Control method and system for self-adaptive electric pulse generating device

Abstract

The embodiment of the invention aims to provide a control method and system for a self-adaptive electric pulse generating device. The method comprises the following steps: acquiring inflammation-related data of a user; according to the inflammation related data, extracting a biological feature vector of the user through a feature extraction module, and according to the biological feature vector, determining a disease course classification of the user through a disease course classification module, according to the biological feature vectors and the disease course classification, corresponding electric pulse parameters are determined through a parameter generation module; and according to the electric pulse parameters, applying a corresponding wide-frequency-domain electric pulse stimulation scheme to the electric stimulation applying part of the user through the self-adaptive electric pulse generating device. Through data fusion and closed-loop optimization, precise personalization and dynamic adaptability of parameter generation are realized, and the electrical pulse parameter configuration scheme is promoted to span from a static template to an intelligent dynamic mode.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the fields of electrical pulse devices and artificial intelligence, and particularly to a control technology for an adaptive electrical pulse generating device. Background Art

[0002] In the field of bioelectrical stimulation, existing devices generally adopt a preset parameter mode to achieve the basic electrical pulse output function. For example, the basic electrical stimulation function is achieved by presetting a frequency range and fixed waveform parameters, and its parameter settings are mainly based on general electrical pulse parameters formed by clinical experience. Such devices usually provide a limited adjustable parameter range, and their frequency output ranges mostly concentrate on the low-frequency band, and lack the multi-dimensional perception ability of the deep tissue state, resulting in a significant deviation between parameter adjustment and the real physiological state.

[0003] Moreover, existing electrical stimulation devices generally face the technical limitation of insufficient parameter adaptability in clinical applications. Since core parameters such as current intensity and frequency mostly adopt a static preset mode, it is difficult to dynamically adjust according to the physiological feedback obtained in real time during the physiotherapy process, resulting in a significant deviation between the actual output parameters and the patient's immediate state.

[0004] These technical defects together lead to the difficulty of existing devices in achieving precise and personalized bioelectrical intervention effects when dealing with related complex pathological states such as aseptic inflammation of the musculoskeletal system. Summary of the Invention

[0005] The purpose of embodiments of the present disclosure is to provide a control method and system for an adaptive electrical pulse generating device.

[0006] According to one aspect of the present disclosure, a control method for an adaptive electrical pulse generating device is provided, wherein the method includes the following steps:

[0007] Collect inflammation-related data of the user, including the electrical stimulation application site, user basic data, and user physiological data;

[0008] According to the inflammation-related data, extract the user's biometric vector through a feature extraction module, determine the user's disease course classification through a disease course classification module according to the biometric vector, and then determine corresponding electrical pulse parameters through a parameter generation module according to the biometric vector and the disease course classification;

[0009] Among them, the feature extraction module, the disease course classification module, and the parameter generation module are trained through a case sample set:

[0010] The case sample set includes labeled source domain samples and unlabeled target domain samples. The source domain samples include inflammation-related data of patients, corresponding disease course labels, and electrical pulse parameters;

[0011] Use the source domain samples to initially train the feature extraction module and the disease course classification module;

[0012] Input the inflammation-related data of the source domain samples and the target domain samples into the feature extraction module, and adjust the feature extraction module by minimizing the distribution difference of the output biometric vectors of the two domains;

[0013] Use the inflammation-related data of the source domain samples and the target domain samples with domain labels to perform adversarial training on the feature extraction module and the domain classification module, and at the same time use the source domain samples to optimize the feature extraction module and the disease course classification module;

[0014] Fix the parameters of the feature extraction module, and jointly train the disease course classification module and the parameter generation module according to the biometric vectors of the source domain samples output by the feature extraction module;

[0015] According to the electrical pulse parameters, apply a corresponding broadband electrical pulse stimulation scheme to the user at the electrical stimulation application site through the adaptive electrical pulse generating device.

[0016] According to one aspect of the present disclosure, there is also provided a control system for an adaptive electrical pulse generating device, including an adaptive electrical pulse generating device, a user device, and a network device; wherein, the adaptive electrical pulse generating device is connected with electrode patches, and a plurality of sensors are arranged on the electrode patches; wherein, the adaptive electrical pulse generating device is configured to perform the following operations:

[0017] Collect the inflammation-related data of the user through the sensors, including the user's physiological data;

[0018] According to the electrical pulse parameters, apply a corresponding broadband electrical pulse stimulation scheme to the user at the electrical stimulation application site through the electrode patches;

[0019] The user device is configured to perform the following operations:

[0020] Collect the inflammation-related data of the user, including at least the electrical stimulation application site and the user's basic data;

[0021] The network device is configured to perform the following operations:

[0022] According to the inflammation-related data, a biometric vector of the user is extracted by a feature extraction module, a disease course classification of the user is determined by a disease course classification module according to the biometric vector, and then corresponding electrical pulse parameters are determined by a parameter generation module according to the biometric vector and the disease course classification;

[0023] Among them, the feature extraction module, the disease course classification module, and the parameter generation module are trained by a case sample set:

[0024] The case sample set includes labeled source domain samples and unlabeled target domain samples. The source domain samples include inflammation-related data of patients, corresponding disease course labels, and electrical pulse parameters;

[0025] The source domain samples are used to initially train the feature extraction module and the disease course classification module;

[0026] The inflammation-related data of the source domain samples and the target domain samples are input into the feature extraction module, and the feature extraction module is adjusted by minimizing the distribution difference of the output biometric vectors of the two domains;

[0027] The inflammation-related data of the source domain samples and the target domain samples with domain labels are used to perform adversarial training on the feature extraction module and the domain classification module, and at the same time, the source domain samples are used to optimize the feature extraction module and the disease course classification module;

[0028] The parameters of the feature extraction module are fixed, and the disease course classification module and the parameter generation module are jointly trained according to the biometric vectors of the source domain samples output by the feature extraction module.

[0029] The solution of the present disclosure starts from the comprehensive collection of multimodal data. The bioelectrical impedance, surface temperature, and electromyogram signals of the physiotherapy part are obtained through the micro sensors integrated in the electrode patches, and the user's basic information and pain feedback are recorded at the same time. For example, during knee joint physiotherapy, the sensor continuously captures the local impedance change curve, temperature gradient, and muscle activity spectrum, and constructs a multi-dimensional data input in combination with the user's age and body mass index.

[0030] The feature extraction module converts the original signal into a unified biometric vector and captures the cross-modal association law. For example, the periodic fluctuation of impedance, the temperature rise slope, and the energy distribution of the electromyogram signal are encoded into a 128-dimensional vector, which characterizes the inflammation activity degree and tissue state. This module eliminates the data distribution deviation between different user groups through cross-domain feature alignment. For example, the metabolic rate difference of elderly patients is normalized to a standard inflammation index to ensure the compatibility of new user data with the historical case database.

[0031] The disease course classification module determines the current disease course stage based on the feature vector and removes interference factors such as device acquisition preference through adversarial training. When a patient's characteristics match both the acute and chronic phase patterns, the module outputs a probability distribution and selects the dominant category. For example, when the probability of the acute phase is 65%, the medium-frequency analgesia scheme is locked, rather than the compromise parameters. The classification results and the feature vector are jointly input into the parameter generation module to drive dynamic decision-making.

[0032] The parameter generation module adopts a multi-objective optimization strategy, which not only refers to the biological feature vector, but also combines the confidence of the disease course classification for multi-objective optimization. When there is uncertainty about the classification result, the conservative parameter combination with the lowest risk is automatically selected. This mechanism enables the system to maintain treatment safety in the face of complex cases, such as generating a gradual parameter scheme for patients with both acute and chronic characteristics, rather than mechanically applying a fixed template.

