Individualized electrical stimulation parameter AI data modeling method and system

Abstract

The invention discloses an individualized electrical stimulation parameter AI data modeling method, and belongs to the technical field of artificial intelligence data modeling. According to the method, a multi-modal state feature set is constructed by collecting surface electromyogram signals, stimulation loop impedance, physiological cycle information and user subjective tolerance data; on the basis, an improved support vector machine model based on a physiological cycle sensitive weight is used for modeling the electrical stimulation effectiveness, and an optimal treatment window for effective muscle contraction and user comfort is determined; a fuzzy PID control algorithm fused with physiological tolerance is further combined to dynamically adjust the stimulation frequency, the pulse width and the current intensity; and through historical data iteration modeling, an individualized initial parameter library and a rehabilitation trend evaluation result are formed. According to the invention, on the premise of guaranteeing the safety, individualization, self-adaption and long-term optimization of electrical stimulation parameters can be realized, and the effectiveness, comfort and stability of electrical stimulation treatment are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence data modeling technology, specifically to an AI data modeling method and system for individualized electrical stimulation parameters. Background Technology

[0002] A personalized electrostimulation parameter AI data modeling method refers to a method that uses artificial intelligence and data modeling technology to comprehensively analyze multi-source physiological signals, individual physiological states, and subjective feedback information collected during electrostimulation therapy, establish a mapping relationship between stimulation parameters and human responses, and on this basis, realize the personalized setting and dynamic optimization of electrostimulation parameters. This method can adaptively adjust the stimulation intensity, pulse width, and frequency according to the stimulation response characteristics of different individuals under different physiological states, thereby improving treatment effectiveness and comfort while ensuring safety, and providing stable and sustainable parameter optimization support for long-term rehabilitation.

[0003] However, existing methods for analyzing and modeling individualized electrical stimulation parameters have technical problems. They are based on a single physiological signal or static empirical parameters, making it difficult to simultaneously characterize physiological cycle differences, changes in subjective tolerance, and dynamic response characteristics during stimulation. Furthermore, treatment parameters are difficult to automatically evolve and update with the course of treatment, resulting in initial parameter settings relying on human experience, insufficient individual adaptability, and poor long-term efficacy stability.

[0004] Existing methods for modeling the effectiveness of electrical stimulation have limitations. They rely solely on electromyography amplitude or stimulation intensity thresholds as criteria, fail to adequately consider the modulating effect of physiological cycles on pain threshold and stimulation response, and have a rough identification of the boundary between "ineffective stimulation" and "overstimulation." These limitations can lead to technical problems in practical applications, such as insufficient muscle recruitment or significant discomfort or even pain.

[0005] Existing methods for dynamic parameter adjustment suffer from technical problems such as directly relying on linear adjustment of current amplitude, ignoring charge distribution patterns and individual tolerance differences, easily resulting in strong stimulation steps, large fluctuations in comfort, and difficulty in timely and smooth transitions when fatigue or pain increases. Summary of the Invention

[0006] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a method and system for individualized electrical stimulation parameter AI data modeling. The technical solution adopted by this invention is as follows: This invention provides a method for individualized electrical stimulation parameter AI data modeling, which includes the following steps:

[0007] Step S1: Data Acquisition;

[0008] Step S2: Modeling the effectiveness of electrical stimulation;

[0009] Step S3: Dynamically adjust parameters;

[0010] Step S4: Individualized electrical stimulation data modeling.

[0011] Further, in step S1, the data acquisition is used to construct a multidimensional input vector space containing physiological characteristics, physical environment and subjective perception characteristics, to provide the original basis for subsequent analysis. Specifically, it involves real-time acquisition of surface electromyographic signals and loop impedance values ​​of the pelvic floor muscle group through electrodes, and obtaining the user's physiological cycle phase and current pain threshold score in conjunction with the accompanying APP to obtain initial multidimensional data. Then, by performing timestamp alignment and noise cleaning on the initial multidimensional data, a multimodal state feature set is obtained.

[0012] The multimodal state feature set includes electromyographic features, tissue conductivity, physiological cycle weights, and subjective tolerance parameters.

[0013] Further, in step S2, the electrical stimulation effectiveness modeling is used to determine the actual response quality of the current electrical stimulation parameters in the user's body and identify the boundary between ineffective and excessive stimulation. Specifically, based on the multimodal state feature set and the collected electromyographic signals, muscle recruitment rate and fatigue index features are calculated, and combined with the loop impedance value, the electrode fit status is determined. The muscle recruitment rate and the user's pain threshold score are correlated, and a classification model based on an improved support vector machine with physiological cycle sensitive weights is constructed. Through cross-validation, the parameter range in which the muscle produces effective contraction and the user feels no pain is found, thus obtaining reference data for electrical stimulation effectiveness. This includes the following steps:

[0014] Step S21: Multidimensional feature quantization, used to convert the original physical electrical signal into a discrete feature vector reflecting the muscle functional state. Specifically, it calculates the ratio of the peak-to-peak value of the electromyographic signal induced by electrical stimulation to the maximum voluntary contraction value to obtain the muscle recruitment rate, extracts the rate of change of the median frequency of the surface electromyographic signal to obtain the fatigue index feature, and combines the real-time measured loop impedance value to determine the electrode contact state, thus obtaining a standardized feature input set.

[0015] Step S22: Construction of physiological cycle sensitive kernel function, which is used to introduce biological variation parameters to correct the model's perception accuracy of pain threshold at different physiological stages. Specifically, it is to establish corresponding weight factors according to the user's physiological cycle phase and embed them into the radial basis kernel function to construct a physiological cycle sensitive kernel function, thereby obtaining a kernel feature mapping space with biological adaptive capability.

