Transdermal electrical stimulation equipment based on large language model

Abstract

The invention discloses a transcutaneous electrical stimulation device based on a large language model, and relates to the technical field of wearable devices. The transcutaneous electrical stimulation equipment based on the large language model comprises a data acquisition and preprocessing module, a feature extraction and fusion module, a model training and prediction module and a real-time information processing and response module, real-time data processing and intervention capabilities are integrated, human body data can be analyzed in real time, electrical stimulation parameters can be automatically adjusted, an instant treatment or adjustment scheme is provided for the problems of arrhythmia, chronic pain and the like, and the timeliness and effectiveness of treatment are greatly improved. And secondly, by fusing the physiological signal features and the symptom text features, the equipment can more comprehensively understand the health condition of the user, so that a more accurate treatment scheme is provided. Besides, the equipment also has the functions of anomaly detection and trend analysis, can predict potential health risks, and triggers an alarm in time when a serious anomaly is detected, thereby effectively guaranteeing the life safety of the user.

Description

Technical Field

[0001] The present invention relates to the technical field of wearable devices, and in particular to a transcutaneous electrical stimulation device based on a large language model. Background Art

[0002] With the rapid development of wearable device technology, watches are not only used to display time, but also gradually have the functions of health monitoring and disease prevention. Traditional health monitoring watches mainly focus on the collection of data, such as heart rate, activity level, etc., but the in-depth analysis and application of these data in medical diagnosis and treatment are still relatively limited. In addition, existing devices have not yet formed an effective solution for chronic disease management and real-time medical intervention. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a transcutaneous electrical stimulation device based on a large language model, which solves the technical defects of existing wearable devices.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: A transcutaneous electrical stimulation device based on a large language model, comprising:

[0005] The data acquisition and preprocessing module collects human body data through sensors on the transcutaneous electrical stimulation device, and records and recognizes symptom descriptions through voice input, performs denoising, normalization and text cleaning on the collected data, and converts it into vector representation;

[0006] Feature extraction and fusion module, used to extract physiological signal features and symptom text features, and use a hybrid neural network model for feature fusion;

[0007] Model training and prediction module, builds a deep learning model using a fully connected layer to predict electrical stimulation parameters, and uses the extracted features to train the model;

[0008] The real-time information processing and response module continuously collects and preprocesses physiological signals, extracts key physiological parameters from them, and then performs anomaly detection and trend analysis to predict potential health risks. It automatically adjusts electrical stimulation parameters based on the monitoring results and triggers an alarm when serious health abnormalities are detected.

[0009] Preferably, the work contents of the data acquisition and preprocessing module specifically include:

[0010] S1.1 Data input:

[0011] Sensor data collection: Use electrocardiogram (ECG), photoelectric volumetric pulse wave (PPG), skin electrical activity (EDA) and temperature sensor devices to capture the user's physiological signals in real time;

[0012] Symptom voice input: Users input their symptom descriptions through the interactive module, and the large language model recognizes the symptom descriptions and converts them into standard instructions;

[0013] S1.2 Data preprocessing:

[0014] De-noising: Apply low-pass filter or median filter techniques to physiological signals to eliminate background noise;

[0015] Normalization processing: adjust the physiological signals to a uniform scale;

[0016] Text cleaning and vectorization: Preprocess the symptom text to remove irrelevant information such as stop words and punctuation marks, and use NLP technology to convert the symptom text into a vector representation.

[0017] Preferably, the working contents of the feature extraction and fusion module specifically include:

[0018] S2.1 Feature extraction:

[0019] Physiological signal feature extraction: Extract key features from physiological signals of heart rate and blood pressure. The extracted physiological signal features include:

[0020] Average heart rate: Calculate the average heart rate over a period of time. The formula is: Where n is the number of sampling points;

[0021] Heart rate variability HRV: measures the fluctuation of heart rate over a set period of time, expressed as standard deviation:

[0022] Mean blood pressure: Calculate the mean of systolic and diastolic blood pressure separately:

[0023]

[0024] Symptom text feature extraction: extract sentiment or keyword importance scores related to specific symptoms from text vectors;

[0025] S2.2 Multimodal data fusion: A hybrid neural network model, such as a multi-input neural network, is used to combine the long short-term memory network LSTM to process time series data (such as heart rate, blood pressure) and the Transformer to process text data (such as symptom description). The features of different modalities are fused through the fusion layer to form a unified feature representation. The fusion process is expressed as:

[0026] h=Concat(LSTM(xhr,xbp),Transformer(xsymptoms));

[0027] Among them, Xhr, Xbp, and Xsymptomsx represent heart rate characteristics, blood pressure characteristics, and symptom characteristics, respectively.