[0033] The adaptive electric pulse generator executes the generated parameters, and its wide frequency domain coverage capability supports flexible switching from low-frequency neuromodulation to deep tissue intervention. Furthermore, during the physical therapy process, real-time physiological data flows back to the system, forming a "collection-decision-execution-optimization" closed loop. For example, when the impedance of a user tends to be stable in the later stage of physical therapy, the device automatically reduces the frequency and extends the duty cycle, improving tolerance while maintaining the therapeutic effect.

[0034] The disclosed solution achieves accurate personalization and dynamic adaptability of parameter generation through data fusion and closed-loop optimization. Cross-domain feature alignment ensures the safe transfer of historical physical therapy experience to new users, the adversarial training mechanism eliminates non-pathological interference, and the joint optimization strategy opens up the complete link from data to decision-making. The final output electric pulse parameters are consistent with clinical pathological logic and can evolve in real time according to individual responses, promoting the configuration scheme of electric pulse parameters from static templates to intelligent dynamic modes.

[0035] Furthermore, the present disclosure adopts a data-driven feedback optimization mechanism to achieve continuous learning and optimization of personalized electrical pulse parameters. For example, for patients in the chronic stage, the parameter generation module gives priority to medium-frequency continuous waves and introduces periodic intensity modulation; if a sudden drop in impedance is detected during physical therapy, indicating an acute attack, the reinforcement learning model seamlessly switches to the medium-frequency pulse mode. The reinforcement learning mechanism allows autonomous exploration within a safe threshold. For example, after discovering that a user responds best to a specific duty cycle square wave, it is included in the personalized strategy library. The data feedback after each physical therapy will be used to improve the accuracy of the relevant model, making the control system of the present disclosure not only a static parameter configuration tool, but also a continuously optimized intelligent system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Other features, objects and advantages of the present disclosure will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0037] Figure 1 The flowchart shows a control method for an adaptive electrical pulse generating device according to an embodiment of the present disclosure.

[0038] Identical or similar reference numerals in the drawings represent identical or similar components. Detailed implementation manners

[0039] The specific embodiments of the present disclosure will be further described below with reference to the drawings.

[0040] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments of the present disclosure are described as devices represented by block diagrams and processes or methods represented by flowcharts. Although the flowcharts describe the operation processes of the embodiments of the present disclosure as sequential processes, many of the operations can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the operations can be rearranged. The processes of the embodiments of the present disclosure can be terminated when their operations are completed, but can also include additional steps not shown in the flowcharts. The processes of the embodiments of the present disclosure can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0041] The methods shown by the flowcharts and the devices shown by the block diagrams discussed below can be implemented by hardware, software, firmware, middleware, microcode, hardware description language, or any combination thereof. When implemented by software, firmware, middleware, or microcode, the program code or code segments for performing the necessary tasks can be stored in a machine or a computer-readable medium such as a storage medium. One or more processors can execute the necessary tasks.

[0042] Similarly, it will also be understood that any flowchart, flow diagram, state transition diagram, and the like, representing various processes, can be fully described as program code stored in a computer-readable medium and thus executed by a computer device or a processor, whether or not these computer devices or processors are explicitly shown.

[0043] As used herein, the term "storage medium" can refer to one or more devices for storing data, including read-only memory (ROM), random access memory (RAM), magnetic RAM, core memory, disk storage media, optical storage media, flash devices, and / or other machine-readable media for storing information. The term "computer-readable medium" can include, but is not limited to, portable or fixed storage devices, optical storage devices, and various other media capable of storing and / or containing instructions and / or data.

[0044] A code segment may represent a procedure, function, subroutine, program, routine, subroutine, module, software package, class, or any combination of instructions, data structures, or program descriptions. A code segment can be coupled to another code segment or a hardware circuit by passing and / or receiving information, data, arguments, parameters, or stored contents. Information, arguments, parameters, data, etc. can be passed, forwarded, or transmitted via any suitable means including storage sharing, information passing, token passing, network transmission, etc.

[0045] As used herein, the term "computer device" refers to an electronic device that can perform a predetermined processing process such as numerical calculation and / or logical calculation by running a predetermined program or instruction. It can at least include a processor and a memory, where the processor executes the program instructions pre-stored in the memory to perform the predetermined processing process, or the predetermined processing process is executed by hardware such as an ASIC, FPGA, DSP, etc., or is implemented by a combination of the above two.

[0046] The above-mentioned "computer device" usually appears in the form of a general-purpose computer device, and its components may include but are not limited to: one or more processors or processing units, and a system memory. The system memory may include a computer-readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The "computer device" may further include other removable / non-removable, volatile / non-volatile computer-readable storage media. The memory may include at least one computer program product having a set (e.g., at least one) of program modules configured to perform the functions and / or methods of the various embodiments of the present disclosure. The processor executes various functional applications and data processing by running the programs stored in the memory.

[0047] For example, computer programs for performing the functions and processes of the various embodiments of the present disclosure are stored in the memory, and when the processor executes the corresponding computer programs, the various embodiments of the present disclosure are implemented.

[0048] Typically, a computer device can be, for example, a user device or a network device, or even a combination of both. Among them, the user device includes, but is not limited to, a personal computer (PC), a laptop computer, a mobile terminal, etc., and the mobile terminal includes, but is not limited to, a smart phone, a tablet computer, etc.; the network device includes, but is not limited to, a single network server, a server group composed of multiple network servers, or a cloud composed of a large number of computers or network servers based on cloud computing (Cloud Computing). Among them, cloud computing is a type of distributed computing, which is a super virtual computer composed of a group of loosely coupled computer sets. Among them, the computer device can run alone to implement the embodiments of the present disclosure, or can be connected to a network and implement the embodiments of the present disclosure through interaction with other computer devices in the network. Among them, the network where the computer device is located includes, but is not limited to, the Internet, a wide area network, a metropolitan area network, a local area network, a VPN network, etc.

[0049] It should be noted that the user device, network device, network, etc. are only examples, and other existing or future possible computing devices or networks that can be applied to the embodiments of the present disclosure should also be included within the protection scope of the present disclosure and are hereby incorporated by reference.

[0050] The specific structures and functional details disclosed herein are merely representative and are for the purpose of describing the exemplary embodiments of the present disclosure. However, the embodiments of the present disclosure can be specifically implemented in many alternative forms and should not be construed as being limited only to the embodiments set forth herein.

[0051] It should be understood that although terms such as "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly, the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0052] The terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments. Unless the context clearly indicates otherwise, the singular forms "a", "an" used herein are also intended to include the plural. It should also be understood that the terms "include" and / or "comprise" used herein specify the presence of the stated features, integers, steps, operations, units, and / or components, and do not preclude the presence or addition of one or more other features, integers, steps, operations, units, components, and / or combinations thereof.

[0053] It should also be noted that in some alternative implementations, the functions / actions mentioned may occur in an order different from that indicated in the figures. For example, depending on the functions / actions involved, two consecutively shown figures may actually be executed substantially simultaneously or sometimes in the reverse order.

[0054] In the traditional physical therapy programs for aseptic inflammation, electrical stimulation therapy is one of the main treatment methods. Electrical stimulation therapy is achieved by relying on specialized electrical stimulation equipment and professional personnel to debug the electrical pulse parameters. The electrical pulse parameters include: electrical pulse type (single-phase pulse, biphasic pulse, modulated pulse, etc.), electrical pulse frequency (1 - 9000 Hz), electrical pulse width (45 - 200 μs), duty cycle (1:2, 1:3, 1:5, etc.), temperature (0 - 42 °C), intensity (low, medium, high, etc.), duration (10 - 20 min), and number of implementations (1 - 2 times per day).