[0016] Step S23: Cost-sensitive support vector machine training, used to find the optimal treatment classification boundary while ensuring user comfort. Specifically, a cost-sensitive learning mechanism is adopted to give higher penalty weights to misclassification of pain state in classification errors. The physiological cycle-sensitive kernel function is used to perform nonlinear classification training on the quantified features, and the model hyperparameters are optimized through cross-validation to obtain the stimulus response classification model.

[0017] Step S24: Optimal treatment window threshold extraction, used to extract the safe range of current parameters that can be executed by hardware from the classification space. Specifically, within the decision space generated by the stimulus response classification model, the overlapping part of the effective muscle contraction area and the user's pain-free area is identified, and the corresponding upper and lower limits of current intensity are calculated to obtain the optimal treatment window threshold.

[0018] Step S25: Calculate the electrical stimulation effectiveness evaluation index, which is used to quantitatively evaluate the efficiency of the current electrical stimulation program. Specifically, it involves establishing a comprehensive evaluation function based on muscle recruitment rate, pain threshold score, current intensity and fatigue characteristics, calculating the electrical stimulation effectiveness evaluation index, and using it together with the optimal treatment window threshold to form reference data for electrical stimulation effectiveness.

[0019] The reference data on the effectiveness of electrical stimulation specifically includes an electrical stimulation effectiveness assessment index and the corresponding optimal treatment window threshold.

[0020] Further, in step S3, the dynamic adjustment of parameters is used to optimize the output waveform in real time based on the evaluation results. Specifically, it employs a fuzzy PID control algorithm improved by incorporating physiological tolerance, using reference data on the effectiveness of electrical stimulation as a feedback variable. When the index is lower than the effective threshold, the pulse width is increased smoothly rather than the current intensity is increased directly. When excessive muscle fatigue or an increase in the pain threshold score is detected, it automatically switches to a low-frequency analgesia mode or increases the pulse interval time, while always clamping the output intensity within the safety standard limits for Class II medical devices, thus obtaining adaptive electrical stimulation execution parameters, including the following steps:

[0021] Step S31: Efficacy reference normalization, used to convert treatment efficacy into control variables, specifically by calculating the deviation and rate of change between the real-time efficacy index and the target value, to obtain the normalized control error characteristics;

[0022] Step S32: Adaptive parameter tuning improves fuzzy control, which is used to dynamically adjust the system response speed according to individual differences. Specifically, it introduces a physiological sensitivity factor based on the treatment window width, and corrects the proportional, integral and derivative gain coefficients of the PID controller in real time through fuzzy logic rules to obtain the adaptive control gain matrix.

[0023] Step S33: Based on energy balance, dual-loop decoupling optimization is used to improve stimulation comfort while ensuring effectiveness. Specifically, the total output of the PID controller is preferentially allocated to the pulse width parameter, and the current intensity compensation adjustment is started after the pulse width reaches the preset safety threshold to obtain a smoothly evolving waveform control command.

[0024] Step S34: Fatigue monitoring and control, used to prevent muscle overexertion and relieve discomfort, specifically monitors the fatigue index and pain threshold score. When the value exceeds the limit, the stimulation frequency is automatically reduced and the pulse interval time is extended to obtain the analgesia and recovery protection mode parameters.

[0025] Step S35: Safety clamping parameter output, used to ensure that the output complies with the safety limits of medical devices. Specifically, the waveform control command is compared and truncated with the preset Class II medical device safety standard limits to obtain the final adaptive electrical stimulation execution parameters.

[0026] The adaptive electrical stimulation execution parameters include frequency, pulse width, and current intensity.

[0027] Further, in step S4, the individualized electrical stimulation data modeling is used to establish an individualized historical response benchmark and rehabilitation trend data model. Specifically, the physiological cycles collected in step S1 during each treatment, the reference data on the effectiveness of electrical stimulation determined in step S2, and the finally stable adaptive electrical stimulation execution parameters in step S3 are stored in a local database. A weighted moving average algorithm or linear regression trend analysis is used to calculate the average parameter preference of the user under different physiological cycle phases. Before the next treatment is started, the system automatically retrieves and matches the closest historical optimal parameter combination from the local database as the initial preset value according to the physiological cycle to which the current date belongs, thus obtaining the user's individualized dynamic initial electrical stimulation parameter library and pelvic floor muscle rehabilitation trend evaluation curve.

[0028] This invention provides an AI data modeling system for individualized electrical stimulation parameters, comprising a data acquisition module, an intelligent modeling module, and a data iteration module;

[0029] The data acquisition module is used for data acquisition, obtaining a multimodal state feature set through data acquisition, and sending the multimodal state feature set to the intelligent modeling module and the data iteration module;

[0030] The intelligent modeling module is used for electrical stimulation effectiveness modeling and parameter dynamic adjustment. Through electrical stimulation effectiveness modeling and parameter dynamic adjustment, it obtains electrical stimulation effectiveness reference data and adaptive electrical stimulation execution parameters, and sends the electrical stimulation effectiveness reference data and adaptive electrical stimulation execution parameters to the data iteration module.

[0031] The data iteration module is used for individualized electrical stimulation data modeling. Through individualized electrical stimulation data modeling, an individualized dynamic initial electrical stimulation parameter library and a pelvic floor muscle rehabilitation trend assessment curve are obtained for the user.

