Brain-computer hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke

Abstract

The invention discloses a brain-computer hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke, and relates to the field of artificial intelligence medical instruments. The system is composed of a plurality of key modules: an intelligent swallowing brain-computer interface module for accurately identifying a swallowing intention through multi-modal physiological signal acquisition and processing; the intelligent swallowing functional electrical stimulation module is used for finely regulating and controlling stimulation parameters according to the condition of the patient; the AI auxiliary online integrated module is used for realizing non-invasive brain-controlled swallowing and high-precision AI regulation and control; the closed-loop intelligent feedback regulation and control module is used for real-time monitoring, early warning and self-adaptive optimization of electrical stimulation; the cloud platform remote monitoring and data storage module supports remote diagnosis and treatment of doctors; and the man-machine interaction interface display and printing module optimizes doctor-patient interaction. According to the system, a brain-computer interface and FES are combined, a personalized, safe and efficient rehabilitation scheme is provided for a patient based on data driving and an artificial intelligence algorithm, swallowing function recovery can be accelerated, complications are reduced, and the system has great clinical and social values.

Description

Technical Field

[0001] The present invention belongs to the technical field related to artificial intelligence medical devices, and specifically relates to a brain-computer hybrid intelligent closed-loop swallowing rehabilitation system for treating post-stroke dysphagia. Background Art

[0002] Stroke is the leading cause of dysphagia, with 51%-73% of stroke patients experiencing dysphagia. Dysphagia not only causes complications such as severe pneumonia, malnutrition, and dehydration, but also significantly increases stroke mortality and disability, prolongs hospital stay, and greatly increases medical expenses. With the advent of an aging society, the number of stroke patients has increased dramatically, and the number of patients with dysphagia after stroke is growing larger and larger, making it urgent to solve this problem.

[0003] Normal swallowing is completed in four stages: oral preparation, oral autonomy, pharyngeal, and esophageal stages, of which the oropharyngeal stage is closely related to stroke. The swallowing process is controlled by three parts of the nervous system: the first is the higher central system in the motor area of the cerebral cortex, which is responsible for initiating swallowing and controlling the oropharyngeal stage, and achieves functional regulation by regulating the medullary swallowing center; the second is the medullary swallowing center system, as a functional neuron group, which is interconnected internally. Effective stimulation can excite the corresponding neurons, coordinate and organize the swallowing muscles to perform patterned sequential movements; the third is the sensory afferent and motor efferent system of the peripheral nerves, which plays a coordinating role in controlling the oropharyngeal muscles.

[0004] The pathological mechanism of dysphagia is relatively complex: in the oral stage, loss of tongue sensation, paralysis of the tongue muscles, and dysfunction of the lip or facial muscles can cause saliva accumulation, flow, and dysarthria, affecting chewing and food bolus formation, leading to delayed or inability to swallow, and may also cause food to flow out of the corners of the mouth or overflow into the oropharynx prematurely, causing aspiration; abnormal movement of the front 2 / 3 of the tongue can hinder the lifting, shaping, and propulsion of food, resulting in repeated ineffective swallowing movements, and food is retained on one side of the mouth or overflows and cannot be delivered to the back of the mouth. During the swallowing stage, the weakening of the posterior tongue causes a significant decrease in the force that pushes the food bolus into the hypopharynx, prolonging the duration of the swallowing stage; paresthesia in the throat, disordered pharyngeal muscle movement, and weakened contraction force can cause a delayed or absent swallowing reflex, poor laryngeal closure during swallowing, and cause coughing, choking, aspiration, and dysphonia during eating; the important propulsion force of the food bolus in the hypopharynx comes from the pharyngeal sphincter, and pharyngeal paralysis can lead to pharyngeal retention, and aspiration is the most critical complication of dysphagia. The swallowing stage of the throat plays a decisive role in whether there is aspiration during stroke.

[0005] Dysphagia after stroke is mainly attributed to damage to neural pathways at various levels, including damage to the swallowing motor cortex, damage to the descending cortical projections, damage to the medullary swallowing center, and damage to the cranial nerves related to swallowing, such as the vagus nerve, trigeminal nerve, and hypoglossal nerve.

[0006] However, although the existing rehabilitation treatments for dysphagia after stroke have been widely carried out and achieved certain results, most of them focus on the stimulation of superficial local muscles, while swallowing requires the coordination of multiple groups of neuromuscles. The root cause of dysphagia after stroke is the damage of the central nervous system leading to motor dysfunction, which makes the voluntary movement command (swallowing intention) of swallowing unable to be transmitted to the muscles, thus resulting in the decline of the ability to control swallowing and triggering swallowing dysfunction.

[0007] In view of this, the present invention aims to provide an intelligent device system that closely associates the patient's swallowing intention with the oropharyngeal swallowing action, accurately captures and identifies the physiological signals of electroencephalogram and electromyogram of the swallowing intention through brain-computer hybrid intelligence, and then implements precise and autonomous closed-loop electrical pulse stimulation regulation on the muscle groups performing swallowing movement, so as to promote the generation of effective swallowing actions and greatly reduce ineffective or excessive stimulation. This system covers the technologies of artificial intelligence prediction model construction, artificial intelligence algorithm, electrical stimulation parameter optimization model, and microcontroller digital-to-analog conversion, forms a monitoring and evaluation treatment feedback closed-loop based on objective physiological index data, and creates a personalized swallowing function rehabilitation treatment system.

[0008] By allowing the patient to imagine the swallowing intention, this process has the same neuronal conduction pathway and similar activation of brain motor areas as the normal swallowing movement in the normal physiological state (non-damaged). With continuous swallowing intention imagination training, the patient can induce the plastic reconstruction and functional reorganization of the central nervous system motor neural network and restore to the natural and coordinated normal swallowing function state as soon as possible.

[0009] The present invention integrates multidisciplinary knowledge of neuroscience, rehabilitation medicine, digital signal processing, electrical and electronics, artificial intelligence, and signal transmission, and provides an innovative and efficient treatment method for the restoration of normal function of dysphagia after stroke. Summary of the Invention

[0010] The purpose of the present invention is to provide a brain-computer hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke, so as to solve the problems raised in the above background technology.

[0011] To achieve the above purpose, the present invention provides the following technical solutions:

[0012] A brain-computer hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke, comprising:

[0013] Intelligent swallowing brain-computer interface module, used to monitor and collect multi-modal physiological signals, decode swallowing intentions, and construct a prediction model driven by objective data. The multi-modal physiological signals include electroencephalogram (EEG) signals collected from the frontal, parietal, temporal, and occipital brain regions using a portable wireless wearable EEG signal acquisition device, and surface electromyogram (sEMG) signals collected from the geniohyoid muscle and digastric muscle using a surface EMG acquisition device. The collected EEG signals are denoised using band-pass filtering, independent component analysis (ICA), or wavelet transform, and the EMG signals are subjected to band-pass filtering, power frequency denoising, and short-term smoothing processing. Then, the time-domain features, frequency-domain features, time-frequency features, mutual information, or correlation analysis features of the EEG signals, and the time-domain features, frequency-domain features, and time-frequency features of the EMG signals are extracted respectively. The features extracted from the EEG and EMG are concatenated into a single vector through feature-level fusion and input into a machine learning model, or decision-level fusion is used to train the EEG and EMG prediction models separately and the prediction results are fused using the weighted voting method to identify swallowing intentions. Finally, the EEG data of the subject in the BCI computer is defined as "on" or "off" stimulation by the prediction model and an instruction is sent;