[0028] Preferably, the working contents of the model training and prediction module specifically include:

[0029] S3.1 Model design: A deep learning model is constructed, and a fully connected layer Dense Layer is used to predict the electrical stimulation parameters. The ReLU activation function is used to ensure the non-negativity of the output. The model structure is expressed as follows:

[0030] Output = ReLU(Dense(fused features));

[0031] Among them, the Dense layer represents the fully connected layer, where the weights and biases are optimized by the back-propagation algorithm during the training process;

[0032] S3.2 Loss function selection: The mean square error (MSE) is used as the loss function to measure the difference between the predicted electrical stimulation parameters and the true values. The MSE formula is:

[0033]

[0034] Among them, m is the number of samples, y i is the actual electrical stimulation parameter, y^ i are the parameters predicted by the model;

[0035] S3.3 Model training: The model is trained using the extracted features, with the goal of predicting the optimal electrical stimulation parameters (intensity, frequency, duration).

[0036] Preferably, a feedback mechanism is introduced during the model training process to achieve intelligent adjustment, including:

[0037] S3.3.1 Initial prediction: Use the trained model to extract and fuse features of new physiological signals and symptom texts, and then predict the initial electrical stimulation parameters;

[0038] S3.3.2 Effect evaluation: Apply the predicted electrical stimulation parameters to the patient and collect physiological signals and symptom feedback after treatment;

[0039] S3.3.3 Parameter adjustment: Based on the treatment effect (such as the degree of symptom relief, improvement of physiological indicators, etc.), the model is fine-tuned using reinforcement learning or Bayesian optimization methods to more accurately predict future electrical stimulation parameters;

[0040] S3.3.4 Iterative training: Use the adjusted parameters and treatment effects as new training data to retrain the model to improve its generalization ability and prediction accuracy.

[0041] Preferably, the working contents of the real-time information processing and response module specifically include:

[0042] S4.1 Real-time data collection and processing:

[0043] Real-time acquisition and processing: continuously collect physiological signals, perform filtering and denoising, and improve data quality;

[0044] Real-time feature extraction: extract key physiological parameters from real-time data;

[0045] S4.2 Real-time health monitoring and response:

[0046] Anomaly detection and trend analysis: Use anomaly detection algorithms to identify abnormal patterns in data and perform trend analysis to predict potential health risks;

[0047] Electrical stimulation parameter adjustment: Automatically adjust electrical stimulation parameters according to monitoring results;

[0048] Alerts and notifications: When serious health abnormalities are detected, alarms are triggered and users and healthcare providers are notified;

[0049] S4.3 User feedback collection: After the electrical stimulation is implemented, user feedback on the treatment effect is collected;

[0050] S4.4 Model optimization: Collect user feedback on electrical stimulation therapy (including comfort, treatment effect, etc.); use this feedback to perform online learning or incremental learning of the model, and continuously optimize the prediction and adjustment rules of electrical stimulation parameters.

[0051] Preferably, the model output of the electrical stimulation parameter adjustment includes the intensity I, frequency F and duration T of the electrical stimulation;

[0052] Assume that the output of the deep learning model is the predicted electrical stimulation parameter vector P = [I, F, T];

[0053] According to the real-time monitoring data D and feature extraction results F, the model is expressed as:

[0054] mathbfP=f(F;θ);

[0055] Where f is the deep learning model and θ is the model parameter;

[0056] Then, the parameter adjustment rule is formalized as:

[0057] I new =g I (I, D);

[0058] F new =g F (F, D);

[0059] T new =g T (T, D);

[0060] Among them, g I , g F and g T They are functions for adjusting electrical stimulation parameters I, F, and T according to real-time monitoring data D;

[0061] Parameter verification and adjustment:

[0062] After adjusting the parameters, perform parameter verification to ensure that the new parameter values ​​are within the safe range and meet medical standards;

[0063] If the new parameters do not meet the requirements, they are fine-tuned according to the preset rules until the conditions are met.

[0064] Preferably, the symptom voice input includes:

[0065] Voice input and preprocessing: Users input symptom descriptions through the microphone of the wearable device, and the lightweight STT algorithm is used to convert the voice into text, and then the text is preprocessed, such as removing noise and segmenting words;

[0066] Sentiment analysis and keyword extraction: Use the optimized lightweight LLM to perform sentiment analysis on the preprocessed text, extract sentiment tendencies, and use keyword extraction algorithms (such as TF-IDF, TextRank, etc.) to identify key information in the text;

[0067] Data synchronization and fusion: synchronize the timestamps of data from different sensors (such as heart rate, blood pressure, etc.); use fusion algorithms to integrate sensor data to form a comprehensive health assessment;

[0068] During the electrical stimulation treatment, interaction and feedback are carried out through the interactive module, including: based on the analysis results, showing the user possible disease suggestions or health suggestions through a simplified interface or voice feedback; further interactive options can also be provided, such as asking the user whether he is willing to undergo a more in-depth examination or consult a doctor.