[0032] The beneficial effects achieved by the present invention using the above solution are as follows:

[0033] (1) In view of the existing individualized electrical stimulation parameter data analysis and modeling methods, which are based on a single physiological signal or static empirical parameters, it is difficult to simultaneously characterize physiological cycle differences, subjective tolerance changes and dynamic response characteristics during stimulation, and the treatment parameters are difficult to automatically evolve and update with the course of treatment, resulting in the initial parameter setting relying on human experience, insufficient individual adaptability and poor long-term efficacy stability. This solution creatively adopts a comprehensive data modeling, adjustment and iteration method that combines electrical stimulation effectiveness modeling and parameter dynamic adjustment. It unifies the modeling and closed-loop update of multimodal state characteristics, electrical stimulation effectiveness reference data and historical stable execution parameters, and realizes adaptive initialization, dynamic optimization and long-term individualized evolution of electrical stimulation parameters under different physiological cycles and individual differences, thereby significantly improving the accuracy of parameter recommendation, treatment continuity and the predictability of rehabilitation process;

[0034] (2) In view of the existing methods for modeling the effectiveness of electrical stimulation, there are technical problems such as using only electromyographic amplitude or stimulation intensity threshold as the criterion, not fully considering the modulation effect of physiological cycle on pain threshold and stimulation response, and the rough identification of the boundary between "ineffective stimulation" and "overstimulation", which easily leads to insufficient muscle recruitment or obvious discomfort or even pain in practical applications. This solution creatively constructs a classification model based on physiological cycle sensitive weights and finds the parameter range in which the muscle produces effective contraction and the user has no pain through cross-validation. This modeling of the effectiveness of electrical stimulation achieves refined identification and quantitative expression of the optimal treatment window, so that the selection of electrical stimulation parameters is both physiologically adaptable and takes into account comfort and safety.

[0035] (3) In view of the technical problems in the existing dynamic parameter adjustment methods, which directly rely on the linear adjustment of current amplitude, ignore the charge distribution mode and individual tolerance differences, are prone to strong stimulation step sensation, large comfort fluctuations, and are difficult to smoothly transition in time when fatigue or pain increases, this solution creatively adopts a fuzzy PID control algorithm that integrates physiological tolerance improvement, combines the electrical stimulation effectiveness evaluation index as a feedback variable, and introduces a physiological sensitivity factor based on the treatment window width and a dual-loop decoupled adjustment mechanism of energy. Energy compensation is achieved first through pulse width adjustment, and current follow-up compensation is performed after pulse width saturation. At the same time, with the help of energy gradient smoothing and safety clamping control, the electrical stimulation parameters are smoothly evolved and comfortably output under the premise of ensuring efficacy, effectively reducing the risk of discomfort and improving the safety and clinical usability of dynamic adjustment. Attached Figure Description

[0036] Figure 1 A flowchart illustrating an AI data modeling method for individualized electrical stimulation parameters provided by this invention;

[0037] Figure 2 A schematic diagram of an AI data modeling system for individualized electrical stimulation parameters provided by the present invention;

[0038] Figure 3 A flowchart illustrating the modeling of the effectiveness of electrical stimulation in step S2;

[0039] Figure 4 This is a flowchart illustrating the dynamic adjustment of parameters in step S3.

[0040] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0042] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0043] Example 1, see Figure 1 The present invention provides an AI data modeling method for individualized electrical stimulation parameters, the method comprising the following steps:

[0044] Step S1: Data Acquisition;

[0045] Step S2: Modeling the effectiveness of electrical stimulation;

[0046] Step S3: Dynamically adjust parameters;

[0047] Step S4: Individualized electrical stimulation data modeling.

[0048] By performing the above operations, this solution addresses the technical problems inherent in existing individualized electrical stimulation parameter data analysis and modeling methods. These methods rely solely on single physiological signals or static empirical parameters, making it difficult to simultaneously characterize physiological cycle differences, changes in subjective tolerance, and dynamic response characteristics during stimulation. Furthermore, treatment parameters are difficult to automatically evolve and update with the course of treatment, leading to initial parameter settings dependent on human experience, insufficient individual adaptability, and poor long-term efficacy stability. This solution creatively employs a comprehensive data modeling, adjustment, and iteration method that combines electrical stimulation effectiveness modeling with dynamic parameter adjustment. It unifies the modeling and closed-loop updating of multimodal state characteristics, electrical stimulation effectiveness reference data, and historically stable execution parameters, enabling adaptive initialization, dynamic optimization, and long-term individualized evolution of electrical stimulation parameters under different physiological cycles and individual differences. This significantly improves the accuracy of parameter recommendations, treatment continuity, and the predictability of the rehabilitation process.

[0049] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the data acquisition is used to construct a multi-dimensional input vector space containing physiological characteristics, physical environment and subjective perception characteristics, to provide the original basis for subsequent analysis. Specifically, it involves real-time acquisition of surface electromyography signals and loop impedance values ​​of the pelvic floor muscle group through electrodes, and obtaining the user's physiological cycle phase and current pain threshold score in conjunction with the supporting APP to obtain initial multi-dimensional data. Then, by performing timestamp alignment and noise cleaning on the initial multi-dimensional data, a multimodal state feature set is obtained.

[0050] Preferably, the surface electromyography (EMG) signal is continuously acquired using a fixed sampling frequency, and after acquisition, the original EMG signal is subjected to bandpass filtering and power frequency interference suppression processing to reduce the impact of environmental noise and motion artifacts on signal quality; the loop impedance value is obtained by calculating the voltage and current changes in the electrical stimulation loop, and is used to reflect the changes in electrode adhesion state and tissue electrical properties.