[0055] Taking the electrical pulse stimulation programs for different disease courses of "knee osteoarthritis" as an example, the allocation of electrical pulse parameters will be introduced.

[0056]

[0057] Current modulation description:

[0058] Taking the subacute postoperative period (2 - 6 weeks after knee surgery) as an example,

[0059] Pulse type: Biphasic pulse or modulated pulse (gradually strengthening electrical stimulation to promote the functional recovery of muscles and joints)

[0060] Frequency: 20 - 50 Hz (low frequency, which helps to promote local blood circulation and enhance muscle function)

[0061] Pulse width: 300 μs (longer pulse width stimulation, suitable for promoting blood circulation in deep tissues and soft tissue repair)

[0062] Duty cycle: 50% - 70% (higher duty cycle enhances the physical therapy effect and promotes muscle function recovery)

[0063] Duration: 15 - 20 minutes (to accelerate recovery, moderately extend the physical therapy time)

[0064] Number of implementations: 1 time per day, gradually increasing to 2 times per day (gradually increase the physical therapy frequency according to the recovery progress)

[0065] Electrode patch heating: Heating (38 - 42 °C). Thermotherapy can further enhance blood circulation and promote the recovery of soft tissues and joints

[0066] Intensity: Medium to high (gradually increase the current intensity according to the user's comfort and tolerance to activate muscles and promote recovery)

[0067] As can be seen from the above, the physical therapy program for traditional aseptic inflammation statically classifies the disease course and sets the disease course according to a fixed time period, which cannot truly match the actual disease course progress of each patient. Therefore, it is also very difficult for the electrical pulse stimulation program set according to the fixed disease course to be personalized to meet the real-time physiological state requirements of the patient.

[0068] Multiple embodiments of the present disclosure provide a control system for an adaptive electrical pulse generating device, which can dynamically determine the disease course that the user is currently in and determine a personalized electrical stimulation program for the user. The control system includes an adaptive electrical pulse generating device, an electrical pulse control APP arranged on the user device, and multiple artificial intelligence models arranged on the network device (cloud).

[0069] In some embodiments, the adaptive electrical pulse generating device and the mobile APP can collect inflammation-related data of the user. Among them, the adaptive electrical pulse generating device can collect the user's physiological data through sensors (such as the surface temperature of the human body, the degree of edema, the inflammatory activity state, muscle fatigue and nerve control level, skin resistance, etc.). At the same time, the mobile APP can collect the electrical stimulation application site of the user (such as the inflammation occurrence site), the user's basic data and the user's physiological data (such as the pain level VAS) by receiving user input.

[0070] The artificial intelligence model in the cloud, for example, determines the disease course that the user is currently in and the corresponding electrical pulse parameters (such as pulse type, pulse frequency, temperature, physical therapy intensity, physical therapy duration, number of physical therapy sessions, waveform, duty cycle, etc.) according to the inflammation-related data of the user through a trained model.

[0071] The adaptive electrical pulse generating device applies a corresponding broadband electrical stimulation (1 - 9000 Hz) program to the inflammation occurrence site of the user according to these electrical pulse parameters, and continuously monitors the real-time feedback data of the user during the electrical pulse stimulation process.

[0072] The mobile APP adjusts the electrical pulse parameters for the current user in real time through a reinforcement learning algorithm according to the real-time feedback data of the adaptive electrical pulse generating device.

[0073] When the use of the adaptive electrical pulse generating device ends this time, the artificial intelligence model in the cloud, such as the prediction model, updates the prediction of the user's individual physiological state evolution according to the user's historical physiological state data, electrical pulse parameters, etc., and generates a personalized suggestion for the user.

[0074] Here, the adaptive electric pulse generator is a portable pulse generator that can implement precise physical electrical stimulation to specific inflammation sites (such as knee joints, shoulders, waist, etc.). The portable device provides an intelligent, convenient, and personalized electric pulse stimulation solution, which is suitable for home use and has good adaptability, safety, and physical therapy effects, and can meet the treatment needs of different users.

[0075] The portable device is a small, lightweight device that usually looks like a portable power bank, making it easy to carry and use. The device is designed to be comfortable to hold and easy to operate, with a moderate size and equipped with Type-C and USB charging ports to support long-term convenient use. The shell material of the portable device is made of lightweight, durable, impact-resistant plastic or alloy to ensure its durability and protection in different usage environments.

[0076] The portable device is connected to an electrode sheet, which is made of self-adhesive material and can be repeatedly pasted and remain stable without falling off. When using it, the user only needs to stick the electrode sheet to the part where electrical stimulation is required and start the electrical pulse stimulation program through the portable device.

[0077] The electrode pads are made of flexible materials and are designed to fit various joints and muscle parts of the human body. The electrode pads can be replaced as needed and are stably attached to the site of electrical stimulation application through adhesive or adjustable fixtures. The shape and size of the electrode pads can be designed according to different parts, such as designing electrode pads of different sizes for shoulders, knees, waist, etc., to ensure coverage of the target area and provide the best electrical stimulation effect. The surface of the electrode pads is made of conductive carbon paste or conductive silver paste material, which can effectively conduct low-frequency current and reduce skin irritation to ensure safety and comfort.

[0078] The portable pulse generator has a built-in modulated electrical stimulation module that can provide low-frequency and medium-frequency electrical pulses covering 1-9000Hz. The electrodes are applied to specific parts of the user (such as the knee joint, shoulder, waist, etc.) to regulate muscle contraction, reduce inflammation, and improve blood circulation. The device controls the intensity and frequency of the current, adjusts the electrical stimulation scheme in real time, and provides personalized electrical pulse output according to different disease courses, parts, and individual differences of users.

[0079] Inside the electrode patch, there are also three types of micro sensors integrated to collect the user's physiological data. Among them, four groups of bioelectrical impedance electrodes located at the edge of the electrode patch alternately send 1 kHz and 100 kHz safe AC signals to accurately distinguish the degree of edema in the superficial and deep tissues of the skin; a high-precision infrared temperature probe is embedded in the non-conductive area to continuously monitor the surface temperature fluctuation of the physiotherapy site and reflect the active state of inflammation; during the intermission of the electrical stimulation, the central electrical stimulation application area automatically switches to the surface electromyogram (sEMG) acquisition mode, that is, the conductive layer is reused as the signal acquisition interface, and a differential amplifier circuit is used to capture the micro current signal during muscle contraction to evaluate muscle fatigue and nerve control level. All sensors are hidden inside a 0.8-mm-thick medical silicone substrate, which not only ensures the accuracy of data collection but also maintains the flexible fitting characteristics of the electrode patch. The user only needs to attach and start it to achieve non-invasive intelligent monitoring.

[0080] These sensors can continuously monitor data during the electrical pulse output process, such as skin temperature, skin resistance, sEMG signals, etc., and feedback the data to the mobile APP in real time for the user and doctor to view. The real-time feedback enables the control system to automatically adjust the current intensity and duration according to the user's feedback, ensuring the accuracy and effectiveness of the applied electrical pulse stimulation, and can also automatically detect and alarm abnormal situations (such as too strong current, too high skin temperature, etc.). The user only needs to select the corresponding part for applying the electrical pulse stimulation, and the system will automatically start and recommend the appropriate duration, frequency, and current intensity.

[0081] In some embodiments, the electrical pulse generating device can also be configured with an edge computing module to extract the feature vectors of the real-time data collected by the sensors by running a lightweight model, so that the model on the mobile APP or in the cloud can directly perform relevant calculations based on the feature vectors.

