Nerve regulation and control method and system combined with continuous blood glucose monitoring

Abstract

The invention belongs to the field of electrical stimulation treatment devices, and provides a nerve regulation and control method and system combined with continuous blood glucose monitoring, and the method comprises the steps: constructing a reference state transition probability matrix between a blood glucose level and a motion state at a cloud; acquiring blood sugar time sequence data and kinematics data of the patient in real time, and comparing the data in the latest preset time window with the reference state transition probability matrix; according to a comparison result, carrying out individualized correction on the reference state transition probability matrix; inputting the individualized state transition probability matrix into a cloud electrical stimulation signal generation model, and outputting a corresponding electrical stimulation parameter sequence; and applying the electrical stimulation parameter sequence to a preset part of the patient through an electrical stimulation device worn by the patient.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of electrical stimulation therapy devices, and in particular relates to a neuromodulation method and system combined with continuous glucose monitoring. BACKGROUND

[0002] From the existing medical and engineering research, neuromodulation means can indirectly affect blood glucose levels by acting on the autonomic nervous system, but it cannot achieve direct, stable and controllable blood glucose regulation at present. Existing literature shows that vagus nerve stimulation, sympathetic nerve regulation and electrical stimulation of certain nuclei of the hypothalamus can affect insulin secretion, hepatic glucose output or muscle glucose uptake, thereby causing changes in blood glucose levels. However, such effects usually exhibit indirect effects, and individual differences are large and the mechanism is complex, and a mature technical system for directly maintaining blood glucose homeostasis mainly by neuromodulation has not yet been formed in clinical application. SUMMARY

[0003] In order to solve the problems in the prior art, the present application provides a neuromodulation method combined with continuous glucose monitoring, comprising the following steps: constructing a baseline state transition probability matrix between blood glucose levels and exercise states based on historical data of multiple users in the cloud, the baseline state transition probability matrix being used to represent the transition relationship between blood glucose fluctuation intervals and exercise performance; acquiring blood glucose time series data and kinematic data of the patient in real time during the process of the patient wearing a continuous glucose monitoring sensor and an exercise state acquisition device, and comparing the data in a recent preset time window with the baseline state transition probability matrix; according to the comparison result, individualizing the baseline state transition probability matrix by a Bayesian updating method to obtain an individualized state transition probability matrix of the patient; inputting the individualized state transition probability matrix into a cloud electrical stimulation signal generation model, the electrical stimulation signal generation model outputting a corresponding electrical stimulation parameter sequence according to the individualized state transition probability matrix; applying the electrical stimulation parameter sequence to a preset part of the patient through an electrical stimulation device worn by the patient.

[0004] Further, the neuromodulation method combined with continuous glucose monitoring, characterized in that the baseline state transition probability matrix is constructed by the following steps: detecting inflection points of the blood glucose time series data of multiple users, determining trend turning points based on the sign change of the first order rate and combining the amplitude of the second order rate, and dividing the blood glucose curve into several metabolic time slices with the trend turning points as boundaries; cutting the corresponding kinematic time series under the boundary conditions of the metabolic time slices, so that the blood glucose segments are strictly aligned with the motion segments; extracting multi-dimensional features for each motion segment and performing normalization processing; mapping the multi-dimensional features to a distribution type motion state vector, each dimension of the distribution vector representing the probability of the segment belonging to different motion sub-states; annotating the blood glucose trend of adjacent metabolic time slices and pairing it with the change of the motion state distribution vector to form a cross-modal transition pair; calculating the coupling degree index of the cross-modal transition pair, the coupling degree being determined by the cross information amount and the overlap degree between the distributions; performing normalization processing on the obtained matrix to obtain a baseline state transition probability matrix composed of row vectors constituting probability distributions.

[0005] Further, the baseline state transition probability matrix is stored in a double-layer structure, wherein the group average layer is used to represent the general transition rule, and the dynamic event layer is used to retain the high volatility characteristics appearing in a small number of users.

[0006] Further, the individualized correction step includes: comparing the blood glucose trend and motion state distribution of the patient within a preset time window with the baseline state transition probability matrix item by item to obtain a difference quantification result; determining the correction amplitude according to the frequency of occurrence of the transition mode in the current time window; strengthening or weakening the correction value according to the correlation degree of the blood glucose change rate and the motion state fluctuation rate; gradually reducing the influence of data outside the time window in the correction process to generate an individualized state transition probability matrix.