[0069] The present invention also discloses a transcutaneous electrical stimulation device, comprising:

[0070] A strap is used to be tied to the patient's limbs, and the strap is made of elastic material;

[0071] Electrode sheets, multiple of which are evenly arranged on the inner side of the bandage, and the surface of the electrode sheets is provided with a plurality of convex points;

[0072] A connecting structure is provided at both ends of the strap, and is used to connect the two ends of the strap or to connect with the dial of the smart watch;

[0073] The control module is used to connect and control a plurality of electrode sheets.

[0074] Preferably, the connection structure is a snap button or a plug buckle;

[0075] When snaps are installed at both ends of the strap, the two ends can be connected to each other, or multiple straps can be connected in series;

[0076] When buckles are installed at both ends of the strap, they are used to be inserted into the two sides of the smart watch dial.

[0077] The present invention provides a transcutaneous electrical stimulation device based on a large language model. Compared with the prior art, it has the following beneficial effects:

[0078] 1. The transcutaneous electrical stimulation device based on the large language model integrates real-time data processing and intervention capabilities, can instantly analyze human body data and automatically adjust electrical stimulation parameters, and provide immediate treatment or adjustment solutions for problems such as arrhythmia and chronic pain, greatly improving the timeliness and effectiveness of treatment. Secondly, by integrating physiological signal features and symptom text features, the device can more comprehensively understand the user's health status, thereby providing more accurate treatment plans. In addition, the device also has anomaly detection and trend analysis functions, which can predict potential health risks and trigger alarms in time when serious anomalies are detected, effectively protecting the user's life safety. In summary, the device not only improves the accuracy and timeliness of treatment, but also enhances the user's sense of security, bringing revolutionary progress to the field of medical health.

[0079] 2. This transcutaneous electrical stimulation device based on a large language model achieves comprehensive monitoring of the user's health status by capturing multi-source physiological signals and symptom voice input in real time, thereby improving the richness and accuracy of the data. Secondly, the data preprocessing process effectively eliminates noise, ensures the stability of the model input, and provides a high-quality data foundation for subsequent analysis. In terms of feature extraction, not only key physiological parameters are extracted from physiological signals, but also emotional and keyword information is mined from symptom texts, achieving in-depth insights into the user's health status. Finally, multimodal data fusion is performed through a hybrid neural network model, and feature information from different sources is organically integrated to form a unified feature representation, which provides strong support for subsequent model training and prediction. These improvements jointly improve the overall performance of the system and provide users with more accurate and personalized health monitoring and treatment plans.

[0080] 3. The feedback mechanism introduced in the model training process of the transcutaneous electrical stimulation device based on the large language model makes the prediction of electrical stimulation parameters more accurate, and can be intelligently adjusted according to the patient's real-time physiological signals and symptom feedback, thereby improving the personalization and effectiveness of treatment. The real-time information processing and response module realizes continuous monitoring and immediate response to the patient's health status, and can promptly detect and warn of potential health risks. At the same time, it automatically adjusts the electrical stimulation parameters according to the monitoring results to ensure the safety and comfort of treatment. In addition, the collection and use of user feedback provides data support for the online learning and optimization of the model, enabling the device to continuously adapt to changes in patients, provide more accurate and personalized treatment plans, and further improve treatment effects and user satisfaction.

[0081] 4. This transcutaneous electrical stimulation device based on a large language model predicts and adjusts electrical stimulation parameters in real time through a deep learning model, achieving dynamic optimization of personalized treatment plans and improving treatment effects. At the same time, the introduction of parameter verification and adjustment mechanisms ensures that the adjusted parameters are within a safe range and meet medical standards, effectively avoiding risks that may be caused by improper parameters. This design not only improves the accuracy of treatment and patient comfort, but also provides a more reliable technical guarantee for the widespread application of electrical stimulation therapy.

[0082] 5. This transcutaneous electrical stimulation device based on a large language model achieves a rapid and accurate understanding of the user's symptom description through the lightweight STT algorithm and LLM sentiment analysis and keyword extraction, providing rich information support for treatment. Data synchronization and fusion technology ensures the accurate association and comprehensive analysis of multi-source data, improving the accuracy of health status assessment. The introduction of interactive modules enhances user participation, allowing users to understand their health status more intuitively and respond in a timely manner, thereby improving the overall treatment effect and user satisfaction. And by designing a lightweight large language model, only simple language recognition and analysis functions are retained, reducing computing resources.

[0083] 6. The transcutaneous electrical stimulation device based on the large language model is a ring-shaped belt (equipped with stimulation electrodes and a monitoring host, and data can be uploaded by connecting to a mobile phone via Bluetooth or WiFi). When used on the wrist, the measured data can be transmitted to the large language model of the cloud or local app to generate the parameters and time of transcutaneous electrical stimulation treatment. After the user agrees, transcutaneous electrical stimulation therapy will be implemented. If the user purchases ring belts of different lengths at the same time, they can also be worn on other parts of the body (such as the legs and waist) by splicing and extending them to simultaneously stimulate multiple parts, which improves versatility; and the stimulation electrodes of the ring belt will have a round bead design for auxiliary pressure, which can be combined with pressure at the acupoints to achieve better therapeutic effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A schematic diagram of a state 1 of the transcutaneous electrical stimulation device of the present invention;

[0085] Figure 2 It is a schematic diagram of the second state of the transcutaneous electrical stimulation device of the present invention;

[0086] Figure 3 It is a system principle block diagram of the present invention;

[0087] Figure 4 It is a schematic diagram of the process framework of the steps of the present invention.