[0082] This portable electrical pulse generating device is also configured with a communication module to support data transmission with the user device and the cloud. The portable device can also be built-in with a rechargeable lithium battery to support long-term continuous use. According to the physiotherapy intensity, the battery can work continuously for several hours. The portable device is equipped with a USB charging port or wireless charging function, which is convenient for users to charge at any time.

[0083] The design of this portable device is convenient for home use. The user only needs to wear the electrode patch and start the device to complete the electrical pulse stimulation of a specific part, thus eliminating the need for frequent medical treatment and drug dependence. Moreover, the portable device does not interfere with daily activities, and the user can carry it in work, leisure and other occasions to ensure continuous use.

[0084] The electric pulse control APP arranged in the user equipment has a simple and easy-to-use interaction interface. Users can input personal basic data (such as age, BMI), view the electric pulse parameters such as duration, current intensity, frequency, etc. in real time, and adjust the parameter settings. This APP can be divided into a doctor version and a user version. The user version can be used to receive user input and present the parameters during physical therapy to the user in real time. The doctor version can be used by doctors to view the user's physical therapy records and modify the electric pulse parameters.

[0085] All user data during the process of using the portable electric pulse generating device will be recorded and uploaded to the cloud, which is used to optimize the subsequent electric pulse stimulation plan or feedback the user's treatment effect (physiological state evolution). The artificial intelligence model arranged in the cloud can perform data analysis based on the user's historical data, automatically recommend the most suitable electric pulse stimulation plan, and continuously optimize it.

[0086] See Figure 1 , which shows a flowchart of a control method for an adaptive electric pulse generating device according to an embodiment of the present disclosure.

[0087] As Figure 1 shown, in step S1, the control system collects the user's inflammation-related data, including the electric stimulation application site, user basic data, and user physiological data; in step S2, according to the inflammation-related data, the control system extracts the user's biometric vector through the feature extraction module, determines the user's disease course classification according to the biometric vector through the disease course classification module, and then determines the corresponding electric pulse parameters according to the biometric vector and the disease course classification through the parameter generation module; in step S3, according to the electric pulse parameters, the control system applies the corresponding broadband electric pulse stimulation plan to the user at the electric stimulation application site through the adaptive electric pulse generating device.

[0088] Specifically, in step S1, the control system collects the user's inflammation-related data, including the electric stimulation application site, user basic data, and user physiological data.

[0089] The user first attaches the flexible electrode patch to the electric stimulation application site (such as the front side of the knee joint). After starting the control system, the portable pulse generating device automatically enters the data collection mode.

[0090] The portable pulse generating device collects the user's physiological data through various sensors built into the electrode patch. For example, it measures the initial tissue impedance through a bioelectrical impedance sensor, such as 380 Ω, and simultaneously uses a micro infrared probe to detect the surface temperature of the affected area (32.0 °C). The user selects the electric stimulation application site through the mobile phone APP, fills in the user basic data (age, height, and weight, and calculates the BMI from the height and weight), and feeds back the user's physiological data (such as the current pain level, visual analogue scale VAS from 0 to 10).

[0091] Here, the portable pulse generating device can collect all physiological signal baseline values within 30 seconds and encrypt and transmit the data to the cloud server.

[0092] According to an example, the control system can also obtain the user's electronic physiotherapy records. For example, the mobile phone APP records historical basic data of the patient, the location of inflammation, the process of disease development, and the electrical pulse parameters used during stage physiotherapy.

[0093] In step S2, the control system extracts a biometric feature vector through the feature extraction module based on the user's inflammation-related data, determines the disease course classification of the user through the disease course classification module based on the extracted biometric feature vector, and then determines the corresponding electrical pulse parameters through the parameter generation module based on the extracted biometric feature vector and the determined disease course classification.

[0094] According to an example, the feature extraction module generates a 128-dimensional biometric feature vector based on the electrical stimulation application site, user age, BMI, VAS, temperature, impedance, and sEMG collected by each sensor. The disease course classification module predicts the disease course classification of the user based on this biometric feature vector, such as [72% chronic phase, 25% acute postoperative phase, 2% subacute postoperative phase, 1% postoperative recovery phase]. The parameter generation module generates the corresponding electrical pulse parameters based on this biometric feature vector and the disease course classification "chronic phase", such as a frequency of 4000 Hz, an intensity of 3.8 mA, a square wave waveform, a duty cycle of 30%, and a duration of 28 minutes.

[0095] Here, the feature extraction module, the disease course classification module, and the parameter generation module are jointly trained through a case sample set.

[0096] The case sample set includes two types of sample data: source domain case sample data and target domain case sample data. The source domain case sample data is labeled complete case data, including the complete data of a large number of historical cases, such as multi-modal inflammation-related data, as well as disease course labels and electrical pulse parameters. The target domain case sample data is unlabeled patient raw data, including inflammation-related data of a small number of other patients, such as physiotherapy sites, user basic data, and user physiological data.

[0097] The process of jointly training the feature extraction module, the disease course classification module, and the parameter generation module according to the case sample set is as follows:

[0098] 1. Initial training of the feature extraction module and the disease course classification module

[0099] Input the complete data of the source domain case samples (multi-modal inflammation-related data + disease course labels + electrical pulse parameters) into the feature extraction module. The feature extraction module generates a biometric feature vector based on the inflammation-related data and outputs it to the disease course classification module for supervised training. The disease course classification module predicts the disease course classification based on the biometric feature vector generated by the feature extraction module.

[0100] According to an example, the feature extraction module receives the original inflammation data (such as sensor data like temperature, impedance, electromyogram signals, and basic information like age, BMI, etc.), and automatically extracts hidden biomedical features through a multi-layer neural network. The disease course classification module synchronously receives the extracted feature vector and learns to map it to the disease course labels (acute phase / chronic phase, etc.). For example, it is found that the "synchronous change of body temperature fluctuation and impedance decrease" is a potential sign of the acute phase.

[0101] When the disease course classification module makes a prediction error, backpropagation will simultaneously adjust the parameters of the feature extraction module and the disease course classification module, forcing the feature extraction module to generate features that are easier to classify.

[0102] According to an example, when the disease course classification module makes a prediction error, such as misclassifying the acute phase as the chronic phase, the loss function will backpropagate, forcing the feature extraction module to adjust the parameters: 1) Enhance the features sensitive to the acute phase (such as the rapid impedance decrease pattern); 2) Weaken the irrelevant features (such as the noise introduced by environmental temperature fluctuations).

[0103] In the initial training stage, the feature extraction module compresses the original data into a high-information-density biometric feature vector through end-to-end supervised learning. This vector retains the key biomarkers related to the disease course, filters out the noise caused by individual differences, and also establishes a unified feature representation benchmark across patients. The deep learning-based neural network encodes the complex patterns invisible to the naked eye into classifiable features through multi-layer non-linear transformations.

[0104] Here, the feature extraction module, for example, adopts a transformer structure and is trained to output a 128-dimensional biometric feature vector based on the user's inflammation-related data. The disease course classification module can be, for example, a classification network using a multi-head attention mechanism. For example, it divides the 128-dimensional input vector into 4 groups (each group is 32-dimensional), calculates the attention weights independently for each group, and outputs a four-dimensional probability vector P = [p1, p2, p3, p4] (corresponding to four disease course stages).

[0105] Thus, a preliminarily trained feature extraction module (capable of extracting key features related to the disease course) and a preliminarily available disease course classification module (capable of roughly judging the disease course stage based on features) are obtained. At this time, the feature extraction module obtained through training has established the basic feature representation ability, ensuring a reasonable starting point for subsequent optimization steps.

[0106] 2. Optimization Training of Feature Extraction Module

[0107] Input the inflammation-related data of source domain case samples and the inflammation-related data of target domain case samples into the feature extraction module to generate two batches of biometric vectors respectively. Calculate the distribution differences between the two batches of biometric vectors, and adjust the feature extraction module by minimizing the differences between them.