[0007] Further, the step of outputting the corresponding electrical stimulation parameter sequence includes: extracting transition features related to the blood glucose interval and fluctuation features related to the motion state distribution, and taking the distribution vector as the input feature; increasing the priority of the electrical stimulation mode corresponding to a certain transition path when the probability of the path significantly increases; when multiple transition paths have high probability at the same time, performing weighted fusion according to the path probability to form a combined parameter sequence; The signal generation model includes a basic layer and a dynamic adjustment layer, the basic layer is used to output standardized pulse frequency, pulse width and amplitude parameters according to the stable part of the matrix, and the dynamic adjustment layer is used to generate additional correction signals for high-risk transition paths in the matrix, and after the electrical stimulation parameters are issued to the electrical stimulation device worn by the patient, the subsequent output is dynamically adjusted through real-time feedback.

[0008] The application also provides a neuromodulation system combined with continuous glucose monitoring, comprising the following modules: a cloud modeling module, configured to construct a baseline state transition probability matrix between blood glucose level and exercise state based on historical data of multiple users, the baseline state transition probability matrix being used to represent the transition relationship between blood glucose fluctuation interval and exercise performance; a data acquisition module, configured to acquire blood glucose time series data and kinematic data of a patient in real time during the process of the patient wearing a continuous glucose monitoring sensor and an exercise state acquisition device, and compare the data within a recent preset time window with the baseline state transition probability matrix; an individualized correction module, configured to correct the baseline state transition probability matrix through a Bayesian updating method according to the comparison result, so as to obtain an individualized state transition probability matrix of the patient; a signal generation module, configured to input the individualized state transition probability matrix into a cloud electrical stimulation signal generation model, and output a corresponding electrical stimulation parameter sequence according to the matrix; an execution device, an electrical stimulation device worn by the patient, configured to apply the electrical stimulation parameter sequence to a preset part of the patient.

[0009] Further, the cloud modeling module further comprises: a turning point detection unit, configured to detect turning points of blood glucose time series data of multiple users, determine trend turning points based on the sign change of first-order change rate and the amplitude of second-order change rate; a time slice division unit, configured to divide a blood glucose curve into a plurality of metabolic time slices with the trend turning points as boundaries, and cut and align corresponding kinematic time series data under the boundary conditions of the metabolic time slices; a feature processing unit, configured to extract multi-dimensional features of each exercise segment and perform normalization processing; a distribution modeling unit, configured to map the multi-dimensional features into a distribution type exercise state vector, each dimension of the distribution type vector representing the probability that the segment belongs to different exercise sub-states; a coupling calculation unit, configured to pair the blood glucose trend of adjacent metabolic time slices with the exercise state distribution vector to form a cross-modal transition pair, and calculate a coupling degree index determined by cross information and overlap degree; a matrix generation unit, configured to normalize the obtained matrix to obtain a baseline state transition probability matrix with row vectors constituting a probability distribution.

[0010] Further, the baseline state transition probability matrix constructed by the cloud modeling module is stored in a double-layer structure, wherein the group average layer is used to represent the general transition rule, and the dynamic event layer is used to retain the high fluctuation characteristics appearing in a small number of users.

[0011] Further, the individualized correction module comprises: a comparison unit for comparing the blood glucose change trend and the motion state distribution of the patient within a preset time window with the baseline state transition probability matrix item by item to obtain a difference quantization result; a dynamic weight unit for determining a correction amplitude according to the occurrence frequency of the transition mode within the current time window; a coupling adjustment unit for enhancing or weakening the correction value according to the correlation degree of the blood glucose change rate and the motion state fluctuation rate; a time decay unit for gradually reducing the influence of data outside the time window in the correction process, thereby generating an individualized state transition probability matrix.

[0012] Further, the signal generation module comprises: a feature extraction unit for extracting transition features related to the blood glucose interval and fluctuation features related to the motion state distribution, and taking a distribution vector as an input feature; a scheduling unit for increasing the priority of the electrical stimulation mode corresponding to a certain transition path when the probability of the path is significantly increased, and for weighting and fusing according to the path probability when multiple transition paths have high probabilities at the same time to form a combined parameter sequence; a double-layer generation unit, including a basic layer and a dynamic adjustment layer, the basic layer is used for outputting standardized pulse frequency, pulse width and amplitude parameters according to the stable part of the matrix, and the dynamic adjustment layer is used for generating additional correction signals for the high-risk transition path in the matrix, and dynamically adjusting the subsequent output after the electrical stimulation parameters are issued to the electrical stimulation device worn by the patient through real-time feedback.

[0013] The present application constructs a baseline state transition probability matrix between blood glucose level and motion state based on multi-user data in the cloud, and combines the real-time blood glucose and motion data of the patient for individualized correction, so that the generated intervention strategy can not only reflect the general law of the group, but also dynamically adapt to the recent characteristics of the patient itself, thereby significantly improving the individualization and accuracy of neural regulation. Compared with the traditional fixed parameter or single signal driven stimulation mode, the present application can establish a dynamic correlation between blood glucose fluctuation and motion function, and realize a stimulation output that is more in line with the actual needs of the patient.