[0051] The physiological cycle phases include the follicular phase, ovulation phase, luteal phase, and menstruation.

[0052] More preferably, corresponding cycle weight parameters are set for different physiological cycle phases to characterize the degree of influence of physiological state changes on electrical stimulation response. The cycle weight parameters are used as one of the input features in the subsequent modeling process.

[0053] The multimodal state feature set includes electromyographic features, tissue conductivity, physiological cycle weights, and subjective tolerance parameters.

[0054] As a further improvement to this embodiment, before constructing the multimodal state feature set, the surface electromyography signal, loop impedance data, physiological cycle phase information and pain threshold score are uniformly timestamped and aligned, and outlier removal and noise cleaning are performed on the aligned data to ensure the consistency and reliability of data from different sources on the same time scale; the multimodal state feature set serves as the input feature set for subsequent electrical stimulation effectiveness modeling and dynamic parameter adjustment.

[0055] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the electrical stimulation effectiveness modeling is used to determine the actual response quality of the current electrical stimulation parameters in the user's body and identify the boundary between ineffective stimulation and overstimulation. Specifically, based on the multimodal state feature set and the collected electromyographic signals, the muscle recruitment rate and fatigue index features are calculated, and combined with the loop impedance value, the electrode contact status is determined. The muscle recruitment rate and the user's pain threshold score are correlated and analyzed to construct a classification model based on the improved support vector machine with physiological cycle sensitive weights. Through cross-validation, the parameter range in which the muscle produces effective contraction and the user feels no pain is found, and reference data on the effectiveness of electrical stimulation is obtained. This includes the following steps:

[0056] Step S21: Multidimensional feature quantization, used to convert the original physical electrical signal into a discrete feature vector reflecting the muscle functional state. Specifically, it calculates the ratio of the peak-to-peak value of the electromyographic signal induced by electrical stimulation to the maximum voluntary contraction value to obtain the muscle recruitment rate, extracts the rate of change of the median frequency of the surface electromyographic signal to obtain the fatigue index feature, and combines the real-time measured loop impedance value to determine the electrode contact state, thus obtaining a standardized feature input set.

[0057] Preferably, the formula for calculating the muscle recruitment rate is:

[0058] ;

[0059] In the formula, R acc eEMG is the muscle recruitment rate, used to characterize the relative degree to which the target muscle group is activated under the current electrical stimulation parameters. peak-to-peakIt is the peak-to-peak amplitude of the surface electromyography (EMG) signal induced by electrical stimulation. Specifically, it is the amplitude difference between the maximum positive peak value and the minimum negative peak value of the surface EMG signal within a single stimulation cycle or a preset time window, which is used to reflect the intensity of instantaneous muscle electrical activity induced by electrical stimulation.

[0060] The fatigue index feature is specifically calculated by performing a short-time Fourier transform on the electromyographic signal induced by electrical stimulation and extracting the rate of change of the median frequency.

[0061] More preferably, the step of determining the electrode contact status by combining the real-time measured loop impedance value specifically involves setting a preset threshold. If the real-time measured loop impedance value exceeds the preset threshold, the electrode contact is determined to be poor, and the input vector is penalized and corrected.

[0062] Step S22: Construction of physiological cycle sensitive kernel function, which is used to introduce biological variation parameters to correct the model's perception accuracy of pain threshold at different physiological stages. Specifically, it is to establish corresponding weight factors according to the user's physiological cycle phase and embed them into the radial basis kernel function to construct a physiological cycle sensitive kernel function, thereby obtaining a kernel feature mapping space with biological adaptive capability.

[0063] The formula for calculating the physiological cycle-sensitive kernel function is as follows:

[0064] ;

[0065] In the formula, It is a physiological cycle-sensitive kernel function, x i It is the input sample in the standardized feature input set, x j It is an adjacent input sample. This is a physiological cycle weighting factor used to modulate the sample distance based on the current physiological cycle phase of the user, in order to reflect the differences in pain threshold and stimulus response at different physiological stages. It is the scaling parameter of the kernel function, used to control the smoothness and generalization ability of the radial basis kernel function;

[0066] Step S23: Cost-sensitive support vector machine training, used to find the optimal treatment classification boundary while ensuring user comfort. Specifically, a cost-sensitive learning mechanism is adopted to give higher penalty weights to misclassification of pain state in classification errors. The physiological cycle-sensitive kernel function is used to perform nonlinear classification training on the quantified features, and the model hyperparameters are optimized through cross-validation to obtain the stimulus response classification model.

[0067] The aforementioned cost-sensitive learning mechanism assigns a higher penalty weight to misclassifications of pain states, specifically by introducing a cost-symmetric loss function and constructing an optimization objective function, the calculation formula of which is:

[0068] ;

[0069] In the formula, is the weight vector of the classification hyperplane in the support vector machine, and b is the bias term of the classification hyperplane. Here, is a slack variable, C is a penalty coefficient used to balance the trade-off between model complexity and classification error, n is the total number of samples, and i is the sample index. It is a cost-symmetric loss weight function used to assign different penalty weights to different types of classification errors. Among them, samples that are misclassified as "painful" are given a higher loss weight. It is the slack variable corresponding to the i-th sample, used to allow soft-margin classification in the case of non-linear separability;

[0070] Step S24: Optimal treatment window threshold extraction, used to extract the safe range of current parameters that can be executed by hardware from the classification space. Specifically, within the decision space generated by the stimulus response classification model, the overlapping part of the effective muscle contraction area and the user's pain-free area is identified, and the corresponding upper and lower limits of current intensity are calculated to obtain the optimal treatment window threshold.