[0088] In the figure: 1- strap, 2- electrode sheet, 3- control module, 4- snap button, 5- buckle. DETAILED DESCRIPTION

[0089] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0090] The present invention provides a transcutaneous electrical stimulation device based on a large language model, and discloses the following five technical solutions:

[0091] The first implementation method includes:

[0092] The data acquisition and preprocessing module collects human body data through sensors on the transcutaneous electrical stimulation device, and records and recognizes symptom descriptions through voice input, performs denoising, normalization and text cleaning on the collected data, and converts it into vector representation;

[0093] Feature extraction and fusion module, used to extract physiological signal features and symptom text features, and use a hybrid neural network model for feature fusion;

[0094] Model training and prediction module, builds a deep learning model using a fully connected layer to predict electrical stimulation parameters, and uses the extracted features to train the model;

[0095] The real-time information processing and response module continuously collects and preprocesses physiological signals, extracts key physiological parameters from them, and then performs anomaly detection and trend analysis to predict potential health risks. It automatically adjusts electrical stimulation parameters based on the monitoring results and triggers an alarm when serious health abnormalities are detected.

[0096] The present invention integrates real-time data processing and intervention capabilities, can instantly analyze human body data and automatically adjust electrical stimulation parameters, and provide immediate treatment or adjustment solutions for problems such as arrhythmia and chronic pain, greatly improving the timeliness and effectiveness of treatment. Secondly, by integrating physiological signal features and symptom text features, the device can more comprehensively understand the user's health status, thereby providing more accurate treatment plans. In addition, the device also has anomaly detection and trend analysis functions, which can predict potential health risks and trigger alarms in time when serious anomalies are detected, effectively protecting the user's life safety. In summary, the device not only improves the accuracy and timeliness of treatment, but also enhances the user's sense of security, bringing revolutionary progress to the field of medical health.

[0097] The second implementation mode is mainly different from the first implementation mode in that the working contents of the data acquisition and preprocessing module specifically include:

[0098] S1.1 Data input:

[0099] Sensor data collection: Use electrocardiogram (ECG), photoelectric volumetric pulse wave (PPG), skin electrical activity (EDA) and temperature sensor devices to capture the user's physiological signals in real time;

[0100] Symptom voice input: Users input their symptom descriptions, such as pain, discomfort and other subjective feelings, through the interactive module. The large language model recognizes the symptom descriptions and converts them into standard instructions.

[0101] S1.2 Data preprocessing:

[0102] De-noising: Apply low-pass filter or median filter techniques to physiological signals to eliminate background noise;

[0103] Normalization processing: adjust the physiological signals to a uniform scale to ensure the stability of the model input;

[0104] Text cleaning and vectorization: Preprocess the symptom text to remove irrelevant information such as stop words and punctuation marks, and use NLP technology (such as word embedding and BERT model) to convert the symptom text into a vector representation.

[0105] The working contents of the feature extraction and fusion module specifically include:

[0106] S2.1 Feature extraction:

[0107] Physiological signal feature extraction: Extract key features from physiological signals of heart rate and blood pressure. The extracted physiological signal features include:

[0108] Average heart rate: Calculate the average heart rate over a period of time. The formula is: Where n is the number of sampling points;

[0109] Heart rate variability HRV: measures the fluctuation of heart rate over a set period of time, expressed as standard deviation:

[0110] Mean blood pressure: Calculate the mean of systolic and diastolic blood pressure separately:

[0111]

[0112] Symptom text feature extraction: extract sentiment or keyword importance scores related to specific symptoms from text vectors;

[0113] S2.2 Multimodal data fusion: A hybrid neural network model, such as a multi-input neural network, is used to combine the long short-term memory network LSTM to process time series data (such as heart rate, blood pressure) and the Transformer to process text data (such as symptom description). The features of different modalities are fused through the fusion layer to form a unified feature representation. The fusion process is expressed as:

[0114] h=Concat(LSTM(xhr,xbp),Transformer(xsymptoms));

[0115] Among them, Xhr, Xbp, and Xsymptomsx represent heart rate characteristics, blood pressure characteristics, and symptom characteristics, respectively.