[0014] Intelligent swallowing functional electrical stimulation module, which can intelligently sense, adapt to, and optimize the autonomous adjustment of various stimulation parameters. The stimulation parameters include the stimulation site, where the negative electrode is placed at the center of the muscle belly of the thyrohyoid muscle and geniohyoid muscle, and the positive electrode is placed at the tendon of the corresponding muscle. The stimulation waveforms include asymmetric biphasic waves for stimulating small muscles and symmetric biphasic waves for stimulating large muscles. The stimulation frequency is between 30 - 50 Hz, the stimulation amplitude uses a 1 - 5 mA current stimulation and is slightly adjusted as the swallowing force increases. The stimulation pulse width is between 200 - 500 μs, the duty cycle is 30%, the treatment time is 30 min to 1 h per single session, five times a week to several times a day, and the total is about 40 h in 2 to 3 months. This module also includes a structure for adaptive intelligent regulation composed of a sensing system, a signal processing and intelligent analysis system, and an electrical stimulation execution system, as well as a feedback and adjustment unit for adaptive intelligent regulation, which is used to dynamically optimize the FES stimulation parameters according to the feedback data;

[0015] AI-assisted BCI-FES online integration module, which combines artificial intelligence, brain-computer interface, bio-signal processing, and FES to build an efficient closed-loop system for non-invasive brain-controlled swallowing. Patients only need to imagine the swallowing movement to trigger FES. It has high-precision AI regulation to optimize swallowing recognition based on deep learning, achieving real-time closed-loop regulation including swallowing monitoring + FES stimulation + AI optimization of the entire process closed-loop control. It has personalized adaptation ability to automatically adjust the training plan according to the patient's swallowing recovery progress. Its structure includes a brain-computer interface system for enhancing swallowing recognition, accurately identifying swallowing intention, and synchronously activating swallowing muscle groups. The artificial intelligence regulation system includes a swallowing pattern recognition algorithm, adaptive FES parameter optimization, and a closed-loop control strategy. The FES execution system provides multi-channel FES output, can adjust the stimulation parameters personalizedly, and receives instructions through wireless transmission to execute stimulation. And it uses AI to optimize FES parameters, collects EEG, EMG, and swallowing success rate data through supervised learning, uses SVM, XGBoost, and Random Forest to train the FES parameter prediction model to achieve real-time inference and provide personalized parameters, uses reinforcement learning through reinforcement learning of state, action, and reward, adopts the deep Q network algorithm to adaptively and dynamically regulate FES parameters, automatically triggers FES to improve the ability of autonomous swallowing, continuously optimizes FES to reduce the number of stimulations, improves the rate of autonomous swallowing, and maintains and promotes swallowing recovery;

[0016] Closed-loop intelligent feedback regulation module, including:

[0017] Signal acquisition layer, which monitors multi-modal bio-signals related to swallowing in real time, including electroencephalogram sensors to collect swallowing intention signals to achieve BCI control of FES, surface electromyogram sensors to detect the electrical activities of swallowing-related muscle groups to evaluate muscle responses, acceleration / pressure sensors to monitor laryngeal and hyoid bone movements to judge swallowing quality, pharyngeal pressure sensors to measure the pressure during swallowing to assist in the analysis of swallowing success rate, and the aspiration / choking detection module to identify swallowing abnormalities by combining respiratory patterns and voice analysis;

[0018] AI intelligent analysis layer, based on AI technologies such as deep learning, pattern recognition, and reinforcement learning, realizes swallowing intention recognition and stimulation regulation optimization, covering a swallowing pattern recognition model to improve the accuracy of swallowing detection based on EEG and sEMG signals, aspiration detection and safety warning to identify abnormal swallowing or choking and adjust the FES strategy in time, adaptive electrical stimulation optimization to adjust FES parameters using machine learning to optimize the stimulation effect, and personalized training strategy to dynamically adjust the swallowing training mode according to the patient's rehabilitation progress;

[0019] The FES execution layer, based on the AI analysis results, accurately controls the electrical stimulation of the swallowing-related muscles, including the electrical stimulation unit's precise electrical stimulation of the digastric and hyoid muscles through a multi-channel FES system, and dynamically adjusts the FES stimulation frequency, pulse width, and intensity to ensure the best swallowing recovery effect. Timing synchronization optimization ensures FES stimulation

[0020] Stimulate the swallowing rhythm in accordance with the natural swallowing rhythm to improve swallowing efficiency;

[0021] The cloud platform remote monitoring and data storage module establishes a cloud data storage and remote monitoring system to record BCI-FES swallowing rehabilitation data in real time, including:

[0022] The data collection end collects and uploads the patient's swallowing rehabilitation data, including EEG for swallowing intention recognition, sEMG for swallowing quality assessment, and FES parameters;

[0023] Cloud data storage, secure storage of patient swallowing data, establishment of a time series database to store EEG / EMG continuous signals, a relational database to store FES parameters and training records, a non-relational database to store patient feedback and doctor advice, and data encryption access control;

[0024] AI analysis and remote monitoring, based on AI optimization of FES parameters, provide personalized rehabilitation plans, use supervised learning to predict the optimal FES parameters, strengthen learning to dynamically adjust FES stimulation, automatically calculate swallowing success rate, muscle activation, FES dependence, and generate personalized rehabilitation curves for doctors to view detailed data;

[0025] Doctors and patients can interact with each other by viewing patient data remotely through the Web or APP, adjusting FES plans, monitoring patient swallowing data in real time, adjusting FES parameters remotely, and providing voice / video rehabilitation guidance. Patients can visualize their swallowing progress, obtain AI training suggestions, and provide voice / video feedback to record their training experience through the mobile APP.

[0026] The human-computer interaction interface display and printing module establishes a visual interactive interface, displays swallowing training data in real time, generates personalized rehabilitation curves, and supports data export / printing. It includes patient management, real-time data monitoring, FES remote control, automatic generation of swallowing rehabilitation reports and support for export / printing through the Web and APP on the doctor side, and swallowing training guidance, training result visualization, and doctor's messages on the patient side through mobile APP and touch screen devices.

[0027] In the intelligent swallowing brain-computer interface module, EEG signal processing further includes: decoding the EEG signals recorded from the motor cortex, extracting meaningful features, namely swallowing intention, and using machine learning and deep learning for classification, generating a general prediction model and transfer learning to establish a personalized prediction model.

[0028] In the intelligent swallowing functional electrical stimulation module, in the structural composition adapted for intelligent regulation:

[0029] In the sensing system, the surface electromyogram sensor detects the electrical activities of swallowing-related muscles such as the digastric muscle and the hyoid muscle to identify the swallowing intention, the acceleration / pressure sensor monitors the movements of the hyoid bone and the larynx to confirm the swallowing action, and the physiological signal monitors the heart rate and respiration to ensure the safety of stimulation;

[0030] In the signal processing and intelligent analysis system, the bio-signal preprocessing module filters, denoises, and normalizes the collected signals. The swallowing pattern recognition module uses deep learning or machine learning algorithms to identify swallowing actions and abnormal swallowing conditions. The real-time feedback control module adaptively adjusts the electrical stimulation parameters based on the swallowing state to achieve closed-loop control. The data storage and analysis module records the user's swallowing data and provides personalized rehabilitation suggestions;

[0031] In the electrical stimulation execution system, the multi-channel FES output unit can independently adjust the stimulation parameters of different muscle groups to ensure coordinated control, can adjust the stimulation parameters to adapt to patients with different degrees of swallowing disorders, and the closed-loop feedback adjustment dynamically optimizes the stimulation strategy according to the sensor data to improve the training effect. The wireless / portable stimulation device improves the convenience and comfort of patient use.