[0071] The formula for calculating the optimal treatment window threshold is as follows:

[0072] ;

[0073] In the formula, T window It is the optimal treatment window threshold, I mineffective It is the minimum current intensity threshold that can induce effective contraction of the target muscle group, I maxcomfort It is the maximum current intensity threshold that maintains comfort within the user's subjective tolerance range;

[0074] Step S25: Calculate the electrical stimulation effectiveness evaluation index, which is used to quantitatively evaluate the efficiency of the current electrical stimulation program. Specifically, it involves establishing a comprehensive evaluation function based on muscle recruitment rate, pain threshold score, current intensity and fatigue characteristics, calculating the electrical stimulation effectiveness evaluation index, and using it together with the optimal treatment window threshold to form reference data for electrical stimulation effectiveness.

[0075] The formula for calculating the electrical stimulation effectiveness assessment index is as follows:

[0076] ;

[0077] In the formula, It is an electrical stimulation effectiveness assessment index used to comprehensively quantify the efficacy and output ratio of the current electrical stimulation program. VAS is the user's pain threshold score, typically using a 0-point visual analog scale, where a higher value indicates stronger pain. current It is the currently applied current intensity, used to reflect the level of stimulus energy input, Ifat It is a fatigue index feature used to reflect the degree of accumulated muscle fatigue;

[0078] The reference data on the effectiveness of electrical stimulation specifically includes an electrical stimulation effectiveness assessment index and the corresponding optimal treatment window threshold.

[0079] By performing the above operations, this solution addresses the technical problems in existing electrostimulation effectiveness modeling methods, which rely solely on electromyographic amplitude or stimulation intensity thresholds as criteria, fail to adequately consider the modulating effect of the physiological cycle on pain threshold and stimulation response, and have a coarse identification of the boundary between "ineffective stimulation" and "overstimulation." These problems can easily lead to insufficient muscle recruitment or significant discomfort or even pain in practical applications. This solution creatively constructs a classification model based on an improved support vector machine with physiological cycle-sensitive weights. Through cross-validation, it identifies the parameter range in which muscles contract effectively and the user experiences no pain, thus modeling the effectiveness of electrostimulation. This achieves refined identification and quantitative expression of the optimal treatment window, ensuring that the selection of electrostimulation parameters is both physiologically adaptable and considers comfort and safety.

[0080] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the parameters are dynamically adjusted to optimize the output waveform in real time based on the evaluation results. Specifically, a fuzzy PID control algorithm with physiological tolerance improvement is adopted, using electrical stimulation effectiveness reference data as a feedback variable. When the index is lower than the effective threshold, the pulse width is increased smoothly rather than the current intensity is increased directly. When excessive muscle fatigue or an increase in pain threshold score is detected, the system automatically switches to a low-frequency analgesia mode or increases the pulse interval time, and always clamps the output intensity within the safety standard limit of Class II medical devices to obtain adaptive electrical stimulation execution parameters, including the following steps:

[0081] Step S31: Efficacy reference normalization, used to convert treatment efficacy into control variables, specifically by calculating the deviation and rate of change between the real-time efficacy index and the target value, to obtain the normalized control error characteristics;

[0082] The formula for calculating the deviation between the real-time effectiveness index and the target value is as follows:

[0083] ;

[0084] In the formula, e(t) is the control error at the current moment, used to reflect the deviation between the current electrical stimulation protocol and the target therapeutic effect, and serves as the input variable for the controller. It is the target electrical stimulation effectiveness index, a reference target value that is preset by the system or adaptively determined based on historical treatment data. It is the current moment's electrical stimulation effectiveness assessment index;

[0085] The formula for calculating the rate of change is:

[0086] ;

[0087] In the formula, ec(t) is the rate of change of the control error, used to reflect the trend of the change of the treatment efficacy deviation over time, and serves as an auxiliary input variable for the fuzzy PID controller. It is the derivative of the control error with respect to time, used to describe the rate of error change;

[0088] Step S32: Adaptive parameter tuning improves fuzzy control, which is used to dynamically adjust the system response speed according to individual differences. Specifically, it introduces a physiological sensitivity factor based on the treatment window width, and corrects the proportional, integral and derivative gain coefficients of the PID controller in real time through fuzzy logic rules to obtain the adaptive control gain matrix.

[0089] The formula for calculating the proportional, integral, and derivative gain coefficients of the PID controller in real time by introducing a physiological sensitivity factor based on the treatment window width and using fuzzy logic rules is as follows:

[0090] ;

[0091] ;

[0092] ;

[0093] In the formula, K p K is the proportional gain coefficient of the PID controller in the current control cycle. p0 It is the initial reference value for the proportional gain, which can be obtained through factory calibration, empirical tuning, or initial calibration for a specific user, and is used as the reference center for adaptive adjustment. It is a physiological sensitivity modulation factor used to scale the gain correction amount obtained from fuzzy inference as a whole, controlling the strength of adaptive adjustment. Preferably, it can be obtained by mapping the treatment window width or the current window width. It is the proportional gain correction amount derived from the fuzzy logic rule base based on the error and the rate of change of error; K i K is the integral gain coefficient of the PID controller in the current control cycle. i0 It is the baseline initial value of the integral gain, used to define the system's fundamental convergence capability in terms of steady-state error elimination. The integral gain correction is obtained from fuzzy logic rule base reasoning; K d K is the differential gain coefficient of the PID controller in the current control cycle. d0 It is the initial reference value of the differential gain, used to define the system's basic ability to suppress the trend of error changes. It is the differential gain correction amount obtained by reasoning from the fuzzy logic rule base;

[0094] More preferably, Table 1 is a reference example table of the fuzzy logic rule base. As shown in the table, e is the control error amount, ec is the error change rate, NB / NM / NS represent negative large, negative medium, and negative small respectively, corresponding to the state where the output is too strong and needs to be reduced; ZO is the zero value, corresponding to the current state being ideal and stable; PS / PM / PB represent positive small, positive medium, and positive large respectively, corresponding to the state where the output is insufficient and needs to be increased.