[0116] By capturing multi-source physiological signals and symptom voice input in real time, comprehensive monitoring of the user's health status is achieved, and the richness and accuracy of the data are improved. Secondly, the data preprocessing process effectively eliminates noise, ensures the stability of the model input, and provides a high-quality data foundation for subsequent analysis. In terms of feature extraction, not only key physiological parameters are extracted from physiological signals, but also emotional and keyword information is mined from symptom texts, achieving in-depth insights into the user's health status. Finally, multimodal data fusion is performed through a hybrid neural network model, and feature information from different sources is organically integrated to form a unified feature representation, which provides strong support for subsequent model training and prediction. These improvements jointly improve the overall performance of the system and provide users with more accurate and personalized health monitoring and treatment plans.

[0117] The third implementation mode is mainly different from the second implementation mode in that the working contents of the model training and prediction module specifically include:

[0118] S3.1 Model design: A deep learning model is constructed, and a fully connected layer Dense Layer is used to predict the electrical stimulation parameters. The ReLU activation function is used to ensure the non-negativity of the output. The model structure is expressed as follows:

[0119] Output = ReLU(Dense(fused features));

[0120] Among them, the Dense layer represents the fully connected layer, where the weights and biases are optimized by the back-propagation algorithm during the training process;

[0121] S3.2 Loss function selection: The mean square error (MSE) is used as the loss function to measure the difference between the predicted electrical stimulation parameters and the true values. The MSE formula is:

[0122]

[0123] Among them, m is the number of samples, y i is the actual electrical stimulation parameter, y^ i are the parameters predicted by the model;

[0124] S3.3 Model training: The model is trained using the extracted features, with the goal of predicting the optimal electrical stimulation parameters (intensity, frequency, duration).

[0125] A feedback mechanism is introduced during model training to achieve intelligent adjustment, including:

[0126] S3.3.1 Initial prediction: Use the trained model to extract and fuse features of new physiological signals and symptom texts, and then predict the initial electrical stimulation parameters;

[0127] S3.3.2 Effect evaluation: Apply the predicted electrical stimulation parameters to the patient and collect physiological signals and symptom feedback after treatment;

[0128] S3.3.3 Parameter adjustment: Based on the treatment effect (such as the degree of symptom relief, improvement of physiological indicators, etc.), the model is fine-tuned using reinforcement learning or Bayesian optimization methods to more accurately predict future electrical stimulation parameters;

[0129] S3.3.4 Iterative training: Use the adjusted parameters and treatment effects as new training data to retrain the model to improve its generalization ability and prediction accuracy.

[0130] The work content of the real-time information processing and response module specifically includes:

[0131] S4.1 Real-time data collection and processing:

[0132] Real-time acquisition and processing: continuously collect physiological signals, perform filtering and denoising, and improve data quality;

[0133] Real-time feature extraction: extract key physiological parameters from real-time data;

[0134] S4.2 Real-time health monitoring and response:

[0135] Anomaly detection and trend analysis: Use anomaly detection algorithms to identify abnormal patterns in data and perform trend analysis to predict potential health risks;

[0136] Electrical stimulation parameter adjustment: Automatically adjust electrical stimulation parameters according to monitoring results;

[0137] Alerts and notifications: When serious health abnormalities are detected, alarms are triggered and users and healthcare providers are notified;

[0138] S4.3 User feedback collection: After the electrical stimulation is implemented, user feedback on the treatment effect is collected;

[0139] S4.4 Model optimization: Collect user feedback on electrical stimulation therapy (including comfort, treatment effect, etc.); use this feedback to perform online learning or incremental learning of the model, and continuously optimize the prediction and adjustment rules of electrical stimulation parameters.

[0140] The feedback mechanism introduced during the model training process makes the prediction of electrical stimulation parameters more accurate, and can be intelligently adjusted according to the patient's real-time physiological signals and symptom feedback, improving the personalization and effectiveness of treatment. The real-time information processing and response module realizes continuous monitoring and immediate response to the patient's health status, and can promptly detect and warn of potential health risks. At the same time, it automatically adjusts the electrical stimulation parameters according to the monitoring results to ensure the safety and comfort of treatment. In addition, the collection and use of user feedback provides data support for the online learning and optimization of the model, allowing the equipment to continuously adapt to changes in patients, provide more accurate and personalized treatment plans, and further improve treatment effects and user satisfaction.

[0141] The fourth embodiment is mainly different from the third embodiment in that: the model output of the electrical stimulation parameter adjustment includes the intensity I, frequency F and duration T of the electrical stimulation;

[0142] Assume that the output of the deep learning model is the predicted electrical stimulation parameter vector P = [I, F, T];

[0143] According to the real-time monitoring data D and feature extraction results F, the model is expressed as:

[0144] mathbfP=f(F;θ);

[0145] Where f is the deep learning model and θ is the model parameter;

[0146] Then, the parameter adjustment rule is formalized as:

[0147] I new =g I (I, D);

[0148] F new=g F (F, D);

[0149] T new =g T (T, D);

[0150] Among them, g I , g F and g T They are functions for adjusting electrical stimulation parameters I, F, and T according to real-time monitoring data D;

[0151] Parameter verification and adjustment:

[0152] After adjusting the parameters, perform parameter verification to ensure that the new parameter values ​​are within the safe range and meet medical standards;

[0153] If the new parameters do not meet the requirements, they are fine-tuned according to the preset rules until the conditions are met.