[0014] The electrical stimulation signal generation model of the present application is based on a probability driven mechanism, which can flexibly respond to high-risk transition paths and output dynamically adjusted electrical stimulation parameter sequences. Through the design of a double-layer structure, the model can not only ensure long-term safe and stable stimulation, but also insert correction signals in emergency situations to prevent falls or metabolic complications. The combination of stability and flexibility effectively improves the safety and robustness of clinical intervention.

[0015] In addition, the present application establishes a cross-modal closed-loop feedback mechanism, which collects blood glucose and exercise data in real time after electrical stimulation, and automatically optimizes the stimulation parameters according to the exercise improvement effect and blood glucose fluctuation trend. This feedback loop ensures that the intervention process is not limited to the improvement of a single function, but also achieves a dynamic balance between exercise and metabolism, thereby providing more comprehensive health management and long-term efficacy protection for patients. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is a flowchart of the method of the present application; Figure 2 is a method for constructing a baseline state transition probability matrix of the present application; Figure 3 is a method for individual correction of the present application; Figure 4 is a method for outputting corresponding electrical stimulation parameter sequences of the present application; Figure 5 is a module structure diagram of the system of the present application. DETAILED DESCRIPTION

[0018] The preferred description of the present application will be made below in combination with the drawings and specific embodiments.

[0019] Embodiment one, the present embodiment provides a neural regulation method combined with continuous blood glucose monitoring, the overall flowchart is as shown in Figure 1 , including the following steps: Construction of baseline state transition probability matrix.

[0020] In the cloud server, based on the historical big data of multiple users, a baseline state transition probability matrix between blood glucose level and exercise state is established.

[0021] The data source can be the blood glucose curve collected by the continuous blood glucose monitoring system, and the exercise state data collected by the kinematics device (such as inertial measurement unit, accelerometer, electromyographic sensor or clinical exercise function score).

[0022] The blood glucose level can be divided into several intervals according to medical standards, such as hypoglycemia interval, normal interval, postprandial peak interval and hyperglycemia interval.

[0023] The motor state can be divided into stable state, mild disorder, moderate disorder, and severe disorder, etc.

[0024] On this basis, the transition of a large number of users in different blood glucose levels is counted to form a probability matrix. The matrix reflects the possibility of motor state change under different blood glucose levels, which is a group-level benchmark rule.

[0025] The matrix can be calculated by big data statistical methods such as sliding window statistics, frequency counting or hidden Markov modeling.

[0026] Individual data collection and comparison.

[0027] Patients wear continuous blood glucose monitoring sensors and motor state collection devices in daily life.

[0028] The blood glucose sensor can be a subcutaneous implantable or attached sensor with a sampling period of minutes, which can continuously record the dynamic changes of blood glucose.

[0029] The motor state collection device can be an inertial sensor worn on the wrist or waist, or an electromyographic electrode, which is used to collect motor amplitude, tremor frequency or gait stability.

[0030] The system collects the blood glucose and motor joint data of the patient within a predetermined time window (such as the past 24 hours or the past week).

[0031] The patient's recent data is compared with the benchmark probability matrix, and the comparison method can use similarity measurement, such as comparing the difference between the two transition distributions, to determine the deviation of the patient's actual state transition pattern from the benchmark pattern.

[0032] Individualized correction.

[0033] The cloud server corrects the benchmark state transition probability matrix according to the comparison result.

[0034] The correction method can use the Bayesian inference idea, that is, to take the benchmark probability as a reference and adjust it combined with the individual's recent data.

[0035] For example, if the individual is more likely to have moderate motor disorder in the high blood glucose state in the near future, the corrected matrix will increase the weight of this transition.

[0036] In order to maintain the rationality of the matrix, smoothing processing methods can be used in the correction process, such as weighted average or weighted voting based updating method, so that the result not only retains the statistical characteristics of big data, but also reflects the individual differences.

[0037] The corrected matrix is the individualized state transition probability matrix of the patient, which is more in line with the individual's metabolic and motor characteristics.

[0038] Electrical stimulation signal generation.

[0039] The individualized state transition probability matrix is input into an electrical stimulation signal generation model.

[0040] The electrical stimulation signal generation model can be implemented on the cloud, mapping different state transition probabilities to electrical stimulation parameters through a rule engine, a probability-driven control strategy, or a machine learning-based prediction method.