[0032] Preferably, in the AI-assisted BCI-FES online integration module, the closed-loop control strategy of the artificial intelligence regulation system is: real-time monitoring of the swallowing effect, adjusting the stimulation plan according to the monitoring results, and optimizing the swallowing function recovery process. Specifically, by continuously obtaining and analyzing the data feedback of EEG, sEMG, and pressure sensors, the FES stimulation parameters are dynamically adjusted to achieve the best swallowing training effect.

[0033] In the closed-loop intelligent feedback regulation module:

[0034] The swallowing intention recognition and triggering function is specifically to combine the electromyogram signal and the laryngeal movement data, and BCI+AI improves the accuracy of swallowing recognition and triggers the FES electrical stimulation in real time to promote the recovery of swallowing function;

[0035] The intelligent FES parameter adaptive optimization function automatically adjusts the FES intensity, frequency, and pulse timing based on the biofeedback data, and adopts a personalized training mode to gradually improve the patient's swallowing ability and improve the swallowing training effect;

[0036] The closed-loop control and real-time feedback function also includes aspiration detection and safety mechanism. The aspiration / coughing detection module monitors swallowing abnormalities in real time. Once an abnormality is detected, the FES strategy is immediately adjusted, and the stimulation parameters are paused or changed to ensure the safety of swallowing training;

[0037] The rehabilitation data analysis and remote monitoring function further generates a detailed personalized rehabilitation report by recording the swallowing training data. The report content covers the change trend of swallowing success rate, FES dependence analysis, and muscle activation degree evaluation. Doctors can view the report online through the remote monitoring function and accurately adjust the training plan to improve the efficiency of rehabilitation management.

[0038] Preferably, in the cloud platform remote monitoring and data storage module:

[0039] When the data acquisition terminal collects data, it encrypts the data of EEG, sEMG, and FES parameters in real time and uploads them to the cloud platform using a secure transmission protocol to ensure the security and integrity of the data during the transmission process;

[0040] The time series database for cloud data storage uses an efficient data compression algorithm to reduce storage occupancy while ensuring the integrity of the time series characteristics of the data, facilitating the accurate extraction of dynamic characteristics when performing subsequent AI analysis of swallowing patterns;

[0041] In the AI analysis and remote monitoring link, when supervised learning predicts the optimal FES parameters, it comprehensively analyzes the patient's historical training data and current physiological state data. When reinforcement learning dynamically adjusts the FES stimulation, more environmental factors such as the patient's fatigue level and diet on the day are introduced as adjustment bases, making the optimization of FES stimulation parameters more in line with the actual needs of the patient, and the generated personalized rehabilitation curve has higher reference value.

[0042] Preferably, in the human-computer interaction interface display and printing module:

[0043] Under the real-time data monitoring function of the doctor terminal, in addition to visually displaying EEG / EMG / FES data, it also provides a data comparison and analysis tool that can compare the current patient's data with the average data of patients in the same type of rehabilitation stage or historical rehabilitation data to assist doctors in more accurately judging the patient's rehabilitation progress;

[0044] In the swallowing training guidance function of the patient terminal, according to the patient's personal swallowing habits and rehabilitation progress, AI dynamically adjusts the visual feedback and voice prompt content, such as changing the prompt frequency and optimizing the prompt screen, making the training guidance more targeted and improving the patient's training compliance.

[0045] Preferably, it also includes the circuit of the swallowing neuromuscular electrical stimulation system, which is mainly composed of parts such as power management, electrode drive, current / voltage modulation, and control unit. It uses low-frequency pulsed current to stimulate the laryngeal muscles to promote the recovery of swallowing function, where:

[0046] The power supply module provides multiple voltages. The STM32 control unit requires 3.3V / 5V, the AD9833 DDS chip requires 3.3V, the operational amplifier signal conditioning requires ±12V, and the H-bridge drive current amplification requires ±5V / ±12V;

[0047] The signal generation module is controlled by the STM32 through SPI to generate a square wave signal from the AD9833. The frequency is adjustable from 20Hz to 50Hz. The square wave output is 0V - 3.3V. 16-bit SPI data transmission is used. After the unipolar 0 - 3.3V square wave is output, it is amplified and transformed into a bipolar ±5V square wave;

[0048] The signal conditioning module uses the operational amplifier OPA2134 to achieve signal amplification and bias adjustment, making the output become ±3.3V, and then further amplifying it to ±5V through the H-bridge;

[0049] The H-bridge circuit selects the H-bridge chip DRV8871. By controlling the switch, the direction of current flow is changed. When a high level is input, the output is +5V, and when a low level is input, the output is -5V, forming a bidirectional symmetric square wave;

[0050] The skin electrode & protection module has isolation protection. It uses the optocoupler TLP250 to isolate the MCU and the H-bridge. The TVS diode 5V prevents overvoltage from harming the patient. The current-limiting resistor 1kΩ ensures a safe current. The output frequency is 20 - 50Hz, the square wave amplitude is ±5V, and the output current is 0 - 10mA.

[0051] Preferably, in terms of the operation process of the circuit of the swallowing neuromuscular electrical stimulation system:

[0052] First, power is supplied. The STM32, AD9833, and H-bridge work together to provide the power required by each module;

[0053] Then, the STM32 controls the AD9833 through SPI to generate a square wave signal;

[0054] Then, the operational amplifier conditions the signal and converts the output of the AD9833 to ±5V;

[0055] Next, the H-bridge amplifies the signal to generate a biphasic square wave;

[0056] Finally, the electrode outputs with isolation & protection to stimulate the swallowing muscles.

[0057] Preferably, its working mechanism is to combine a non-invasive brain-computer interface based on EEG data with a non-invasive FES system for swallowing, and use artificial intelligence algorithms and large model technology as the core of monitoring, regulation, and feedback technology to achieve effective swallowing functional movements, while inducing the plasticity reconstruction and functional reorganization of the damaged central nervous system motor neural network, prompting patients to recover their autonomous and coordinated swallowing function to a natural and normal physiological state as soon as possible, using the brain's stronger swallowing movement intention to complete the swallowing movement imagination task, stimulate the corresponding cerebral cortex, and activate the brain's neural plasticity.

[0058] Compared with the prior art, the present invention provides a brain-computer hybrid intelligent closed-loop swallowing rehabilitation system for treating post-stroke dysphagia, which has the following beneficial effects:

[0059] 1. Precise and personalized rehabilitation treatment

[0060] Through the intelligent swallowing brain-computer interface (BCI) module, combined with a portable 8-lead wireless wearable EEG signal acquisition device and a surface electromyography acquisition device, multimodal physiological signals are fully collected. It can not only accurately capture the subtle changes in the brain's electrical activity when the swallowing intention is generated, reflecting the patient's unique swallowing intention, but also monitor the actual movement state of swallowing-related muscles in real time. Based on these rich and accurate data, advanced machine learning and deep learning algorithms are used, such as feature-level fusion to splice EEG and electromyography features into the model, or decision-level fusion weighted voting to judge swallowing intention, to establish a general and personalized swallowing intention prediction model. This enables the system to deeply understand the swallowing intention of each patient, laying a solid foundation for subsequent precision treatment, and overcoming the shortcomings of traditional treatment methods that are based only on surface symptoms and lack accurate insight into the intrinsic activities of individual swallowing nerves and muscles of patients.