[0154] By using a deep learning model to predict and adjust electrical stimulation parameters in real time, dynamic optimization of personalized treatment plans is achieved, improving treatment effects. At the same time, the introduction of parameter verification and adjustment mechanisms ensures that the adjusted parameters are within a safe range and meet medical standards, effectively avoiding risks that may be caused by improper parameters. This design not only improves the accuracy of treatment and patient comfort, but also provides a more reliable technical guarantee for the widespread application of electrical stimulation therapy.

[0155] The fifth implementation mode is mainly different from the first implementation mode in that the symptom voice input includes:

[0156] Voice input and preprocessing: Users input symptom descriptions through the microphone of the wearable device, and the lightweight STT algorithm is used to convert the voice into text, and then the text is preprocessed, such as removing noise and segmenting words;

[0157] Sentiment analysis and keyword extraction: Use the optimized lightweight LLM to perform sentiment analysis on the preprocessed text, extract sentiment tendencies, and use keyword extraction algorithms (such as TF-IDF, TextRank, etc.) to identify key information in the text;

[0158] Data synchronization and fusion: synchronize the timestamps of data from different sensors (such as heart rate, blood pressure, etc.); use fusion algorithms to integrate sensor data to form a comprehensive health assessment;

[0159] During the electrical stimulation treatment, interaction and feedback are carried out through the interactive module, including: based on the analysis results, showing the user possible disease suggestions or health suggestions through a simplified interface or voice feedback; further interactive options can also be provided, such as asking the user whether he is willing to undergo a more in-depth examination or consult a doctor.

[0160] Through the lightweight STT algorithm and LLM sentiment analysis and keyword extraction, a quick and accurate understanding of the user's symptom description is achieved, providing rich information support for treatment. Data synchronization and fusion technology ensures the accurate association and comprehensive analysis of multi-source data, improving the accuracy of health status assessment. The introduction of interactive modules enhances user participation, allowing users to understand their health status more intuitively and respond in a timely manner, thereby improving the overall treatment effect and user satisfaction. And by designing a lightweight large language model, only simple language recognition and analysis functions are retained, reducing computing resources.

[0161] The present invention also discloses a transcutaneous electrical stimulation device, comprising:

[0162] The strap 1 is used to be tied to the patient's limbs, and the strap 1 is made of elastic material. The elastic design can better fit the skin and maintain stable data collection. Various sensors are also integrated inside the strap 1;

[0163] The electrode sheet 2 is evenly arranged on the inner side of the bandage 1, and the surface of the electrode sheet 2 is provided with a plurality of convex points; the convex points are hemispherical, which are used to achieve auxiliary pressure on the meridian acupoints and ensure good conductivity and appropriate hardness. The following material options can be considered:

[0164] 1. Stainless steel

[0165] Electrical conductivity: Stainless steel has good electrical conductivity and can effectively transmit electrical stimulation.

[0166] Hardness: Stainless steel has high hardness, is wear-resistant, and can withstand a certain amount of mechanical pressure, making it suitable for pressurized stimulation of acupoints.

[0167] Biocompatibility: High-quality medical stainless steel generally has good biocompatibility and is suitable for prolonged contact with the skin.

[0168] Considerations: Make sure the type of stainless steel you choose (such as 304 or 316) does not cause allergic reactions to the skin.

[0169] 2. Silver

[0170] Conductive properties: Silver has the highest electrical conductivity of all metals and is suitable for applications requiring efficient electrical signal transmission.

[0171] Hardness: Pure silver is soft, but its hardness and durability can be increased by alloying it with other metals, such as copper.

[0172] Biocompatibility: Silver has certain antibacterial properties, which helps reduce the risk of infection at skin contact sites.

[0173] Considerations: Silver is expensive and the silver material needs to be prevented from oxidizing and turning black.

[0174] 3. Gold-plated stainless steel

[0175] Conductive Properties: The gold plating provides excellent conductivity while maintaining the mechanical properties of stainless steel.

[0176] Hardness: Stainless steel provides the required hardness and durability.

[0177] Biocompatibility: Gold is a very good biocompatible material and rarely causes skin allergic reactions.

[0178] Considerations: Metal plating can wear away, and the thickness and durability of the plating need to be considered.

[0179] Comprehensive advice: When selecting materials, not only conductivity and hardness should be considered, but also cost, processing technology, biocompatibility and environmental adaptability. Stainless steel or gold-plated stainless steel is a more economical and practical choice, providing good mechanical properties and conductivity, while silver and conductive silicone can be used as options for specific needs. When designing, the size, shape and fixing method of the beads also need to be considered to ensure comfort and functionality during use.