[0108] According to an example, input the inflammation-related data of the source domain and the target domain into the feature extraction module simultaneously to generate two batches of biometric vectors respectively. Calculate the distribution differences between the two batches of biometric vectors (such as statistical mean / variance differences), and adjust the parameters of the feature extraction module through an optimization algorithm to narrow the differences between them and achieve feature alignment between the source domain and the target domain. For example, if the impedance of patients in the target domain is generally lower than that in the source domain, the feature extraction module will learn to eliminate this systematic bias.

[0109] In this way, a feature extraction module that can generate "cross-domain general" features (the feature distributions of the source domain and the target domain are close) is obtained. The feature extraction module after optimization training can eliminate feature offsets caused by device differences and population differences, and improve the adaptability of the model to new patients.

[0110] In this optimization training step, retain the training of the disease course classification module to prevent damage to the existing classification ability of the disease course classification module.

[0111] 3. Adversarial Training of Feature Extraction Module and Domain Classification Module

[0112] Input the inflammation-related data of source domain case samples and the inflammation-related data of target domain case samples with their respective domain labels (source domain = 0 / target domain = 1) into the feature extraction module to generate biometric vectors. Input the generated biometric vectors into the domain classification module to determine whether the input biometric vectors are from the source domain or the target domain.

[0113] In adversarial training, the learning objective of the feature extractor is to extract features strongly related to the patient's physiological state (such as inflammation activity level, degree of edema) from the original data (inflammation-related data) and ignore the statistical biases between groups (such as the average BMI of patients in the source domain is relatively high), focusing on cross-group universal biomarkers (such as impedance-temperature correlation patterns). The adversarial objective of the domain classification module is to try to identify the data source from the input feature vectors (such as determining whether the features are from the source domain or the target domain), and through the Gradient Reversal Layer, the gradient direction received by the feature extractor is opposite to that of the domain classifier: when the domain classification module successfully distinguishes the feature source → the feature extraction module is punished and its parameters are adjusted; when the domain classification module fails to distinguish the feature source → the feature extraction module receives positive feedback.

[0114] By reversing the gradients of the domain classification module during backpropagation, the feature extraction module is forced to generate features that make the domain classification module unable to distinguish the source. Eventually, the accuracy of the output of the domain classification module approaches random guessing at about 50% (indicating that the features are fully cross-domain aligned). For example, the "impedance-temperature correlation pattern" of source domain patients is generalized into a common pattern that can also be understood in the target domain. By confusing the domain classification module (which cannot distinguish between the source domain / target domain), implicit feature encodings related to the device can also be indirectly eliminated.

[0115] In this way, a feature extraction module that can generate "domain-invariant" features (features that cannot be used to determine which domain a patient comes from) is obtained, and the training of the feature extraction module is completed. Through domain adversarial training, feature generalization is strengthened to ensure that the feature extraction module eliminates the implicit distribution differences between the source domain and the target domain (which may include mixed factors such as devices, patient groups, and environments), and retains inflammation-related biomarkers that are universal across patients.

[0116] For example, there are device differences between the source domain data and the target domain data. The source domain data comes from an old device (sampling rate 1 kHz, impedance detection error ±5 Ω). The target domain data comes from a new device (sampling rate 10 kHz, impedance error ±1 Ω). After adversarial training, the feature extraction module automatically normalizes the impedance value to a relative change rate (such as "impedance drops 10% per minute"), rather than relying on absolute dimensions. The sampling rate differences between the old and new devices are converted into unified time-frequency features (such as the main frequency of fluctuations, energy spectrum entropy).

[0117] Another example is that there are population differences between the patient data in the source domain and the patient data in the target domain. Most of the patients in the source domain are middle-aged and young (BMI 22-26, with strong inflammatory responses). The patients in the target domain are the elderly (BMI 28-32, with slow inflammatory responses). After adversarial training, the feature extraction module suppresses BMI-related features (such as the basic impedance value) and strengthens dynamic indicators of inflammatory activity (such as the goodness of fit of the temperature-impedance curve). The unique metabolic rate features of the elderly are encoded into a standardized inflammation index comparable to that of middle-aged and young people.

[0118] In the adversarial training step of the feature extraction module and the domain classification module, the biometric vectors generated by the feature extraction module based on the inflammation-related data of the source domain case samples can also be input into the disease course classification module to further optimize the disease course classification (using the true labels of the source domain data). At this time, the optimization objective of the disease course classification module (maximizing the disease course prediction accuracy) and the objective of the domain classification module (minimizing the domain classification accuracy) reach a dynamic balance through the parameter update of the feature extraction module. When the domain classification module attempts to better distinguish the source domain / target domain, the gradient received by the feature extraction module forces it to weaken the domain-related features; when the domain classification module cannot distinguish the domain source, the feature extraction module maintains the current parameter update direction. The error of the disease course classification module for the source domain data drives the feature extraction module to strengthen the inflammation-related features. Each round of parameter adjustment of the feature extraction module is jointly affected by the disease course classification error and the domain confusion degree, forming a two-objective optimization.

[0119] 4. Joint Optimization of the Disease Course Classification Module and the Parameter Generation Module

[0120] With the parameters of the feature extraction module fixed, the disease course classification module and the parameter generation module share the biometric vectors of the source domain after alignment output by the feature extraction module. The disease course classification module and the parameter generation module are jointly trained based on the biometric vectors of the source domain after alignment, as well as the disease course labels and electrical pulse parameters in the corresponding case samples.

[0121] The biometric vectors of the source domain after alignment output by the feature extraction module (reflecting the patient's physiological state) and the disease course classification output by the disease course classification module (providing prior knowledge of the disease course stage) are input into the parameter generation module, and the electrical pulse settings matching the historical physiotherapy parameters (electrical pulse parameters in the source domain) are learned through a regression task.

[0122] According to an example, the parameter generation module adopts a multi-branch conditional generation architecture to generate safe and controllable electrical pulse parameters based on the patient's individual characteristics and disease course stage. For example, the input layer receives a 128-dimensional domain-invariant feature vector and a 4-dimensional disease course probability vector, and realizes the personalized adaptation of the electrical pulse parameters through the gating mechanism of feature splicing and sub-parameter regression branches (such as the pulse frequency branch, the stimulation intensity branch, etc.).

[0123] By combining the loss function to constrain the accuracy of the course classification and the error of parameter generation, a collaborative cross-task collaborative optimization mechanism is constructed. For example, when the course classification module misclassifies a patient's chronic phase as the postoperative acute phase, the parameter generation module will generate medium-frequency pulse parameters based on the wrong classification, resulting in a significant deviation between the output parameters and the chronic phase medium-frequency physiotherapy plan actually required by the patient. At this time, the course classification error and the parameter regression error form a superposition effect in the loss function. The back propagation process simultaneously injects the classification logic correction gradient into the course classification module and transmits the parameter mapping rule adjustment signal to the parameter generation module. This two-way error correction mechanism forces the course classification module to improve its sensitivity to chronic phase biomarkers, while guiding the parameter generation module to establish a parameter fault-tolerant strategy under classification uncertainty. For example, when the classification confidence is lower than the threshold, the parameter template that is safe across the course of disease is preferred. Through loss coupling, the two modules form a closed-loop optimization of "classification guides parameter generation, and parameter feedback strengthens classification", ensuring that even under the constraint of a fixed feature space, the system can continue to improve the overall decision accuracy through task collaboration and avoid systematic deviations caused by overfitting of a single task.