[0041] The electrical stimulation parameters can include pulse frequency, pulse width, stimulation current intensity, and waveform mode.

[0042] For example, when the matrix shows a high probability of transitioning from a normal blood glucose state to mild motor impairment, the system can select a lower frequency and medium intensity electrical stimulation mode; when the matrix shows a high probability of transitioning from a high blood glucose state to moderate motor impairment, the system can select a high frequency and wide pulse width stimulation mode.

[0043] The electrical stimulation parameters are generated in a time series and can be dynamically adjusted according to the patient's state.

[0044] Electrical stimulation device implementation.

[0045] The generated electrical stimulation parameter sequence is delivered to the patient's worn electrical stimulation device via a mobile application.

[0046] The mobile application communicates with the cloud server in both directions, receiving updated parameter sequences and uploading real-time patient data.

[0047] The electrical stimulation device includes a host and an electrode belt, with a pulse generation circuit, a power management module, and a wireless communication module built-in.

[0048] The electrical stimulation device can be a transcutaneous electrical stimulation device worn on the wrist, neck, waist, lower limbs, ears, or other body parts, and can be in the form of a wristband, neck ring, patch, or ear hanging. The device includes an electrode patch, a fixing structure, and a pulse generation module, which can apply the electrical stimulation parameter sequence to the radial nerve, median nerve, ulnar nerve, vagus nerve, sciatic nerve, femoral nerve, or auricular vagus nerve, etc. target nerve area, to achieve coordinated regulation of blood glucose level and motor state.

[0049] The device connects to the mobile terminal through Bluetooth or WiFi and executes the stimulation signal immediately after receiving the parameters.

[0050] The system has safety protection mechanisms, such as maximum output intensity limitation, automatic electrode contact detection, and abnormal alarm function, to ensure safe use.

[0051] Through the above steps, the embodiment realizes a neuromodulation method based on big data modeling and individualized correction. The electrical stimulation signal is driven by a probability model, which can not only suppress movement disorder symptoms, but also regulate metabolism according to the blood glucose fluctuation trend, thereby realizing a cross-system combined closed-loop treatment.

[0052] In this embodiment, a specific method for constructing a reference state transition probability matrix is given, as shown in Figure 2 The overall process includes the following steps: During the dynamic change of blood glucose level, there are often significant stage characteristics. The stage characteristics are that the metabolic trend is relatively stable within a certain time range, and there is a significant change inflection point at the stage transition. The metabolic characteristics in different stages can directly affect the physiological load and sports performance of the patient, so effective capture of these stage characteristics is a prerequisite for constructing the reference state transition probability matrix.

[0053] To achieve this goal, the embodiment first performs inflection point detection on the blood glucose time series data. The basic idea of inflection point detection is to identify the trend change of the blood glucose curve through the mathematical derivative information: In the blood glucose curve, the first-order difference calculation is performed on the continuous sampling points to obtain the time series sequence of the blood glucose change rate. When the sign of the first-order change rate changes, it means that the blood glucose curve changes from an upward trend to a downward trend, or from a downward trend to an upward trend, and the point is marked as a potential inflection point.

[0054] To avoid false inflection points caused by minor fluctuations, the embodiment further introduces the second-order change rate as a correction condition. When the absolute value of the second-order change rate exceeds the set threshold, the point is confirmed as a real trend turning point, indicating that the metabolic trend has significantly accelerated or slowed down at this position.

[0055] Based on the above method, the entire blood glucose time series curve is divided into several intervals, each interval is defined by the adjacent trend turning points before and after, and a series of metabolic time slices are formed. Each time slice represents a relatively stable stage in the metabolic sense, and the change direction and amplitude of blood glucose in the stage remain consistent, thereby more accurately describing the evolution process of the metabolic state.

[0056] After obtaining the metabolic time slices, the embodiment further strictly time-aligns the time series data of the motion state acquisition device with these segments. The motion state acquisition device includes but is not limited to an inertial measurement unit, a surface electromyography sensor, a gait analysis system, or a tremor monitoring device, and its data acquisition has a high sampling frequency.

[0057] To ensure the correspondence of cross-modal information, the embodiment realizes alignment through the following steps: Firstly, the motion data is divided according to the time boundary of the blood glucose segment to ensure that each segment of motion data corresponds to a metabolic time slice.

[0058] During the division process, if the motion data has missing points or the sampling rate is not synchronized with the blood glucose data, linear interpolation or piecewise time normalization method is used to fill in and resample to ensure the consistency of the two types of data on the time axis.

[0059] Through this strict time alignment processing, a cross-modal time slice pair composed of blood glucose segments and motion segments is finally formed, and each pair of time slices can be analyzed synchronously in the two dimensions of metabolic trend and motion performance.