[0180] The connection structure is provided at both ends of the strap 1 and is used to connect the two ends of the strap 1 or to connect with the dial of the smart watch;

[0181] The control module 3 is used to connect and control a plurality of electrode sheets 2. The control module 3 can be connected to a mobile phone APP and synchronized with the smart phone APP to provide detailed health reports and setting options; or synchronize data to the cloud via Bluetooth or Wi-Fi to support remote medical services; it also has a built-in high-efficiency rechargeable battery to support long-term operation and wireless charging for user convenience; it has a built-in high-performance microprocessor to support high-speed data processing and machine learning operations; and it has built-in storage for storing historical data and user configurations.

[0182] The connection structure is a snap button 4 or a plug button 5;

[0183] When snap buttons 4 are installed at both ends of the strap 1, the two ends can be connected to each other, or multiple straps 1 can be connected in series;

[0184] When buckles 5 are installed at both ends of the strap 1, they are used to be inserted into the two sides of the smart watch dial.

[0185] The product form of this device is a ring belt equipped with stimulation electrodes and a monitoring host. The data can be uploaded by connecting to a mobile phone via Bluetooth or WiFi. When used on the wrist, the measured data can be transmitted to the cloud or the large language model of the local app to generate the parameters and time of transcutaneous electrical stimulation treatment. After the user agrees, transcutaneous electrical stimulation therapy will be implemented. If the user purchases ring belts of different lengths at the same time, they can also be worn on other parts of the body such as the legs and waist by splicing and extending them, which improves versatility; and the stimulation electrodes of the ring belt will have a round bead design for auxiliary pressure, which can be combined with pressure at the acupoints to achieve better therapeutic effects.

[0186] Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0187] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0188] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A transcutaneous electrical stimulation device based on a large language model, characterized in that: include: The data acquisition and preprocessing module collects human body data through sensors on the transcutaneous electrical stimulation device, and records and recognizes symptom descriptions through voice input, performs denoising, normalization and text cleaning on the collected data, and converts it into vector representation; Feature extraction and fusion module, used to extract physiological signal features and symptom text features, and use a hybrid neural network model for feature fusion; Model training and prediction module, which builds a deep learning model using a fully connected layer to predict electrical stimulation parameters and uses the extracted features to train the model; The real-time information processing and response module continuously collects and preprocesses physiological signals, extracts key physiological parameters from them, and then performs anomaly detection and trend analysis to predict potential health risks. It automatically adjusts electrical stimulation parameters based on the monitoring results and triggers an alarm when serious health abnormalities are detected.

2. A transcutaneous electrical stimulation device based on a large language model according to claim 1, characterized in that: The work content of the data acquisition and preprocessing module specifically includes: S1.1 Data input: Sensor data collection: Use electrocardiogram (ECG), photoelectric volumetric pulse wave (PPG), skin electrical activity (EDA) and temperature sensor devices to capture the user's physiological signals in real time; Symptom voice input: Users input their symptom descriptions through the interactive module, and the large language model recognizes the symptom descriptions and converts them into standard instructions; S1.2 Data preprocessing: De-noising: Apply low-pass filter or median filter techniques to physiological signals to eliminate background noise; Normalization processing: adjust the physiological signals to a uniform scale; Text cleaning and vectorization: Preprocess the symptom text to remove irrelevant information such as stop words and punctuation marks, and use NLP technology to convert the symptom text into a vector representation.

3. A transcutaneous electrical stimulation device based on a large language model according to claim 1, characterized in that: The working contents of the feature extraction and fusion module specifically include: S2.1 Feature extraction: Physiological signal feature extraction: Extract key features from physiological signals of heart rate and blood pressure. The extracted physiological signal features include: Average heart rate: Calculate the average heart rate over a period of time. The formula is: Where n is the number of sampling points; Heart rate variability HRV: measures the fluctuation of heart rate over a set period of time, expressed as standard deviation: Mean blood pressure: Calculate the mean of systolic and diastolic blood pressure separately: Symptom text feature extraction: extract sentiment or keyword importance scores related to specific symptoms from text vectors; S2.2 Multimodal data fusion: A hybrid neural network model is used to combine the long short-term memory network LSTM to process time series data and the Transformer to process text data. The features of different modalities are fused through the fusion layer to form a unified feature representation. The fusion process is expressed as: h=Concat(LSTM(xhr,xbp),Transformer(xsymptoms)); Among them, Xhr, Xbp, and Xsymptomsx represent heart rate characteristics, blood pressure characteristics, and symptom characteristics, respectively.