[0124] This results in a disease course classification module that can dynamically determine the current disease course based on the patient's medical characteristics, and a parameter generation module that can generate personalized electric pulse parameters based on the patient's medical characteristics and disease course, thereby opening up the complete decision-making chain of "physiological data → disease course judgment → physical therapy plan" and ensuring that the generation of electric pulse parameters conforms to medical logic.

[0125] At this point, the training of the feature extraction module, disease course classification module and parameter generation module can be completed.

[0126] In some embodiments, the feature extraction module, the disease course classification module and the parameter generation module can be further fine-tuned through the following stage 5.

[0127] 5. Self-training optimization of feature extraction module, disease course classification module and parameter generation module

[0128] The inflammation-related data (without labels) of the target domain are input into the feature extraction module to generate a biological feature vector, and then the corresponding disease course classification is predicted by the disease course classification module. The biological feature vector and the disease course classification are input into the parameter generation module to obtain the electric pulse parameters. The disease course classification predicted by the disease course classification module and the electric pulse parameters generated by the parameter generation module are used as the "pseudo-labels" of the corresponding target domain data.

[0129] Select high-confidence prediction results from these results (such as samples with a prediction probability > 90%), and add them as temporary samples to the training dataset. Use the extended training dataset (source domain samples + temporary samples) to fine-tune the entire model (including the feature extraction module, disease course classification module, and parameter generation module) to enhance the adaptability to the unique patterns of the target domain. For example, if it is found that a certain type of new patient is more sensitive to low-frequency stimuli, then adjust the parameter generation strategy.

[0130] In the fine-tuning stage, the feature extraction module, disease course classification module, and parameter generation module are adjusted synchronously, but limited by the domain-invariant feature basis trained earlier to ensure that the optimization direction does not deviate from cross-domain consistency.

[0131] Thus, the entire model has the ability to make optimized decisions for target domain patients, can break through the limitations of source domain data, and use the actual data of the target domain to continuously improve the accuracy of personalized electrical pulse parameters.

[0132] Through the multi-stage collaborative training mechanism, the entire model has achieved an essential innovation in the parameter generation logic. Driven by the cross-domain feature alignment technology, the feature extraction module can map multi-source heterogeneous sensor data into a unified high-dimensional feature vector, breaking through the limitation that traditional devices rely on manually setting parameter combinations. For users who use the electrical pulse generation device for the first time, even if the distribution of their sensor data has a systematic deviation from the training source domain, the feature space optimized by domain adversarial training can still accurately capture the correlation patterns of impedance-temperature-electromyogram signals and generate electrical pulse parameters adapted to individual biological characteristics. For example, when it is detected that the electrical stimulation application site of a certain user shows medium-frequency impedance fluctuations coupled with low-frequency resonance of electromyogram signals, the model automatically matches the wide-frequency domain modulation strategy to improve the medium-frequency energy penetration efficiency while maintaining the comfort of the superficial nerves. For users who have used the device, the electrical pulse parameters can also be generated according to the real-time biological signals during each physiotherapy session, making the physiotherapy more in line with the user's current physiological state. Moreover, for users who have used the device, the model not only obtains the real-time biological signals but also retrieves the user's historical feature vectors and parameter response records, and analyzes the physiological state change trajectory through a time series model. For example, when a chronic inflammation user undergoes the third physiotherapy session, the model detects that the impedance fluctuation amplitude in this session is 35% lower than the previous two sessions, and combines the trend of the improvement of the medium-frequency energy absorption efficiency in the historical data to automatically adjust the pulse waveform duty cycle and introduce intermittent medium-frequency components.

[0133] The multi-modal data fusion architecture endows the model with decision-making dimensions beyond conventional parameter preset solutions. By analyzing the implicit correlations between sensor data, the model can autonomously discover efficient parameter combinations that are difficult to cover by artificial experience. For example, in some user groups, the combined pattern of intermittent medium-frequency pulses (7500 Hz) and low-frequency background waveforms (50 Hz) generated by the model can increase the energy density of the electrode patch by 2.1 times compared to traditional single-frequency outputs while maintaining the same somatic comfort level. This innovative parameter generation strategy stems from the model's in-depth exploration of impedance-frequency response curves in cross-domain cases, rather than relying on a preset rule base.

[0134] The cross-device compatibility of the model is achieved through feature-level abstraction. The differences in the original signals collected by different models of electrode patches are transformed into standardized biometric feature vectors during the feature extraction stage, ensuring the decision consistency of the parameter generation module. When the device is connected to a new model of electrode patch, after online feature alignment with a small number of samples, the stability of the output parameters can be maintained. This feature significantly reduces the model reconstruction cost during hardware iteration, enabling the model to have the ability to continuously evolve across generations of devices.

[0135] Compared with the mutually independent signal acquisition, parameter calculation, and output control modules in traditional devices, the model constructs an end-to-end optimization path from feature representation to parameter generation and has the ability of collaborative optimization. This architectural innovation enables the model to autonomously explore the efficient regions in the parameter space, such as discovering the non-linear response characteristics of a specific user group in the 3000 - 4500 Hz frequency band and generating corresponding parameters accordingly. This data-driven parameter discovery mechanism fundamentally changes the development paradigm of electrostimulation devices.

[0136] In some embodiments, before outputting the physiotherapy parameters, the parameter generation module can also search for similar cases in the case library according to the disease course classification results output by the disease course classification module, and the parameter generation module adjusts the generated electrical pulse parameters according to the distribution of physiotherapy parameters of these similar cases.

[0137] For example, the stimulation current calculated by the parameter generation module is 22 A, but the maximum safe intensity used by users with the same disease course in the first 50 similar cases in the case library is 20 A, then the pulse current of the current user is also reduced to 20 A.

[0138] The distribution of physiotherapy parameters of these similar cases can also be understood as physical boundary constraints, and the parameter generation module needs to adjust the generated electrical pulse parameters according to these constraints.

[0139] In some embodiments, before outputting the electrical pulse parameters, the system also performs boundary correction on the relevant electrical pulse parameters according to physiological safety rules. For example, for elderly users (>65 years old), even if the recommended intensity for similar cases is 15 mA, the system will automatically lower it to 12 mA according to the age attenuation coefficient; if abnormal skin sensitivity is detected (initial impedance < 100 Ω), the electrode patch heating function will be disabled. The final output parameters inherit the group treatment experience and are dynamically calibrated through individual physiological characteristics to achieve a balance between efficacy and safety.

[0140] According to another example, the user's historical feedback data can also be preferentially used to fine-tune the parameters. For example, if a certain patient was previously overly sensitive to a frequency of 50 Hz, the electrical pulse parameters this time will shift to the 60 Hz frequency band.

[0141] Subsequently, in step S3, the control system applies a corresponding broadband electrical pulse stimulation scheme to the user at the electrical stimulation application site according to the determined electrical pulse parameters.

[0142] The portable pulse generating device generates a corresponding electrical pulse stimulation scheme according to the electrical pulse parameters determined in step S2, which can cover the low and medium frequency bands of 1 - 9000 Hz, and its broadband modulation ability can accurately adapt to the treatment requirements of different tissue depths.

[0143] The current is evenly distributed to the physical therapy area through the conductive layer of the electrode patch, and the intensity gradually rises from 0 mA to the target value (such as 12 mA) to avoid sudden increase in stimulation. The impedance sensor built in the electrode patch real-time monitors the uniformity of the current distribution. If local impedance abnormality is detected (such as a sudden 20% increase in impedance in a certain area), the electrode polarity is automatically fine-tuned to optimize the electric field coverage.

[0144] The infrared sensor continuously monitors the temperature of the affected area. If it exceeds 39 °C or the temperature rises > 1 °C within 5 minutes, the physical therapy is immediately paused and the cooling protocol is initiated.