[0060] Through the construction of the above cross-modal time slice pair, the time slice not only contains the boundary information of the blood glucose trend, but also retains the dynamic performance of the motion state in the metabolic phase. Since the change of metabolic level often has a direct driving or restraining effect on the motion function, this division and alignment method can ensure that blood glucose and motion data are compared in the same semantic unit, avoiding the misalignment problem of different modalities in the time granularity, thereby significantly enhancing the causal explanation in the cross-modal modeling process.

[0061] In other words, this method realizes the division of the metabolic phase at the mathematical level and ensures the alignment consistency of multi-modal data at the engineering implementation level through segmentation based on trend inflection points and cross-modal time alignment, thereby providing a complete and reliable input basis for subsequent state transition modeling and probability matrix construction.

[0062] Within each metabolic time slice, the patient's motion state often presents a complex performance of multi-dimension and multi-level. These performances not only manifest in the differences of macroscopic motion patterns, such as the rhythmicity of gait and the stability of stride, but also include the fluctuations of physiological parameters at the microscopic level, such as the electromyographic signal of muscle activity, the slight changes of joint angular velocity, the tremor amplitude of hands or lower limbs, and the motion coordination, etc. Therefore, if only a single label (such as normal or abnormal) is used to describe the motion state, a large amount of information will inevitably be lost, making it difficult to fully reflect the interaction between different characteristic dimensions. Based on this, the embodiment proposes a distributed state representation method, which enables the characterization of motion state to take into account multi-dimensionality, fuzziness and continuity.

[0063] In the process of implementation, first, the various motion characteristic dimensions are normalized. Different motion characteristics often have different dimensions and numerical ranges, for example, tremor amplitude is usually expressed in millimeters, joint angular velocity is expressed in radians per second, and electromyographic signal is expressed in millivolt potential fluctuations. If not normalized, the numerical value of a certain dimension may dominate during aggregation, resulting in state representation deviation. Therefore, the embodiment maps all features to a unified numerical interval through the normalization method, so that they have comparability and balance in subsequent calculations.

[0064] After normalization, the embodiment uses a fuzzy aggregation method based on probability density estimation to map multi-dimensional features. Specifically, by estimating the joint distribution of multi-dimensional motion characteristics, the probability density distribution in a certain feature space is obtained. In order to avoid the boundary fragmentation problem caused by hard classification, the embodiment introduces a fuzzy aggregation mechanism to map the feature vector of each motion segment into a distribution vector. Each dimension of the distribution vector represents the membership degree or occurrence probability of the segment in a certain motion sub-state, rather than a single deterministic label.

[0065] For example, when a motion segment shows both slight tremor and gait instability, the traditional method may classify it as one of tremor or gait abnormalities, while the embodiment generates a probability vector for it, where the tremor sub-state dimension may account for 0.6 and the gait abnormality sub-state dimension accounts for 0.4, thus more accurately expressing its overlap in multiple sub-states.

[0066] In the matrix representation layer, the motion state no longer enters the state transition matrix in the form of a single label, but is expressed in the form of a distribution type probability vector. The definition of state transition is transformed from state A to state B to a smooth transition from one probability distribution vector to another probability distribution vector. This modeling method can reflect the fuzziness and continuity of motion state in the matrix, thereby avoiding the information loss caused by discretization in traditional methods.

[0067] The core principle of this method is to use the ability of fuzzy aggregation to unify multi-dimensional heterogeneous features in the probability space. In this way, the motion state no longer depends on rigid classification boundaries, but exists in the form of continuous transition in the probability space. In this way, the state transition process not only captures the differences between different motion performances, but also describes the diversity and gradualness of motion performance. For example, when a patient gradually transitions from stable gait to mild instability, the state transition is no longer abrupt, but shows continuous change in the probability space, so that the system can more sensitively identify the early evolution of motion state.

[0068] The method has the advantages that multi-dimensional information in the motion state can be comprehensively retained, and a distributed probability vector is taken as a unified expression form, so that more abundant and more detailed input data are provided for subsequent reference state transition probability matrix construction. In this way, cross-modal modeling can better reflect the complex coupling relationship between blood glucose metabolism and the motion state, and the scientificity and interpretability of the modeling result are improved.

[0069] To establish the state transition relationship between blood glucose level and motion state, the embodiment proposes a dynamic coupling aggregation mechanism. The core idea of the mechanism is to correspond the probability distribution of metabolic trend and motion state across modalities, to establish the transition strength expression form between the two by introducing an information theory measurement method, and to aggregate on a cross-user level to form a transition matrix that can reflect individual differences and group commonality.

