. Stimulus parameters with an absolute correlation coefficient greater than a threshold are selected as sensitive parameters; The difference-standardized feature vector and the third swallowing electrical stimulation sequence template are used as the current state S; Choose the action that maximizes Q(S,A). , where Q(S,A) represents the action value function of taking action A in state S; Actions that maximize Q(S,A) This serves as a parameter for updating the second swallowing electrical stimulation sequence template.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the swallowing biomimetic electrical stimulation timing generation method based on multimodal BCI as described in any one of claims 1-9.

Images