[0145] When the impedance < 100 Ω, electrode patch detachment or skin damage may occur, and the current output is cut off within 0.2 seconds.

[0146] According to an example, the APP interface synchronously displays the current electrical pulse parameters, such as waveform / intensity / frequency / remaining duration, and uses color gradient to prompt the physical therapy progress, such as red → green representing a decrease in inflammation activity.

[0147] In some embodiments, Figure 1 The process shown may also include step S4 (not shown). In step S4, the control system adjusts the electrical pulse parameters output by the portable pulse generating device in real time according to the real-time feedback data of the user during the current electrical pulse stimulation process.

[0148] Since the electrical pulses output by the portable pulse generation device can cover the low and medium frequency bands of 1 - 9000 Hz, its wide - frequency - domain modulation ability can accurately adapt to the physiotherapy needs of different tissue depths. For example, the low - frequency band (1 - 1000 Hz) is for superficial neuromuscular regulation, and the medium - frequency band (1000 - 9000 Hz) penetrates to deep - layer inflammatory lesions. By combining real - time feedback data, the optimal action frequency band is dynamically selected to achieve all - round bio - electrical intervention from the epidermis to the joint capsule.

[0149] During each physiotherapy process, the control system deeply integrates multi - modal sensor data and user feedback, and adjusts the electrical pulse parameters in real time through a reinforcement learning algorithm to dynamically optimize the treatment effect.

[0150] According to an example, the portable device runs a lightweight model through edge computing to extract the feature vectors of the sensors, ensuring that the inference latency of the reinforcement learning model deployed in the mobile phone APP is less than 50 milliseconds, meeting the real - time requirements.

[0151] According to an example, the reinforcement learning model deployed in the mobile phone APP is used to perform local real - time calculations to optimize the electrical pulse parameters for the user. The cloud can uniformly iterate the reinforcement learning model, so that the model of the mobile phone APP is updated according to the iteration of the cloud.

[0152] When the user starts the electrical pulse physiotherapy, the portable device first continuously collects real - time data of the physiotherapy site where electrical stimulation is applied through the sensors on the electrode patches, including physiological indicators such as skin temperature, impedance change, and muscle electrical signal intensity, and at the same time records the current output electrical pulse parameters (such as frequency, intensity, waveform). These raw data are preliminarily cleaned and standardized by the edge - computing module, removing abnormal noise interference, and unifying the signals of different sensors into a time - aligned digital sequence. For example, when a local temperature increase of 0.5 °C suddenly occurs during the physiotherapy of a certain user, the system immediately detects this change and marks it as a key event.

[0153] The processed data is input into a pre - trained reinforcement learning model, which converts it into a "feature vector" representing the current physiotherapy state. This vector contains implicit indicators such as inflammation activity, tissue response efficiency, and user tolerance. Based on this state feature and combined with historical physiotherapy records (such as the best parameter combinations for the user's past three physiotherapies), the model selects the next action from the preset safe parameter range: maintaining the current parameters, fine - tuning the frequency or intensity, switching the waveform mode, etc. For example, when it is detected that the impedance value continues to decrease and the user's feedback pain score increases, the model may decide to increase the frequency from 4000 Hz to 6000 Hz and at the same time reduce the intensity by 5% to balance comfort. If it is a new user using it for the first time, the model can generate 2 - 3 sets of candidate parameters (such as 4200 Hz / 4 mA, 3800 Hz / 3 mA) within the safe range and preferentially select the option with the highest expected therapeutic effect gain.

[0154] After executing the new pulse parameters, the system continues to collect physiological feedback data within the next 5 minutes, calculates the "treatment efficacy gain" (such as the edema index decline rate) and the "side effect cost" (such as the degree of skin flushing) of this adjustment, and comprehensively generates a reward signal. For example, if a certain parameter adjustment reduces the pain score by 2 points but causes mild muscle twitching, the reward signal will weigh these two results.

[0155] The reward signal triggers the model parameter update mechanism, and the reinforcement learning algorithm dynamically adjusts the decision-making strategy according to the physiotherapy effect. If a certain parameter combination continuously obtains high rewards three times (such as the "5000Hz square wave + medium intensity" is generally effective in chronic patients), the model will increase its selection priority; conversely, the parameter combinations that frequently cause discomfort will be suppressed. The updated model preferentially calls the optimized policy library in the next decision, forming a continuous optimization closed-loop of "execution → observation → learning".

[0156] The control system of the present disclosure realizes the dynamic matching of electrical stimulation parameters and the user's physiological state through reinforcement learning, and its core value lies in the real-time personalized adaptation ability. Traditional devices rely on fixed programs or manual parameter adjustment, and it is difficult to cope with the non-linear changes in tissue response during physiotherapy. However, the control system of the present disclosure can autonomously explore the optimal parameter range in each physiotherapy. For example, for a certain arthritis patient, due to the recurrence of inflammation during the mid-term of physiotherapy, the impedance mutates. The system automatically completes three parameter iterations within 10 minutes and finally locks in the 8000Hz medium-frequency short pulse mode that is effective for him, and the treatment effect is improved by 40% compared with the traditional preset scheme.

[0157] Furthermore, the system autonomously discovers cross-population rules that are difficult for humans to summarize through a large amount of individual data. For example, for the obese user group with BMI>30, the model finds that the combination of "low-frequency carrier wave + intermittent medium-frequency pulse" can effectively penetrate the fat layer. In addition, the user's manual intervention behavior (such as actively reducing the intensity) is converted into safety constraint rules to form a personalized risk barrier. After a user manually reduces the intensity three times, the system automatically lowers the intensity upper limit by 15%, reducing the incidence of adverse reactions in subsequent physiotherapy to zero. This intelligent dynamic balance mechanism not only improves the treatment effect and safety, but also promotes the paradigm innovation of electrical stimulation physiotherapy from "programmed preset" to "autonomous evolution".

[0158] In some embodiments, when ending the use of the portable pulse generating device this time, the control system updates the individual physiological state evolution prediction of the user according to the user's historical physiotherapy data (such as physiological state data, electrical pulse parameters), and generates a personalized recommendation for the user.

[0159] After this physiotherapy ends, the control system constructs a dynamic treatment effect prediction (i.e., physiological state evolution prediction) model based on the user's historical data to achieve full-cycle health management.

[0160] First, integrate the user's multi-source treatment data, such as historical disease process sequences (the evolution path of disease course labels in the past 30 days, e.g., "acute phase → subacute phase → chronic phase"), physiological state response curves (average daily improvement rate of impedance, weekly decline rate of pain score, muscle function recovery slope based on sEMG), and external environmental factors (seasonal temperature changes, user's exercise intensity log). After standardizing and aligning these data with time, input them into a pre-trained Transformer time series prediction model.

[0161] The prediction model captures long-range dependencies through the self-attention mechanism, such as identifying individualized physiotherapy response patterns like "impedance improvement lags behind pain relief", and then outputs two core predictions:

[0162] 1) Inflammation recurrence risk: Calculate the recurrence probability in the next two weeks (e.g., 58%). If the probability exceeds 50%, the system automatically recommends preventive intervention measures, such as increasing the physiotherapy frequency from 3 times a week to 5 times and superimposing low-frequency maintenance stimulation (10 Hz, 200 μs).

[0163] 2) Efficacy saturation warning: Analyze the decay trend of the repair rate (e.g., the impedance improvement rate drops from 2% / day to 0.5% / day in the past week), and dynamically adjust the treatment plan, such as prompting "the current repair has entered a plateau phase, it is recommended to change to physiotherapy every other day and combine with hot compress".