[0070] In the specific implementation process, first, the blood glucose level of adjacent time slices is analyzed. As a continuous time series signal, blood glucose level is often jointly affected by dietary intake, exercise intensity, and individual metabolic rate. In order to eliminate the interference of instantaneous fluctuations, the embodiment introduces a sliding window and trend extraction algorithm in the analysis process, and the blood glucose change pattern in each time slice is summarized into three categories: rising, falling and stable. Rising refers to a sustained upward trend in blood glucose values in the time slice sequence; falling corresponds to a sustained downward trend; and stable indicates that the fluctuation amplitude is small within a certain interval and does not significantly deviate from the reference value. This trend extraction step lays a clear metabolic semantic foundation for subsequent correspondence with the motion state.

[0071] After obtaining the blood glucose change trend, the embodiment corresponds it with the motion state distribution vector in the same time slice. The motion state distribution vector is a probability type representation generated based on fuzzy aggregation, and each dimension represents the membership degree of the motion segment in a certain sub-state. Therefore, in the correspondence process of the trend and the distribution vector, each blood glucose trend label will form a cross-modal transition pair with a group of motion state distribution changes. For example, when the blood glucose trend is falling, and the probability vector of the motion state changes from stable gait to a mixed performance of tremor and abnormal gait, they form an effective transition pair.

[0072] In the process of statistical cross-modal transition relationship, the embodiment introduces the coupling degree between distributions as a measure of transition strength. The definition of coupling degree is derived from the combination of information theory and probability statistics, and its calculation is based on two parts: one is the cross information, which is used to measure the degree of information overlap between two probability distributions; the second is the overlap degree, which is used to measure the size of the cross region of the distribution in the multi-dimensional feature space. By jointly calculating the cross information and the overlap degree, a numerical index can be obtained to reflect the extent to which the blood glucose change drives the change of the exercise state. The larger the index, the closer the relationship between the blood glucose trend and the exercise state transition, and the contribution degree in the subsequent matrix also increases.

[0073] In the aggregation process across users, the embodiment adopts a dynamic weighting strategy. Traditional aggregation methods often give the same weight to different individuals, which can easily lead to the averaging of significant difference events and make it difficult to highlight key trends. The improved method of the embodiment introduces metabolic trend amplitude and exercise state fluctuation synchronicity as weighting factors. The metabolic trend amplitude refers to the absolute change amplitude of blood glucose level between adjacent time slices, which is used to quantify the significance of the trend; the exercise state fluctuation synchronicity is measured by comparing the distribution change rates of different feature dimensions, which is used to represent the activity of the exercise state as a whole. When the metabolic trend amplitude is large and the exercise state distribution fluctuation is obvious, the transition relationship corresponding to this time slice will obtain a higher weight in the aggregation process, thereby enhancing its influence in the overall matrix.

[0074] Through the above steps, the final cross-modal transition matrix not only contains the corresponding relationship between blood glucose level and exercise state, but also highlights high correlation events through a dynamic weighting mechanism, so that the transition matrix has the cross-modal driving characteristic. The principle is to use information theory to measure the linkage strength between blood glucose and exercise state, and then retain key events through weighted aggregation, thereby realizing the dynamic coupling modeling between metabolic state and exercise performance.

[0075] For example, in an actual scenario, when a patient has a rapid increase in blood glucose after a meal, the exercise state distribution tends to transition from a stable gait to a mixed state of high-amplitude tremor and gait imbalance. In this case, the cross information calculation result shows that the distributions of the two have a high degree of overlap, the overlap degree is also at a significant level, so the coupling degree value is high. At the same time, since the blood glucose rise amplitude is large and the exercise state change shows strong synchronicity, the weight of this transition relationship in the overall matrix is significantly improved, thereby effectively enhancing the description of such metabolic-exercise linkage events.

[0076] Through the dynamic coupling aggregation mechanism, the final transition matrix can not only reflect the static corresponding relationship between blood glucose and exercise state, but also present the driving effect of the two in the time evolution process, making the expression of cross-modal relationship more detailed, flexible and with engineering realizability.

[0077] After dynamic aggregation is completed, the obtained transition matrix needs to be further normalized and structured. First, at the probability constraint level, this embodiment specifies that for each blood glucose level interval, the distribution type transition value of all possible exercise states corresponding to the interval is uniformly normalized, so that each row vector strictly meets the probability distribution requirement. In this way, it is ensured that the output result of any blood glucose state transition in the matrix is a legal probability combination, avoiding the occurrence of numerical overflow or invalid weight.