[0095] The fuzzy logic rule base (as shown in Table 1) is the core logical link for realizing adaptive adjustment of electrical stimulation parameters. It dynamically decides the correction amount of proportional, integral and differential gain by real-time monitoring of control error and error change rate.

[0096] The rule base is designed with full consideration of the stringent requirements for safety and comfort during pelvic floor muscle electrical stimulation. The specific logic is explained as follows:

[0097] First, regarding the situation where the control error is in the negative range (i.e., the current electrical stimulation efficacy exceeds the expected target): when the system detects that the effectiveness index is higher than the set threshold, it means that there may be a risk of overstimulation or a decrease in the user's subjective tolerance; at this time, the rule base adopts a "safety first" strategy. If the error shows a rapid expansion trend, the logic rule will assign a large negative correction to the integral gain to eliminate the cumulative effect and prevent overshoot; at the same time, the proportional gain is finely adjusted so that the system can quickly and smoothly reduce the current intensity or pulse width, ensuring that the stimulation energy returns to the safe window as soon as possible, avoiding causing pain to the user;

[0098] Secondly, for situations where the control error is in the positive range (i.e., the current electrical stimulation efficacy has not yet reached the target): when the system detects a low muscle recruitment rate or insufficient treatment intensity, the logic rules focus on "steady improvement in efficacy." At this time, the rule base will assign a positive correction to the proportional gain based on the magnitude of the error to improve the system's response speed and allow the energy output to quickly approach the target value. To prevent the parameters from rising too quickly and causing users to experience stinging or abrupt sensations, the rule base will adjust the differential gain in a damping manner while adjusting the proportional gain, playing a role in predicting and suppressing fluctuations, thereby achieving a linear and gentle increase in electrical stimulation energy.

[0099] Secondly, regarding the stable region where the control error approaches zero (i.e., currently within the ideal treatment window): when both the control error and its rate of change are in a very small range, the system determines that it is currently in the optimal treatment state; at this time, the output of the rule base tends to be stable, and all gain correction values ​​are set to zero or fine-tuning values. This logic aims to maintain the currently effective electrical stimulation waveform and avoid frequent parameter adjustments by the system due to minor physiological noise interference, thereby ensuring the continuity of the treatment process and the stability of the user's somatosensory experience;

[0100] Finally, combined with the global modulation of the physiological sensitivity factor: the various correction values ​​output by the above rule base will be weighted twice by the physiological sensitivity factor; this mechanism ensures that even when fuzzy rules trigger large adjustment values, the system can still globally clamp the adjustment range according to the sensitivity of the user's current physiological cycle (such as high sensitivity during menstruation or low sensitivity during the follicular phase) and the actual width of the treatment window, thus achieving a deep coupling between "AI logical reasoning" and "actual biological characteristics".

[0101] Table 1. Reference Examples of Fuzzy Logic Rule Base

[0102]

[0103] Step S33: Energy balance-based dual-loop decoupling optimization, used to improve stimulation comfort while ensuring effectiveness. Specifically, the total output of the PID controller is preferentially allocated to the pulse width parameter, and the current intensity compensation adjustment is initiated after the pulse width reaches a preset safety threshold, resulting in a smoothly evolving waveform control command. This includes the following sub-steps:

[0104] Step S331: Charge increment mapping, used to determine the scale of energy change required for treatment, specifically by linearly mapping the PID control output to the target charge increment to obtain the total energy regulation requirement;

[0105] The formula for calculating the target charge increment is:

[0106] ;

[0107] In the formula, It is the target charge increment, used to characterize the amount of stimulus energy change required to achieve the desired therapeutic efficacy, k map is the mapping coefficient, used to convert the controller output into a charge increment scale, reflecting the proportional relationship between the control signal and the stimulus energy. u(t) is the total amplitude of the control output of the PID controller at the current moment.

[0108] Step S332: Comfort-based pulse width priority adjustment is used to improve stimulation efficacy without increasing current density. Specifically, while keeping the current intensity constant, the pulse width adjustment amount required to achieve the target charge increment is calculated, and it is determined whether it is within the preset pulse width safety upper limit threshold to obtain the first stage pulse width adjustment parameters.

[0109] The formula for calculating the pulse width adjustment is:

[0110] ;

[0111] In the formula, Under the condition of keeping the current intensity constant, I is the theoretical pulse width increment required to achieve the target charge increment. curr It is the current intensity value of the current being applied during the current stimulation.

[0112] The formula for calculating the pulse width adjustment parameter in the first stage is:

[0113] ;

[0114] In the formula, W stage1 This is the pulse width parameter value after the first stage of pulse width adjustment, W. curr This is the pulse width parameter value of the current electrical stimulation, W. max It is a preset safe upper limit threshold for pulse width, used to prevent discomfort or safety risks caused by excessive stimulation time;

[0115] Step S333: Current follow-up compensation after pulse width saturation is used to make up for the performance after the pulse width reaches the upper limit of adjustment. Specifically, it calculates the charge gap that cannot be met in the first stage and converts it into incremental compensation of current intensity to obtain the current intensity parameters in the second stage.