[0061] The intelligent swallowing functional electrical stimulation (FES) module finely adjusts key parameters such as stimulation site, waveform, frequency, amplitude, pulse width, duty cycle and treatment time according to the individual patient's swallowing function, muscle characteristics and rehabilitation process. For example, based on the principle of motor nerve plasticity, the optimal stimulation sites of the thyrohyoid muscle and mylohyoid muscle are accurately selected, with the negative electrode placed in the center of the muscle belly and the positive electrode at the tendon to ensure the maximum stimulation effect; based on the muscle size, depth and required stimulation intensity, the asymmetric biphasic wave or symmetric biphasic wave is intelligently selected, and the stimulation frequency of 30-50Hz, the stimulation amplitude of 1-5mA and the stimulation pulse width of 200-500μs are finely adjusted to meet the personalized needs of patients with different degrees of swallowing disorders. Compared with the traditional unified mode of electrical stimulation treatment, this personalized electrical stimulation program can more accurately activate the corresponding swallowing muscle groups, avoid ineffective or excessive stimulation, and greatly improve the pertinence and effectiveness of the treatment.

[0062] 2. Efficient swallowing function recovery

[0063] The closed-loop integrated working mechanism constructed by this system closely combines the swallowing function detection technology and the regulation and stimulation technology of brain-computer hybrid intelligence. When the patient has the intention to swallow, the BCI module quickly and accurately identifies it and immediately sends instructions to the FES module. The FES module dynamically optimizes and adjusts the stimulation parameters according to the real-time physiological signal feedback of the patient. For example, it continuously optimizes the current, frequency, pulse width, etc. through reinforcement learning to achieve immediate and accurate electrical pulse stimulation regulation of the swallowing muscles, prompting the rapid generation of effective swallowing actions. Through repeated training like this, the synergy between the brain's swallowing-related neural pathways and muscles is strengthened, accelerating the plasticity reconstruction and functional reorganization of the motor neural network in the central nervous system, enabling the patient to quickly recover to a coordinated swallowing function in the natural normal physiological state. Compared with the traditional rehabilitation treatment that simply relies on passive training or fixed-mode electrical stimulation and lacks a real-time feedback adjustment mechanism, the rehabilitation cycle is greatly shortened.

[0064] The AI-assisted BCI-FES online integration module realizes non-invasive brain-controlled swallowing. The patient only needs to imagine the swallowing action to trigger the FES, greatly mobilizing the patient's enthusiasm for actively participating in rehabilitation training. At the same time, the high-precision AI regulation optimizes swallowing recognition based on deep learning. The real-time closed-loop regulation covers the entire process closed-loop control of swallowing monitoring + FES stimulation + AI optimization, ensuring that each swallowing training can be accurately adjusted according to the patient's current actual state. This combination of active participation and precise regulation significantly improves the efficiency and effect of swallowing training compared with the situation in traditional treatment where the patient passive receives treatment and the treatment means lack dynamic optimization, prompting the patient's swallowing function to recover faster and better.

[0065] III. Omnidirectional Real-time Monitoring and Safety Guarantee

[0066] The signal acquisition layer of the closed-loop intelligent feedback regulation module real-time monitors multi-modal biological signals related to swallowing, including all-round monitoring means such as electroencephalogram, surface electromyogram, acceleration / pressure, pharyngeal pressure sensors, and aspiration / cough detection modules. These sensors work together to not only accurately identify the swallowing intention, judge the swallowing quality in real time, but also promptly detect abnormal swallowing conditions such as aspiration and cough. For example, the pharyngeal pressure sensor accurately measures the pressure changes during swallowing, combined with the muscle contraction situation reflected by the surface electromyogram signal, to accurately judge whether the food bolus smoothly passes through the pharynx; the aspiration / cough detection module detects swallowing abnormalities at the first time by analyzing the breathing pattern and sound characteristics, providing a key basis for subsequent safety regulation. This real-time and comprehensive monitoring function is not available in traditional treatment means, effectively avoiding serious complications such as pneumonia and asphyxia caused by the failure to detect swallowing abnormalities in time, and greatly guaranteeing the safety of the patient.

[0067] Once abnormal swallowing is detected, the AI intelligent analysis layer immediately activates the mis-swallowing detection and safety warning function, quickly identifies abnormal swallowing or choking, and adjusts the FES strategy in time according to the preset strategy, such as pausing or changing the stimulation parameters to prevent the dangerous situation from further deteriorating. At the same time, the entire system strictly follows safety regulations during the electrical stimulation process, from the hardware-level optocoupler isolation MCU and H-bridge, TVS diode to prevent overvoltage damage, current-limiting resistor to ensure safe current, to the software-level real-time closed-loop control, fatigue management, etc., to fully protect the safety of patients during treatment, so that patients can receive rehabilitation treatment with peace of mind.

[0068] 4. Convenient telemedicine support

[0069] The cloud platform remote monitoring and data storage module has built a powerful telemedicine platform. When patients undergo swallowing rehabilitation training at home or other places, the data collection terminal collects and encrypts key data such as EEG, sEMG, FES parameters, etc. and uploads them to the cloud in real time. No matter where the doctor is, he can view the patient's data remotely through the Web terminal or APP, and understand the patient's rehabilitation progress in real time, such as swallowing success rate, muscle activation, FES dependence and other detailed information. Based on these real-time data, doctors can accurately judge the patient's current status and adjust the FES plan remotely in time, such as increasing the stimulation intensity, optimizing the stimulation frequency, etc., to ensure that the treatment plan always fits the patient's rehabilitation needs. This telemedicine model breaks the limitations of time and space, making it convenient for patients to receive professional treatment guidance anytime and anywhere, avoiding the inconvenience of patients frequently traveling to and from the hospital, which is especially significant for patients with limited mobility or living in remote areas.

[0070] The human-computer interaction interface display and printing module further optimizes the doctor-patient interaction experience. The doctor-side interface is rich in functions. In addition to real-time data monitoring and FES remote control, it can also automatically generate detailed swallowing rehabilitation reports, and support export and printing in multiple formats such as PDF and Excel, which is convenient for doctors to record and analyze the patient's rehabilitation status, and provide strong support for the formulation of scientific and reasonable follow-up treatment plans. The patient side uses mobile APP and touch screen devices to realize the visualization and personalization of swallowing training guidance, such as real-time swallowing feedback to intuitively display the swallowing success with the growth of the progress bar, AI provides targeted swallowing skills guidance, and receives remote messages and voice guidance from doctors. This convenient and intuitive human-computer interaction interface improves patients' compliance and confidence in rehabilitation training, promotes patients to actively cooperate with treatment, and further improves the rehabilitation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a system schematic diagram of the present invention.

[0072] Figure 2 It is the working flow diagram of the equipment of the present invention.

[0073] Figure 3 Partial enlargement of the device working flowchart of the present invention Figure 1 。

[0074] Figure 4 Partial enlargement of the device working flowchart of the present invention Figure 2 。

[0075] Figure 5 Main body circuit diagram of the device of the present invention.