[0078] At the structure design level, the transition matrix is constructed as a double-layer structure to enhance flexibility and scalability in different user groups and individualized applications. The first layer is the group average layer, which reflects the common transition rules across users. This layer is aggregated from a large number of user historical blood glucose and exercise segment data, and can present overall trends with statistical stability. The second layer is the dynamic event layer, which retains high volatility characteristics that occur in a small number of user groups, such as rare events of rapid blood glucose decline accompanied by abnormal exercise imbalance. Through this design, the matrix can have universality and retain sensitivity to extreme scenarios.

[0079] In the storage and calling link of the cloud database, the double-layer matrix is saved in a structured manner. Each user can inherit the regularity knowledge deposited in the group average layer and use the dynamic event layer to provide additional reference for subsequent individualized correction when calling. In actual operation, the system dynamically adjusts the weight ratio of the group layer and the event layer according to the real-time data collected by the user, thereby achieving a balance between general patterns and individual characteristics.

[0080] In this way, the final transition matrix not only meets the mathematical constraints of probability distribution, but also has the adaptability and flexibility of a double-layer structure, providing a solid data foundation for individualized correction and real-time dynamic updating.

[0081] As shown in Figure 3 , this embodiment further provides a specific method for generating an individualized state transition probability matrix.

[0082] In this embodiment, in order to further reflect the individual differences of patients on the basis of group rules, an individualized state transition probability matrix generation method based on dynamic comparison and correction is proposed. The method includes the following steps: In the process of wearing continuous glucose monitoring sensor and motion state acquisition device by the patient, the system first establishes a data acquisition channel to record the blood glucose concentration change and motion state parameters in real time. The blood glucose monitoring sensor can continuously collect the glucose concentration in the patient's body fluid with a time resolution of minutes or less, and form a time-evolving blood glucose time series data sequence. At the same time, the motion state acquisition device (such as an accelerometer, gyroscope or heart rate monitoring component) captures the patient's kinematic information, including step frequency, acceleration amplitude, motion intensity level, activity duration and other indicators, thereby forming kinematic time series data corresponding to blood glucose data on the time axis.

[0083] The system takes the latest preset time window as the analysis interval, which can be several minutes, several hours or a daytime period, and the specific length can be flexibly set according to the clinical scene or individual characteristics. Within this time window, the system performs trend analysis on the blood glucose time series data to extract the direction and amplitude characteristics of the blood glucose change, such as determining whether it presents a continuous rise, gradual decline or basically stable metabolic trend in this interval. At the same time, the system also performs statistical modeling on the kinematic time series data to extract the distribution form of the motion state as a multi-dimensional vector, which is used to represent the frequency and proportion of different motion patterns (such as rest, light activity, moderate intensity exercise or high intensity exercise) in this time window.

[0084] After extracting the blood glucose trend and motion distribution, the system compares the cross-modal transfer pattern formed by the combination of the two with the reference state transfer probability matrix stored in the database. This reference matrix is pre-constructed based on large-scale population sample data and can reflect the universal transfer relationship between blood glucose change and motion state, i.e. the typical transfer probability of each motion state distribution under different blood glucose trends. Through comparison, the system can identify the difference between the observed transfer pattern of the current patient in this time window and the average rule of the population. For example, when the population reference matrix shows that blood glucose is likely to decrease after exercise, but the actual observed pattern of the patient shows that blood glucose does not decrease significantly or even slightly increases after exercise, it can be judged that the individual deviates from the population rule.

[0085] This item-by-item comparison not only reveals whether the patient follows the typical population rule, but also further locates the specific deviation link, including whether the deviation occurs in the direction of blood glucose trend or in the transfer intensity of motion state distribution. Through this comparison mechanism, the system realizes the fine characterization of the relationship between the individual and the population, providing data basis for subsequent difference measurement and individualized correction.

[0086] In the comparison process, the system not only statistically analyzes the actual transfer frequency of the patient within the preset time window, but also further introduces a cross-modal coordination measurement method. Specifically, the system calculates the blood glucose change rate and the motion state fluctuation rate respectively, and evaluates the coordination degree between the two on this basis. The blood glucose change rate is used to describe the change amplitude and direction of blood glucose concentration in unit time, which can represent the significance of blood glucose metabolism trend; the motion state fluctuation rate is obtained by standardizing the kinematic parameters (such as step frequency, acceleration or heart rate variability), which can reflect the switching degree and intensity difference between different motion modes. When the blood glucose change rate and the motion state fluctuation rate show a high consistency or synchronous fluctuation on the time axis, it is considered that the cross-modal coupling strength between the two is strong, indicating that the blood glucose change has a significant driving effect on the motion state distribution in the current time window.