[0116] The formula for calculating the charge gap that cannot be satisfied in the first stage is:

[0117] ;

[0118] In the formula, It is the remaining charge gap that is not satisfied even after the pulse width reaches the upper limit of adjustment;

[0119] The formula for calculating the incremental compensation of the current intensity is:

[0120] ;

[0121] In the formula, This is the increase in current intensity required to fill the remaining charge gap;

[0122] The formula for calculating the current intensity parameter in the second stage is:

[0123] ;

[0124] In the formula, I next It is the current intensity parameter value at the next moment after the compensation adjustment is completed;

[0125] Step S334: Cross-dimensional energy gradient smoothing control, used to eliminate the perceptible step during parameter switching. Specifically, an energy transition factor is introduced to nonlinearly weight the pulse width increment and current increment, resulting in a smoothly evolving waveform control command. The calculation formula is as follows:

[0126] ;

[0127] In the formula, W(t) is the smoothed real-time pulse width control command value, and I(t) is the smoothed real-time current intensity control command value. It is an energy transition factor, specifically a logistic smoothing factor based on an error limit, calculated using the following formula: Where Wcurr-Wmax is the distance between the current pulse width parameter and the safety upper limit, used to reflect whether the pulse width adjustment is close to saturation;

[0128] Step S34: Fatigue monitoring and control, used to prevent muscle overexertion and relieve discomfort, specifically monitors the fatigue index and pain threshold score. When the value exceeds the limit, the stimulation frequency is automatically reduced and the pulse interval time is extended to obtain the analgesia and recovery protection mode parameters.

[0129] Step S35: Safety clamping parameter output, used to ensure that the output complies with the safety limits of medical devices. Specifically, the waveform control command is compared and truncated with the preset Class II medical device safety standard limits to obtain the final adaptive electrical stimulation execution parameters.

[0130] Preferably, the calculation formula for the comparison truncation is:

[0131] ;

[0132] In the formula, Output is the final current control parameter value output to the electrical stimulation execution module, and I... new The candidate current parameter value I is obtained through step S33. safetylimit It is the upper limit of current intensity that complies with the safety specifications for Class II medical devices, T window It is the optimal treatment window threshold range calculated in step S24;

[0133] The adaptive electrical stimulation execution parameters include frequency, pulse width, and current intensity.

[0134] By performing the above operations, this solution addresses the technical problems of existing dynamic parameter adjustment methods, which rely directly on linear adjustment of current amplitude, ignore charge distribution and individual tolerance differences, easily produce strong stimulation step sensations and large fluctuations in comfort, and are difficult to smoothly transition in time when fatigue or pain increases. This solution creatively adopts a fuzzy PID control algorithm that integrates physiological tolerance improvement, combines the electrical stimulation effectiveness evaluation index as a feedback variable, and introduces a physiological sensitivity factor based on the treatment window width and a dual-loop decoupled adjustment mechanism for energy. It prioritizes energy compensation through pulse width adjustment, and then performs current follow-up compensation after pulse width saturation. At the same time, it is combined with energy gradient smoothing and safety clamping control to achieve smooth evolution and comfortable output of electrical stimulation parameters while ensuring efficacy. This effectively reduces the risk of discomfort and improves the safety and clinical usability of dynamic adjustment.

[0135] Example 5, see Figure 1 , Figure 2 This embodiment is based on the above embodiment. In step S4, the individualized electrical stimulation data modeling is used to establish an individualized historical response benchmark and rehabilitation trend data model. Specifically, the physiological cycle collected in step S1, the reference data on the effectiveness of electrical stimulation determined in step S2, and the finally stable adaptive electrical stimulation execution parameters in step S3 are stored in the local database. The weighted moving average algorithm or linear regression trend analysis is used to calculate the average parameter preference of the user under different physiological cycle phases. Before the next treatment is started, the system automatically retrieves and matches the closest historical optimal parameter combination from the local database according to the physiological cycle to which the current date belongs as the initial preset value, so as to obtain the user's individualized dynamic initial electrical stimulation parameter library and pelvic floor muscle rehabilitation trend evaluation curve.

[0136] Example 6, see Figure 1 and Figure 2 Based on the above embodiments, this embodiment provides an AI data modeling system for individualized electrical stimulation parameters, including a data acquisition module, an intelligent modeling module, and a data iteration module.

[0137] The data acquisition module is used for data acquisition, obtaining a multimodal state feature set through data acquisition, and sending the multimodal state feature set to the intelligent modeling module and the data iteration module;

[0138] The intelligent modeling module is used for electrical stimulation effectiveness modeling and parameter dynamic adjustment. Through electrical stimulation effectiveness modeling and parameter dynamic adjustment, it obtains electrical stimulation effectiveness reference data and adaptive electrical stimulation execution parameters, and sends the electrical stimulation effectiveness reference data and adaptive electrical stimulation execution parameters to the data iteration module.

[0139] The data iteration module is used for individualized electrical stimulation data modeling. Through individualized electrical stimulation data modeling, an individualized dynamic initial electrical stimulation parameter library and a pelvic floor muscle rehabilitation trend assessment curve are obtained for the user.