[0076] Figure 6 Partial enlargement of the main body circuit diagram of the device of the present invention Figure 1 。

[0077] Figure 7 Partial enlargement of the main body circuit diagram of the device of the present invention Figure 2 。 Specific implementation manners

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0079] The present invention provides a brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke as shown in Figures 1-7 , including:

[0080] An intelligent swallowing brain-computer interface (BCI) module, which is used to monitor and collect multi-modal physiological signals, decode swallowing intentions, and construct a prediction model driven by objective data. The multi-modal physiological signals include electroencephalogram signals collected from the frontal, parietal, temporal, and occipital brain regions using a portable wireless wearable electroencephalogram signal acquisition device, and surface electromyogram signals collected from the mylohyoid muscle and digastric muscle using a surface electromyogram acquisition device. The collected electroencephalogram signals are denoised using band-pass filtering, ICA, or wavelet transform, and the electromyogram signals are subjected to band-pass filtering, power frequency denoising, and short-time smoothing processing. Then, the time-domain features, frequency-domain features, time-frequency features, mutual information, or correlation analysis features of the electroencephalogram signals, and the time-domain features, frequency-domain features, and time-frequency features of the electromyogram signals are extracted respectively. The features extracted from the electroencephalogram and electromyogram are spliced into a single vector through feature-level fusion and input into a machine learning model, or decision-level fusion is used to train the electroencephalogram and electromyogram prediction models respectively and the prediction results are fused using the weighted voting method to identify swallowing intentions. Finally, the electroencephalogram data in the subject's BCI computer is defined as "on" or "off" of the stimulus by the prediction model and an instruction is sent;

[0081] Intelligent Swallowing Functional Electrical Stimulation (FES) Module, which can intelligently sense, adapt to, and optimize the autonomous adjustment of various stimulation parameters. The stimulation parameters include the stimulation site, where the negative electrode is placed at the center of the thyrohyoid muscle and the mylohyoid muscle belly, and the positive electrode is placed at the tendon of the corresponding muscle. The stimulation waveforms include an asymmetric biphasic wave for stimulating small muscles and a symmetric biphasic wave for stimulating large muscles. The stimulation frequency is between 30 - 50 Hz, the stimulation amplitude uses a 1 - 5 mA current stimulation and is slightly adjusted as the swallowing force increases. The stimulation pulse width is between 200 - 500 μs, and the duty cycle is 30%. The treatment time is 30 min to 1 h per session, five times a week to several times a day, and a total of about 40 h over 2 weeks to 3 months. This module also includes a structure for adaptive intelligent regulation consisting of a sensing system, a signal processing and intelligent analysis system, and an electrical stimulation execution system, as well as a feedback and adjustment unit for adaptive intelligent regulation, which is used to dynamically optimize the FES stimulation parameters according to the feedback data; AI-assisted BCI-FES Online Integration Module, which combines artificial intelligence, brain-computer interface, biological signal processing, and FES to build an efficient closed-loop system to achieve non-invasive brain-controlled swallowing. Patients only need to imagine the swallowing action to trigger FES. It has high-precision AI regulation to optimize swallowing recognition based on deep learning, achieving real-time closed-loop regulation including swallowing monitoring + FES stimulation + AI optimization of the entire process closed-loop control, and having personalized adaptation ability to automatically adjust the training plan according to the patient's swallowing recovery progress. Its structure includes a brain-computer interface system for enhancing swallowing recognition, accurately identifying swallowing intent, and synchronously activating swallowing muscle groups. The artificial intelligence regulation system includes a swallowing pattern recognition algorithm, adaptive FES parameter optimization, and a closed-loop control strategy. The FES execution system provides multi-channel FES output, can adjust the stimulation parameters personalizedly, and receives instructions through wireless transmission to execute stimulation, and uses AI to optimize the FES parameters. By supervised learning to collect EEG, EMG, and swallowing success rate data, use SVM, XGBoost, and Random Forest to train the FES parameter prediction model to achieve real-time inference and provide personalized parameters. Use reinforcement learning to adaptively and dynamically regulate the FES parameters through reinforcement learning of state, action, and reward, automatically trigger FES to improve the autonomous swallowing ability, continuously optimize FES to reduce the number of stimulations, improve the autonomous swallowing rate, and maintain and promote swallowing recovery;

[0082] Closed-loop Intelligent Feedback Regulation Module, including:

[0083] Signal Acquisition Layer, which real-time monitors multi-modal biological signals related to swallowing, including electroencephalogram sensors to collect swallowing intent signals to achieve BCI control of FES, surface electromyogram sensors to detect the electrical activities of swallowing-related muscle groups to evaluate muscle responses, acceleration / pressure sensors to monitor laryngeal and hyoid bone movements to judge swallowing quality, pharyngeal pressure sensors to measure the pressure during swallowing to assist in the analysis of swallowing success rate, and the aspiration / cough detection module to identify swallowing abnormalities by combining respiratory patterns and voice analysis;

[0084] The AI intelligent analysis layer, based on AI technologies such as deep learning, pattern recognition, and reinforcement learning, realizes swallowing intention recognition and stimulation regulation optimization, including swallowing pattern recognition model to improve swallowing detection accuracy based on EEG and sEMG signals, mis-swallowing detection and safety warning to identify abnormal swallowing or choking and adjust FES strategy in time, adaptive electrical stimulation optimization to use machine learning to adjust FES parameters to optimize stimulation effect, and personalized training strategy to dynamically adjust swallowing training mode according to the patient's rehabilitation progress;

[0085] The FES execution layer, based on the AI analysis results, accurately controls the electrical stimulation of the swallowing-related muscles, including the electrical stimulation unit's precise electrical stimulation of the digastric and hyoid muscles through a multi-channel FES system, and dynamically adjusts the FES stimulation frequency, pulse width, and intensity to ensure the best swallowing recovery effect. Timing synchronization optimization ensures FES stimulation

[0086] Stimulate the swallowing rhythm in accordance with the natural swallowing rhythm to improve swallowing efficiency;

[0087] Cloud platform remote monitoring and data storage module, establish cloud data storage and remote monitoring system, and record BCI-FES swallowing rehabilitation data in real time, including:

[0088] The data collection end collects and uploads the patient's swallowing rehabilitation data, including EEG for swallowing intention recognition, sEMG for swallowing quality assessment, and FES parameters;

[0089] Cloud data storage, secure storage of patient swallowing data, establishment of a time series database to store EEG / EMG continuous signals, a relational database to store FES parameters and training records, a non-relational database to store patient feedback and doctor advice, and data encryption access control;

[0090] AI analysis and remote monitoring, based on AI optimization of FES parameters, provide personalized rehabilitation plans, use supervised learning to predict the optimal FES parameters, strengthen learning to dynamically adjust FES stimulation, automatically calculate swallowing success rate, muscle activation, FES dependence, and generate personalized rehabilitation curves for doctors to view detailed data;

[0091] Doctors and patients can interact with each other by viewing patient data remotely through the Web or APP, adjusting FES plans, monitoring patient swallowing data in real time, adjusting FES parameters remotely, and providing voice / video rehabilitation guidance. Patients can visualize their swallowing progress, obtain AI training suggestions, and provide voice / video feedback to record their training experience through the mobile APP.

[0092] Human-computer interaction interface display and printing module, which establishes a visual interaction interface, displays swallowing training data in real time, generates personalized rehabilitation curves, and supports data export / printing. It includes that on the doctor side, through the Web and APP, patient management, real-time data monitoring, FES remote control, automatic generation of swallowing rehabilitation reports and support for export / printing are realized. On the patient side, through the mobile APP and touch screen devices, swallowing training guidance, visualization of training results, and obtaining doctor messages are achieved.

[0093] In the intelligent swallowing brain-computer interface (BCI) module, the electroencephalogram signal processing further includes: decoding the electroencephalogram signals recorded from the motor cortex (C3 / C4 / FCZ), extracting meaningful features, namely swallowing intentions, and using machine learning and deep learning for classification to generate a general prediction model and establishing a personalized prediction model through transfer learning.