4. A transcutaneous electrical stimulation device based on a large language model according to claim 1, characterized in that: The working contents of the model training and prediction module specifically include: S3.1 Model design: A deep learning model is constructed, and a fully connected layer Dense Layer is used to predict the electrical stimulation parameters. The ReLU activation function is used to ensure the non-negativity of the output. The model structure is expressed as follows: Output = ReLU(Dense(fused features)); Among them, the Dense layer represents the fully connected layer, where the weights and biases are optimized by the back-propagation algorithm during the training process; S3.2 Loss function selection: The mean square error (MSE) is used as the loss function to measure the difference between the predicted electrical stimulation parameters and the true values. The MSE formula is: Among them, m is the number of samples, y i is the actual electrical stimulation parameter, y^ i are the parameters predicted by the model; S3.3 Model training: The model is trained using the extracted features, with the goal of predicting the optimal electrical stimulation parameters.

5. A transcutaneous electrical stimulation device based on a large language model according to claim 4, characterized in that: The feedback mechanism is introduced into the model training process to realize intelligent adjustment, including: S3.3.1 Initial prediction: Use the trained model to extract and fuse features of new physiological signals and symptom texts, and then predict the initial electrical stimulation parameters; S3.3.2 Effect evaluation: Apply the predicted electrical stimulation parameters to the patient and collect physiological signals and symptom feedback after treatment; S3.3.3 Parameter adjustment: Based on the treatment effect, the model can be fine-tuned using reinforcement learning or Bayesian optimization methods to more accurately predict future electrical stimulation parameters; S3.3.4 Iterative training: Use the adjusted parameters and treatment effects as new training data to retrain the model.

6. A transcutaneous electrical stimulation device based on a large language model according to claim 1, characterized in that: The work content of the real-time information processing and response module specifically includes: S4.1 Real-time data collection and processing: Real-time acquisition and processing: continuously collect physiological signals, perform filtering and denoising; Real-time feature extraction: extract key physiological parameters from real-time data; S4.2 Real-time health monitoring and response: Anomaly detection and trend analysis: Use anomaly detection algorithms to identify abnormal patterns in data and perform trend analysis to predict potential health risks; Electrical stimulation parameter adjustment: Automatically adjust electrical stimulation parameters according to monitoring results; Alerts and notifications: When serious health abnormalities are detected, alarms are triggered and users and healthcare providers are notified; S4.3 User feedback collection: After the electrical stimulation is implemented, user feedback on the treatment effect is collected; S4.4 Model optimization: Collect user feedback on electrical stimulation therapy; use the feedback to perform online learning or incremental learning of the model to continuously optimize the prediction and adjustment rules of electrical stimulation parameters.

7. A transcutaneous electrical stimulation device based on a large language model according to claim 6, characterized in that: The model output of the electrical stimulation parameter adjustment includes the intensity I, frequency F and duration T of the electrical stimulation; Assume that the output of the deep learning model is the predicted electrical stimulation parameter vector P = [I, F, T]; According to the real-time monitoring data D and feature extraction results F, the model is expressed as: mathbfP=f(F;θ); Where f is the deep learning model and θ is the model parameter; Then, the parameter adjustment rule is formalized as: I new =g I (I,D); F new =g F (F,D); T new =g T (T,D); Among them, g I , g F and g T They are functions for adjusting electrical stimulation parameters I, F, and T according to real-time monitoring data D; Parameter verification and adjustment: After adjusting the parameters, perform parameter verification to ensure that the new parameter values ​​are within the safe range and meet medical standards; If the new parameters do not meet the requirements, they are fine-tuned according to the preset rules until the conditions are met.

8. A transcutaneous electrical stimulation device based on a large language model according to claim 1, characterized in that: The symptom voice input includes: Voice input and preprocessing: Users input symptom descriptions through the microphone of the wearable device, and the lightweight STT algorithm is used to convert the voice into text, and then the text is preprocessed; Sentiment analysis and keyword extraction: Use the optimized lightweight LLM to perform sentiment analysis on the preprocessed text, extract sentiment tendencies, and use the keyword extraction algorithm to identify key information in the text; Data synchronization and fusion: synchronize the timestamps of data from different sensors; use fusion algorithms to integrate sensor data to form a comprehensive health assessment; During the electrical stimulation treatment, interaction and feedback are carried out through the interactive module, including: based on the analysis results, showing possible disease suggestions or health suggestions to the user through a simplified interface or voice feedback.

9. A transcutaneous electrical stimulation device based on a large language model according to claim 1, characterized in that: The transcutaneous electrical stimulation device comprises: A strap is used to be tied to the patient's limbs, and the strap is made of elastic material; Electrode sheets, multiple of which are evenly arranged on the inner side of the bandage, and the surface of the electrode sheets is provided with a plurality of convex points; A connecting structure is provided at both ends of the strap, and is used to connect the two ends of the strap or to connect with the dial of the smart watch; The control module is used to connect and control a plurality of electrode sheets.

10. A transcutaneous electrical stimulation device based on a large language model according to claim 9, characterized in that: The connecting structure is a snap button or a plug button; When snaps are installed at both ends of the strap, the two ends can be connected to each other, or multiple straps can be connected in series; When buckles are installed at both ends of the strap, they are used to be inserted into the two sides of the smart watch dial.

Images