[0140] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0141] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0142] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for AI data modeling of individualized electrical stimulation parameters, characterized in that: The method includes the following steps: Step S1: Data acquisition, collecting initial multidimensional data including physiological characteristics, physical environment and subjective perception, and processing it to obtain a multimodal state feature set; Step S2: Modeling the effectiveness of electrical stimulation. Construct a classification model based on an improved support vector machine with physiological cycle sensitive weights, and through cross-validation, find the parameter range in which muscles produce effective contraction and the user feels no pain, to obtain reference data on the effectiveness of electrical stimulation. Step S3: Parameter dynamic adjustment. A fuzzy PID control algorithm with physiological tolerance improvement is adopted. The reference data of electrical stimulation effectiveness is used as a feedback variable. When the index is lower than the effective threshold, the pulse width is increased smoothly rather than the current intensity is increased directly. When the muscle fatigue level exceeds the standard or the pain threshold score rises, the system automatically switches to low-frequency analgesia mode or increases the pulse interval time, and always clamps the output intensity within the safety standard limit of Class II medical devices to obtain adaptive electrical stimulation execution parameters. Step S4: Individualized electrical stimulation data modeling, linking and storing historical response data with physiological cycles, and establishing an individualized dynamic initial electrical stimulation parameter library.

2. The method for individualized electrical stimulation parameter AI data modeling according to claim 1, characterized in that: In step S1, a multimodal state feature set is obtained by performing timestamp alignment and noise cleaning on the initial multidimensional data. The multimodal state feature set includes electromyographic feature values, tissue conductivity, physiological cycle weights, and subjective tolerance parameters.

3. The method for individualized electrical stimulation parameter AI data modeling according to claim 2, characterized in that: In step S2, specifically, the ratio of the peak-to-peak value of the electromyographic signal induced by electrical stimulation to the maximum voluntary contraction value is calculated to obtain the muscle recruitment rate, the rate of change of the median frequency of the surface electromyographic signal is extracted to obtain the fatigue index feature, and the electrode contact state is determined by combining the real-time measured loop impedance value to obtain the standardized feature input set.

4. The method for individualized electrical stimulation parameter AI data modeling according to claim 3, characterized in that: In step S2, the improved support vector machine classification model based on physiological cycle sensitive weights specifically introduces biological variation parameters to correct the model's perception accuracy of pain threshold at different physiological stages. It establishes corresponding weight factors according to the user's physiological cycle phase and embeds them into the radial basis kernel function to construct a physiological cycle sensitive kernel function, thereby obtaining a kernel feature mapping space with biological adaptive capabilities.

5. The method for individualized electrical stimulation parameter AI data modeling according to claim 4, characterized in that: In step S2, the improved support vector machine classification model based on physiological cycle sensitive weights adopts a cost-sensitive learning mechanism to assign higher penalty weights to misclassifications of pain states in classification errors. The physiological cycle sensitive kernel function is used to perform nonlinear classification training on the quantified features, and the model hyperparameters are optimized through cross-validation to obtain the stimulus response classification model.

6. The method for individualized electrical stimulation parameter AI data modeling according to claim 5, characterized in that: In step S2, the reference data for the effectiveness of electrical stimulation specifically includes the electrical stimulation effectiveness assessment index and the corresponding optimal treatment window threshold.

7. The method for individualized electrical stimulation parameter AI data modeling according to claim 6, characterized in that: In step S3, the parameter is dynamically adjusted, including the following steps: Step S31: Efficacy reference normalization, used to convert treatment efficacy into control variables, specifically by calculating the deviation and rate of change between the real-time efficacy index and the target value, to obtain the normalized control error characteristics; Step S32: Adaptive parameter tuning improves fuzzy control, which is used to dynamically adjust the system response speed according to individual differences. Specifically, it introduces a physiological sensitivity factor based on the treatment window width, and corrects the proportional, integral and derivative gain coefficients of the PID controller in real time through fuzzy logic rules to obtain the adaptive control gain matrix. Step S33: Based on energy balance, dual-loop decoupling optimization is used to improve stimulation comfort while ensuring effectiveness. Specifically, the total output of the PID controller is preferentially allocated to the pulse width parameter, and the current intensity compensation adjustment is started after the pulse width reaches the preset safety threshold to obtain a smoothly evolving waveform control command. Step S34: Fatigue monitoring and control, used to prevent muscle overexertion and relieve discomfort, specifically monitors the fatigue index and pain threshold score. When the value exceeds the limit, the stimulation frequency is automatically reduced and the pulse interval time is extended to obtain the analgesia and recovery protection mode parameters. Step S35: Safety clamping parameter output, used to ensure that the output complies with the safety limits of medical devices. Specifically, the waveform control command is compared and truncated with the preset Class II medical device safety standard limits to obtain the final adaptive electrical stimulation execution parameters. The adaptive electrical stimulation execution parameters include frequency, pulse width, and current intensity.

8. A personalized electrical stimulation parameter AI data modeling system, used to implement the personalized electrical stimulation parameter AI data modeling method as described in any one of claims 1-7, characterized in that: It includes a data acquisition module, an intelligent modeling module, and a data iteration module.

9. The personalized electrical stimulation parameter AI data modeling system according to claim 8, characterized in that: The data acquisition module is used for data acquisition, obtaining a multimodal state feature set through data acquisition, and sending the multimodal state feature set to the intelligent modeling module and the data iteration module; The intelligent modeling module is used for electrical stimulation effectiveness modeling and parameter dynamic adjustment. Through electrical stimulation effectiveness modeling and parameter dynamic adjustment, it obtains electrical stimulation effectiveness reference data and adaptive electrical stimulation execution parameters, and sends the electrical stimulation effectiveness reference data and adaptive electrical stimulation execution parameters to the data iteration module. The data iteration module is used for individualized electrical stimulation data modeling. Through individualized electrical stimulation data modeling, an individualized dynamic initial electrical stimulation parameter library and a pelvic floor muscle rehabilitation trend assessment curve are obtained for the user.

Images