[0094] In the intelligent swallowing functional electrical stimulation (FES) module, in the structural composition adapted for intelligent regulation:

[0095] The surface electromyogram sensor in the sensing system detects the electrical activities of swallowing-related muscles such as the digastric muscle and the hyoid muscle, identifies swallowing intentions, the acceleration / pressure sensor monitors the movements of the hyoid bone and the larynx to confirm swallowing actions, and the physiological signal monitors the heart rate and respiration to ensure stimulation safety;

[0096] In the signal processing and intelligent analysis system, the bio-signal preprocessing module filters, denoises, and normalizes the collected signals. The swallowing pattern recognition module uses deep learning or machine learning algorithms to identify swallowing actions and abnormal swallowing conditions. The real-time feedback control module adaptively adjusts the electrical stimulation parameters based on the swallowing state to achieve closed-loop control. The data storage and analysis module records the user's swallowing data and provides personalized rehabilitation suggestions;

[0097] In the electrical stimulation execution system, the multi-channel FES output unit can independently adjust the stimulation parameters of different muscle groups to ensure coordinated control, can adjust the stimulation parameters to adapt to patients with different degrees of swallowing disorders, and the closed-loop feedback adjustment dynamically optimizes the stimulation strategy according to the sensor data to improve the training effect. The wireless / portable stimulation device improves the convenience and comfort of patient use.

[0098] In the AI-assisted BCI-FES online integration module, the closed-loop control strategy of the artificial intelligence regulation system is: real-time monitoring of swallowing effects, adjusting the stimulation plan according to the monitoring results, and optimizing the process of swallowing function recovery. Specifically, by continuously obtaining and analyzing the data feedback of EEG, sEMG, and pressure sensors, the FES stimulation parameters are dynamically adjusted to achieve the best swallowing training effect.

[0099] In the closed-loop intelligent feedback regulation module:

[0100] The swallowing intention recognition and triggering function specifically combines electromyogram signals and laryngeal movement data. BCI+AI improves the accuracy of swallowing recognition and triggers FES electrical stimulation in real time to promote the recovery of swallowing function.

[0101] The intelligent FES parameter adaptive optimization function automatically adjusts the FES intensity, frequency, and pulse timing by AI based on biofeedback data, and adopts a personalized training mode to gradually improve the patient's swallowing ability and enhance the effect of swallowing training.

[0102] The closed-loop control and real-time feedback function also includes aspiration detection and safety mechanisms. The swallowing abnormality is monitored in real time through the aspiration / cough detection module. Once an abnormality is detected, the FES strategy is immediately adjusted, and the stimulation parameters are paused or changed to ensure the safety of swallowing training.

[0103] The rehabilitation data analysis and remote monitoring function further records the swallowing training data to generate a detailed personalized rehabilitation report. The report content covers the change trend of swallowing success rate, FES dependence analysis, and muscle activation degree evaluation. Doctors can view the report online through the remote monitoring function and accurately adjust the training plan to improve the efficiency of rehabilitation management.

[0104] In the cloud platform remote monitoring and data storage module:

[0105] When the data acquisition terminal collects data, the data of EEG, sEMG, and FES parameters are encrypted in real time and uploaded to the cloud platform using a secure transmission protocol to ensure the security and integrity of the data during the transmission process.

[0106] The time series database for cloud data storage uses an efficient data compression algorithm to reduce storage occupancy, and at the same time ensures the integrity of the time series characteristics of the data, which is convenient for accurately extracting dynamic characteristics when AI analyzes swallowing patterns later.

[0107] In the AI analysis and remote monitoring link, when the supervised learning predicts the optimal FES parameters, it comprehensively analyzes the patient's historical training data and current physiological state data. When the reinforcement learning dynamically adjusts the FES stimulation, more environmental factors such as the patient's fatigue degree and diet on the day are introduced as adjustment bases, so that the optimization of FES stimulation parameters is more in line with the actual needs of the patient, and the generated personalized rehabilitation curve has higher reference value.

[0108] In the human-computer interaction interface display and printing module:

[0109] Under the real-time data monitoring function of the doctor terminal, in addition to visually displaying EEG / EMG / FES data, it also provides a data comparison and analysis tool, which can compare the current patient data with the average data of patients in the same type of rehabilitation stage or historical rehabilitation data to assist doctors in more accurately judging the patient's rehabilitation progress.

[0110] In the swallowing training guidance function of the patient terminal, according to the patient's personal swallowing habits and rehabilitation progress, the AI dynamically adjusts the visual feedback and voice prompt content, such as changing the prompt frequency and optimizing the prompt screen, to make the training guidance more targeted and improve the patient's training compliance.

[0111] It also includes the circuit of the swallowing neuromuscular electrical stimulation system, which is mainly composed of power management, electrode drive, current / voltage modulation, and part of the control unit. It uses low-frequency pulsed current to stimulate the laryngeal muscles to promote the recovery of swallowing function, where:

[0112] The power supply module provides multiple voltages. The STM32 control unit requires 3.3V / 5V, the AD9833 DDS chip requires 3.3V, the operational amplifier signal conditioning requires ±12V, and the H-bridge drive current amplification requires ±5V / ±12V;

[0113] The signal generation module is controlled by the STM32 through SPI to control the AD9833 to generate a square wave signal with an adjustable frequency in the range of 20Hz - 50Hz. The square wave output is 0V - 3.3V, and 16-bit SPI data transmission is used. After the single-pole 0 - 3.3V square wave is output, it is amplified and converted into a bipolar ±5V square wave;

[0114] The signal conditioning module uses the operational amplifier OPA2134 to achieve signal amplification and bias adjustment, making the output become ±3.3V, and then further amplifying it to ±5V through the H-bridge;

[0115] The H-bridge circuit selects the H-bridge chip DRV8871, and changes the direction of current flow by controlling the switch. When a high level is input, the output is +5V, and when a low level is input, the output is -5V, forming a bidirectional symmetric square wave;

[0116] The skin electrode & protection module has isolation protection. It uses the optocoupler TLP250 to isolate the MCU and the H-bridge. The TVS diode 5V prevents overvoltage from harming the patient, and the current-limiting resistor 1kΩ ensures a safe current. The output frequency is 20 - 50Hz, the square wave amplitude is ±5V, and the output current is 0 - 10mA.

[0117] The circuit of the swallowing neuromuscular electrical stimulation system has the following operation process:

[0118] First, power is supplied. The STM32, AD9833, and H-bridge work together to provide the power required for each module;

[0119] Then, the STM32 controls the AD9833 through SPI to generate a square wave signal;

[0120] Next, the operational amplifier conditions the signal and converts the output of the AD9833 into ±5V;

[0121] Then, the H-bridge amplifies the signal to generate a bipolar square wave;

[0122] Final electrode output, with isolation & protection, stimulating the swallowing muscles.

[0123] Its working mechanism is to combine a non-invasive brain-computer interface based on electroencephalogram data with a non-invasive FES system for swallowing, using artificial intelligence algorithms and large model technologies as the core of monitoring, regulation, and feedback technologies to achieve effective swallowing function movements. At the same time, it induces the plastic reconstruction and functional reorganization of the damaged central nervous system motor neural network, prompting the patient to quickly recover to the natural and normal physiological state of autonomous and coordinated swallowing function, using the stronger swallowing movement intention of the brain to complete the swallowing movement imagination task, and stimulating the corresponding cerebral cortex to activate the neuroplasticity of the brain.