[0164] The prediction results of the model can be visually presented through the user APP. For example, present a rehabilitation progress bar to the user, horizontally compare the expected recovery curve with the actual progress, and mark key nodes, such as "5 more physiotherapy sessions are needed to reach a knee flexion angle of 90°". It can also provide environmental sensitivity tips to the user, such as generating a warning by combining external data, specifically like "a cold snap is coming in the next week, the risk of arthritis recurrence increases by 15%, it is recommended to increase the duration of home physiotherapy".

[0165] According to an example, after the end of this physiotherapy session, the control system synchronizes the user's historical physiotherapy data to the doctor; receives the doctor's modification of the user's disease course and / or the user's electrical pulse parameters.

[0166] The user's historical physiotherapy data is also synchronized to the doctor side. Through the doctor version of the electrical pulse control APP, the doctor can query the user's physiotherapy data and modify the disease course or electrical pulse parameters determined by the control system for the user, enabling the doctor to remotely guide the patient's inflammation physiotherapy.

[0167] It should be noted that the various embodiments of the present disclosure may be implemented in software and / or a combination of software and hardware, for example, may be implemented using an application specific integrated circuit (ASIC), a general purpose computer, or any other similar hardware device. In one embodiment, the software program of the various embodiments of the present disclosure may be executed by a processor to implement the steps or functions described above. Similarly, the software program of the various embodiments of the present disclosure (including related data structures) may be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive, or a floppy disk and the like. In addition, some steps or functions of the various embodiments of the present disclosure may be implemented using hardware, for example, as a circuit that cooperates with a processor to perform various steps or functions.

[0168] In addition, at least a portion of the embodiments of the present disclosure may be applied as a computer program product, such as a computer program instruction, which, when executed by a computing device, can call or provide the method and / or technical solution according to the embodiments of the present disclosure through the operation of the computing device. The program instruction for calling / providing the method of the embodiments of the present disclosure may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in a working memory of a computing device that runs according to the program instruction.

[0169] For those skilled in the art, it is obvious that the embodiments of the present disclosure are not limited to the details of the above exemplary embodiments, and the embodiments of the present disclosure can be implemented in other specific forms without departing from the spirit or basic features of the embodiments of the present disclosure. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the embodiments of the present disclosure is limited by the attached claims rather than the above description, so it is intended to include all changes within the meaning and scope of the equivalent elements of the claims in the embodiments of the present disclosure. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any specific order.

Claims

1. A control method for an adaptive electric pulse generating device, wherein, The method includes the following steps: Collect inflammation-related data of the user, including the electrostimulation application site, the user's basic data, and the user's physiological data; According to the inflammation-related data, extract the biometric vector of the user through a feature extraction module, determine the disease course classification of the user through a disease course classification module based on the biometric vector, and then determine the corresponding electro-pulse parameters through a parameter generation module based on the biometric vector and the disease course classification; Among them, the feature extraction module, the disease course classification module, and the parameter generation module are trained through a case sample set: The case sample set includes labeled source domain samples and unlabeled target domain samples. The source domain samples include the inflammation-related data of the patient, the corresponding disease course label, and the electro-pulse parameters; Use the source domain samples to initially train the feature extraction module and the disease course classification module; Input the inflammation-related data of the source domain samples and the target domain samples into the feature extraction module, and adjust the feature extraction module by minimizing the distribution difference of the biometric vectors output in the two domains; Use the inflammation-related data of the source domain samples and the target domain samples with domain labels to perform adversarial training on the feature extraction module and the domain classification module, and at the same time use the source domain samples to optimize the feature extraction module and the disease course classification module; Fix the parameters of the feature extraction module, and jointly train the disease course classification module and the parameter generation module based on the biometric vector of the source domain samples output by the feature extraction module; According to the electro-pulse parameters, apply a corresponding wide-frequency domain electro-pulse stimulation scheme to the user at the electrostimulation application site through the adaptive electro-pulse generating device.

2. The method according to claim 1, wherein The training of the feature extraction module, the disease course classification module, and the parameter generation module further includes: Use the high-confidence prediction results generated based on the target domain samples as temporary samples to be added to the case sample set, and adjust the feature extraction module, the disease course classification module, and the parameter generation module through the expanded case sample set.

3. The method according to claim 1, wherein, The method further includes the following steps: According to the real-time feedback data of the user during the current electro-pulse stimulation process, adjust the electro-pulse parameters in real time.

4. The method according to claim 1, wherein, The method further includes the following steps: When ending the current use of the adaptive electro-pulse generating device, update the prediction of the user's individual physiological state evolution based on the user's historical physiotherapy data and generate personalized suggestions for the user.

5. The method according to claim 1, wherein, Before outputting the electro-pulse parameters to the adaptive electro-pulse generating device, the method further includes the following steps: The parameter generation module searches for similar cases in the case library according to the disease course classification determined by the disease course classification module, and then adjusts the generated electro-pulse parameters according to the electro-pulse parameter distribution of the similar cases.

6. The method according to claim 1 or 5, wherein The method further includes the following steps: The parameter generation module performs boundary correction on the relevant electro-pulse parameters according to the physiological safety rules.

7. The method according to claim 1, wherein The method further includes the following steps: After finishing the use of the adaptive electro-pulse generating device, synchronize the user's historical physiotherapy data to the doctor's dedicated channel; receive the doctor's modification of the user's disease course classification and / or the user's electro-pulse parameters.

8. A control system for an adaptive electrical pulse generating device, comprising an adaptive electrical pulse generating device, a user device, and a network device; wherein, The adaptive electro-pulse generating device is connected with electrode patches, and a plurality of sensors are arranged on the electrode patches; wherein, the adaptive electro-pulse generating device is configured to perform the following operations: Collect the user's inflammation-related data through the sensors, including the user's physiological data; According to the electro-pulse parameters, apply a corresponding wide-frequency domain electro-pulse stimulation scheme to the user at the electro-stimulation application site through the electrode patches; The user device is configured to perform the following operations: Collect the user's inflammation-related data, including at least the electro-stimulation application site and the user's basic data; The network device is configured to perform the following operations: According to the inflammation-related data, extract the user's biometric feature vector through the feature extraction module, determine the user's disease course classification through the disease course classification module according to the biometric feature vector, and then determine the corresponding electro-pulse parameters through the parameter generation module according to the biometric feature vector and the disease course classification; Among them, the feature extraction module, the disease course classification module and the parameter generation module are trained through a case sample set: The case sample set includes labeled source domain samples and unlabeled target domain samples, and the source domain samples include the user's inflammation-related data, corresponding disease course labels and electro-pulse parameters; Use the source domain samples to perform initial training on the feature extraction module and the disease course classification module; Input the inflammation-related data of the source domain samples and the target domain samples into the feature extraction module, and adjust the feature extraction module by minimizing the distribution difference of the biometric feature vectors output in the two domains; Perform adversarial training on the feature extraction module and the domain classification module by using the inflammation-related data of the source domain samples and the target domain samples with domain labels, and at the same time optimize the feature extraction module and the disease course classification module by using the source domain samples; Fix the parameters of the feature extraction module, and jointly train the disease course classification module and the parameter generation module according to the biometric feature vector of the source domain samples output by the feature extraction module.

9. The system according to claim 8, wherein, The training of the feature extraction module, the disease course classification module and the parameter generation module further includes: Use the high-confidence prediction results generated based on the target domain samples as temporary samples to be added to the case sample set, and adjust the feature extraction module, the disease course classification module and the parameter generation module through the updated case sample set.

10. The system according to claim 8 or 9, wherein The user device is further configured to perform the following operations: According to the real-time feedback data of the user during the current electro-pulse stimulation process, adjust the electro-pulse parameters in real time.

Images