[0124] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke, characterized in that, Including: An intelligent swallowing brain-computer interface module, which is used to monitor and collect multi-modal physiological signals, decode swallowing intentions, and construct a prediction model driven by objective data. The multi-modal physiological signals include electroencephalogram (EEG) signals collected from the frontal, parietal, temporal, and occipital brain regions using a portable wireless wearable EEG signal acquisition device, and surface electromyogram (sEMG) signals collected from the geniohyoid muscle and digastric muscle using a surface EMG acquisition device. The collected EEG signals are denoised using band-pass filtering, independent component analysis (ICA), or wavelet transform, and the EMG signals are subjected to band-pass filtering, power frequency denoising, and short-term smoothing processing. Then, the time-domain features, frequency-domain features, time-frequency features, mutual information, or correlation analysis features of the EEG signals, and the time-domain features, frequency-domain features, and time-frequency features of the EMG signals are extracted respectively. The features extracted from the EEG and EMG are concatenated into a single vector through feature-level fusion and input into a machine learning model, or decision-level fusion is used to train the EEG and EMG prediction models respectively and the prediction results are fused using the weighted voting method to identify swallowing intentions. Finally, the EEG data in the subject's BCI computer is defined as "on" or "off" stimulation by the prediction model and an instruction is sent; An intelligent swallowing functional electrical stimulation module, which can intelligently sense, adapt to, and optimize the autonomous adjustment of various stimulation parameters. The stimulation parameters include the stimulation site, where the negative electrode is placed at the center of the geniohyoid muscle and mylohyoid muscle bellies, and the positive electrode is placed at the tendon of the corresponding muscle. The stimulation waveforms include asymmetric biphasic waves for stimulating small muscles and symmetric biphasic waves for stimulating large muscles. The stimulation frequency is between 30 - 50 Hz, the stimulation amplitude is 1 - 5 mA current stimulation and is slightly adjusted as the swallowing force increases, the stimulation pulse width is between 200 - 500 μs, the duty cycle is 30%, the treatment time is 30 min to 1 h per session, five times a week to several times a day, and the total treatment time is about 40 h in 2 to 3 months. This module also includes a structure for adaptive intelligent regulation composed of a sensing system, a signal processing and intelligent analysis system, and an electrical stimulation execution system, as well as an adaptive intelligent regulation feedback and adjustment unit for dynamically optimizing the FES stimulation parameters according to the feedback data; AI-assisted BCI-FES online integration module, which combines artificial intelligence, brain-computer interface, bio-signal processing, and FES to build an efficient closed-loop system, enabling non-invasive brain-controlled swallowing. Patients only need to imagine the swallowing movement to trigger FES. It has high-precision AI regulation to optimize swallowing recognition based on deep learning, achieving real-time closed-loop regulation including swallowing monitoring + FES stimulation + AI optimization of the entire process closed-loop control. It has personalized adaptation ability to automatically adjust the training plan according to the patient's swallowing recovery progress. Its structure includes a brain-computer interface system for enhancing swallowing recognition, accurately identifying swallowing intention, and synchronously activating swallowing muscle groups. The artificial intelligence regulation system includes swallowing pattern recognition algorithms, adaptive FES parameter optimization, and closed-loop control strategies. The FES execution system provides multi-channel FES output, can adjust the stimulation parameters personalizedly, and receives instructions through wireless transmission to execute stimulation. And it uses AI to optimize FES parameters, collects EEG, EMG, and swallowing success rate data through supervised learning, uses SVM, XGBoost, and Random Forest to train the FES parameter prediction model to achieve real-time inference and provide personalized parameters, uses reinforcement learning through reinforcement learning of state, action, and reward, adopts the deep Q-network algorithm to adaptively and dynamically regulate FES parameters, automatically triggers FES to improve the ability of autonomous swallowing, continuously optimizes FES to reduce the number of stimulations, improves the rate of autonomous swallowing, and maintains and promotes swallowing recovery; Closed-loop intelligent feedback regulation module, including: Signal acquisition layer, which monitors multi-modal bio-signals related to swallowing in real time, including electroencephalogram sensors to collect swallowing intention signals to achieve BCI control of FES, surface electromyogram sensors to detect the electrical activities of swallowing-related muscle groups to evaluate muscle responses, acceleration / pressure sensors to monitor laryngeal and hyoid bone movements to judge swallowing quality, pharyngeal pressure sensors to measure the pressure during swallowing to assist in the analysis of swallowing success rate, and aspiration / cough detection module to identify swallowing abnormalities by combining respiratory patterns and voice analysis; AI intelligent analysis layer, based on AI technologies such as deep learning, pattern recognition, and reinforcement learning, realizes swallowing intention recognition and stimulation regulation optimization, covering swallowing pattern recognition models to improve the accuracy of swallowing detection based on EEG and sEMG signals, aspiration detection and safety warning to identify abnormal swallowing or cough and adjust FES strategies in time, adaptive electrical stimulation optimization to adjust FES parameters using machine learning to optimize the stimulation effect, and personalized training strategies to dynamically adjust the swallowing training mode according to the patient's rehabilitation progress; FES execution layer, according to the AI analysis results, accurately controls the electrical stimulation of swallowing-related muscle groups, including electrical stimulation units to accurately stimulate the digastric muscle and hyoid muscle through a multi-channel FES system, dynamic parameter regulation to adjust the stimulation frequency, pulse width, and intensity of FES to ensure the best swallowing recovery effect, and timing synchronization optimization to ensure that the FES stimulation is consistent with the natural swallowing rhythm to improve swallowing efficiency; Cloud platform remote monitoring and data storage module, establishing a cloud data storage and remote monitoring system, and recording BCI-FES swallowing rehabilitation data in real time, including: The data acquisition terminal collects and uploads the patient's swallowing rehabilitation data, including EEG for swallowing intention recognition, sEMG for swallowing quality assessment, and FES parameters; Cloud data storage securely stores the patient's swallowing data, establishes a time-series database to store the continuous signals of EEG / EMG, a relational database to store FES parameters and training records, and a non-relational database to store patient feedback and doctor suggestions, and performs data encryption access control; AI analysis and remote monitoring optimize FES parameters based on AI, provide personalized rehabilitation programs, use supervised learning to predict the optimal FES parameters, use reinforcement learning to dynamically adjust FES stimulation, automatically calculate the swallowing success rate, muscle activation degree, and FES dependence, and generate personalized rehabilitation curves for doctors to view detailed data; Doctor and patient interaction terminals allow doctors to remotely view patient data through the Web and APP, adjust the FES program, real-time monitor the patient's swallowing data, remotely adjust FES parameters, and provide voice / video rehabilitation guidance. Patients can visualize their swallowing progress through the mobile APP, obtain AI training suggestions, and record their training feelings through voice / video feedback; The human-computer interaction interface display and printing module establishes a visual interaction interface, real-time displays swallowing training data, generates personalized rehabilitation curves, and supports data export / printing. For the doctor side, through the Web and APP, it can achieve patient management, real-time data monitoring, remote FES regulation, automatically generate swallowing rehabilitation reports, and support export / printing. For the patient side, through the mobile APP and touch-screen devices, it can achieve swallowing training guidance, visualization of training results, and obtain doctor messages.

2. The brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke according to claim 1, wherein: In the intelligent swallowing brain-computer interface module, the electroencephalogram signal processing further includes: decoding the electroencephalogram signals recorded from the motor cortex, extracting meaningful features, i.e., swallowing intention, and using machine learning and deep learning for classification to generate a general prediction model and using transfer learning to establish a personalized prediction model.

3. The brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke according to claim 1, wherein: In the intelligent swallowing functional electrical stimulation module, the structural composition for adaptive intelligent regulation includes: In the sensing system, the surface electromyography sensor detects the electrical activities of swallowing-related muscles such as the digastric muscle and hyoid muscle to identify swallowing intention, the acceleration / pressure sensor monitors the movement of the hyoid bone and larynx to confirm swallowing actions, and the physiological signal monitors heart rate and respiration to ensure stimulation safety; In the signal processing and intelligent analysis system, the bio-signal preprocessing module filters, denoises, and normalizes the collected signals. The swallowing pattern recognition module uses deep learning or machine learning algorithms to identify swallowing actions and abnormal swallowing conditions. The real-time feedback control module adaptively adjusts the electrical stimulation parameters based on the swallowing state to achieve closed-loop control. The data storage and analysis module records the user's swallowing data and provides personalized rehabilitation suggestions; In the electrical stimulation execution system, the multi-channel FES output unit can independently adjust the stimulation parameters of different muscle groups to ensure coordinated control, can adjust the stimulation parameters to adapt to patients with different degrees of swallowing disorders, and the closed-loop feedback adjustment dynamically optimizes the stimulation strategy according to the sensor data to improve the training effect. The wireless / portable stimulation device improves the convenience and comfort of patient use.

4. A brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke, characterized in that: In the described AI-assisted BCI-FES online integration module, the closed-loop control strategy of the artificial intelligence regulation system is as follows: continuously monitor the swallowing effect, adjust the stimulation plan according to the monitoring results, and optimize the swallowing function recovery process. Specifically, by continuously acquiring and analyzing the data feedback of EEG, sEMG, and pressure sensors, dynamically adjust the FES stimulation parameters to achieve the best swallowing training effect.

5. The brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke according to claim 1, wherein: In the described closed-loop intelligent feedback regulation module: The swallowing intention recognition and triggering function is specifically as follows. By combining the myoelectric signal and laryngeal movement data, BCI+AI improves the accuracy of swallowing recognition and triggers the FES electrical stimulation in real time to promote the recovery of swallowing function. The intelligent FES parameter adaptive optimization function automatically adjusts the FES intensity, frequency, and pulse timing based on the biofeedback data by AI, and adopts a personalized training mode to gradually improve the patient's swallowing ability and enhance the swallowing training effect. The closed-loop control and real-time feedback function also includes aspiration detection and safety mechanism. The aspiration / coughing detection module continuously monitors swallowing abnormalities in real time. Once an abnormality is detected, immediately adjust the FES strategy, pause or change the stimulation parameters to ensure the safety of swallowing training. The rehabilitation data analysis and remote monitoring function further includes recording the swallowing training data to generate a detailed personalized rehabilitation report. The report content covers the change trend of swallowing success rate, FES dependence analysis, and muscle activation degree assessment. Doctors can view the report online through the remote monitoring function and accurately adjust the training plan to improve the rehabilitation management efficiency.

6. The brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke according to claim 1, wherein: In the described cloud platform remote monitoring and data storage module: When the data acquisition terminal collects data, it encrypts the data of EEG, sEMG, and FES parameters in real time and uploads them to the cloud platform using a secure transmission protocol to ensure the security and integrity of the data transmission process. The time series database for cloud data storage uses an efficient data compression algorithm to reduce storage occupancy, while ensuring the integrity of the time series characteristics of the data, which is convenient for accurately extracting dynamic characteristics when AI analyzes the swallowing pattern later. In the AI analysis and remote monitoring link, when the supervised learning predicts the optimal FES parameters, it comprehensively analyzes the patient's historical training data and current physiological state data. When the reinforcement learning dynamically adjusts the FES stimulation, more environmental factors such as the patient's fatigue degree and diet on the day are introduced as adjustment bases, so that the optimization of the FES stimulation parameters is more in line with the actual needs of the patient, and the generated personalized rehabilitation curve has higher reference value.

7. A brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke, characterized in that: In the described human-computer interaction interface display and printing module: Under the real-time data monitoring function of the doctor terminal, in addition to visually displaying the EEG / EMG / FES data, it also provides a data comparison and analysis tool that can compare the current patient data with the average data of patients in the same type of rehabilitation stage or historical rehabilitation data to assist doctors in more accurately judging the patient's rehabilitation progress. In the swallowing training guidance function of the patient terminal, according to the patient's personal swallowing habits and rehabilitation progress, AI dynamically adjusts the visual feedback and voice prompt content, such as changing the prompt frequency and optimizing the prompt screen, to make the training guidance more targeted and improve the patient's training compliance.

8. A brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke according to any one of claims 1-7, characterized in that: It also includes the circuit of the swallowing neuromuscular electrical stimulation system, which mainly consists of parts such as power management, electrode drive, current / voltage modulation, and control unit. It uses low-frequency pulsed current to stimulate the laryngeal muscles to promote the recovery of swallowing function, where: The power supply module provides multiple voltages. The STM32 control unit requires 3.3V / 5V, the AD9833 DDS chip requires 3.3V, the operational amplifier signal conditioning requires ±12V, and the H-bridge drive current amplification requires ±5V / ±12V; The signal generation module is controlled by the STM32 through SPI to control the AD9833 to generate a square wave signal. The frequency is adjustable from 20Hz to 50Hz. The square wave outputs 0V - 3.3V. It uses 16-bit SPI data transmission. After outputting a unipolar 0 - 3.3V square wave, it is amplified and converted into a bipolar ±5V square wave; The signal conditioning module uses the operational amplifier OPA2134 to achieve signal amplification and bias adjustment, making the output become ±3.3V, and then further amplifying it to ±5V through the H-bridge; The H-bridge circuit selects the H-bridge chip DRV8871. By controlling the switch, the direction of current flow is changed. When a high level is input, the output is +5V, and when a low level is input, the output is -5V, forming a bidirectional symmetric square wave; The skin electrode & protection module has isolation protection. It uses the optocoupler TLP250 to isolate the MCU and the H-bridge. The TVS diode 5V prevents overvoltage from harming the patient, and the current-limiting resistor 1kΩ ensures a safe current. The output frequency is 20 - 50Hz, the square wave amplitude is ±5V, and the output current is 0 - 10mA.

9. A brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke, characterized in that: The circuit of the swallowing neuromuscular electrical stimulation system in the operation process: First, power is supplied. The STM32, AD9833, and H-bridge work together to provide the power required by each module; Then the STM32 controls the AD9833 through SPI to generate a square wave signal; Then the operational amplifier conditions the signal to convert the output of the AD9833 into ±5V; Then the H-bridge amplifies the signal to generate a biphasic square wave; Finally, the electrode outputs, with isolation & protection, to stimulate the swallowing muscles.

10. A brain-machine hybrid intelligent closed-loop swallowing rehabilitation system for treating dysphagia after stroke according to any one of claims 1-9, characterized in that: Its working mechanism is to combine a non-invasive brain-computer interface based on electroencephalogram data with a non-invasive FES system for swallowing. It uses artificial intelligence algorithms and large model technologies as the core of monitoring, regulation, and feedback technologies to achieve effective swallowing function movements. At the same time, it induces the plastic reconstruction and functional reorganization of the damaged central nervous system motor neural network, prompting the patient to quickly recover to the natural and normal physiological state of autonomous and coordinated swallowing function, and using the stronger swallowing movement intention of the brain to complete the swallowing movement imagination task, stimulating the corresponding cerebral cortex and activating the neural plasticity of the brain.

Images