[0087] In addition to calculating the coupling strength, the system also performs multi-dimensional feature recognition on the patient's transfer mode. First, the system identifies the frequency of the mode within the time window to determine whether it is an incidental event or a regular trend. Second, the system analyzes the fluctuation amplitude of the transfer mode in the numerical level, i.e., whether the change amplitude of blood glucose level is significant and whether the switching between different categories of motion state distribution is intense. Third, the system measures the duration of the transfer mode to reflect whether a certain state transfer is a short-term phenomenon or a long-term existence. Through the joint consideration of the three dimensions of frequency, amplitude and duration, the system can form a multi-angle description of the difference of the transfer mode.

[0088] Finally, these difference quantification results will be used as the core input factors for the subsequent correction process. Specifically, if the observed mode of the patient deviates significantly from the group regularity in terms of frequency, the system will reduce the weight of the corresponding path in the baseline state transfer matrix; if the observed mode shows abnormal enhancement or weakening in terms of amplitude or duration, the system will correct the numerical value of the corresponding transfer probability in the matrix through dynamic weighting. In this way, the corrected individualized transfer matrix can more truly reflect the individual metabolism regularity and motion state characteristics of the patient, providing reliable data support for subsequent risk assessment and personalized intervention.

[0089] On the basis of difference measurement, the system performs dynamic correction operation on the baseline state transfer probability matrix to generate an individualized state transfer matrix that is more consistent with the individual metabolism regularity of the patient. The correction process introduces a multi-level control mechanism to ensure that the correction result not only maintains the stability of the group regularity, but also reflects the individual difference characteristics exhibited by the patient in the near future. Specifically, the correction process includes dynamic weight mechanism, cross-modal coupling mechanism and time decay mechanism.

[0090] Firstly, the dynamic weight mechanism is used to solve the problem of the difference in the frequency of observation patterns. The system will count the number of occurrences of each transition pattern within a preset time window, and adjust the weight in the correction matrix according to its relative frequency. If a transition pattern is observed multiple times within the current time window, it means that this pattern has higher stability and representativeness in the patient's recent physiological state, so it is given a greater influence factor in the correction, so that it can occupy a more prominent position in the individualized matrix. Conversely, if a pattern occurs only occasionally at individual time, the system will correspondingly reduce its influence weight to avoid accidental data causing excessive deviation of the matrix structure. The introduction of this mechanism ensures that the correction result can adaptively balance the contradiction between long-term regularity and short-term contingency.

[0091] Secondly, the cross-modal coupling mechanism is used to highlight the interactive driving effect between blood glucose changes and exercise state. During the correction process, the system not only examines the transition characteristics in a single mode, but also identifies the coupling relationship between cross-modal by calculating the degree of cooperation between blood glucose rate of change and exercise state fluctuation rate. When two modalities show strong correlation within the time window, for example, rapid blood glucose rise and significant exercise state fluctuation occur simultaneously, it means that this cross-modal driving pattern plays an important role in the patient's current state. The system will accordingly increase the correction amplitude of this transition pattern, so that the individualized matrix can more sensitively reflect the inherent coupling characteristics between different modalities. Through this mechanism, the correction process not only reflects the regularity of a single mode, but also enhances the adaptability of the matrix to complex interactive relationships.

[0092] Thirdly, the time decay mechanism is used to deal with the influence of historical data on the current correction. Since the patient's metabolic state and exercise habits may change dynamically over time, if the influence of past data on the current correction process is too large, it may cause the matrix to lag behind the actual state. Therefore, the system introduces a time weighting factor in the correction process, which applies a decreasing influence to data that exceeds the preset time window. As the time interval increases, the weight of historical observation data in the correction process gradually decreases until it is completely disabled. This approach ensures that the correction process mainly relies on recent observations, avoiding excessive interference from long-term data, while still maintaining some historical continuity.

[0093] Through the synergistic effect of the dynamic weight mechanism, the cross-modal coupling mechanism and the time decay mechanism, the corrected individualized state transition probability matrix not only inherits the group common regularity reflected in the baseline matrix, but also fully absorbs the individual differences exhibited by the patient in the near future. The final generated matrix establishes a balanced relationship between the group average structure and the individual dynamic characteristics, which has both stability and flexibility, and can provide fine support for subsequent individual risk assessment and intervention strategies.

[0094] After the dynamic correction is completed, the system generates an individualized state transition probability matrix for the current patient. This matrix continues the group average layer design of the benchmark matrix in the overall structure, retains the common rules between blood glucose state and exercise state revealed by large-scale sample data, thereby ensuring the stability and interpretability of the results. At the same time, the matrix introduces adaptive adjustment to the patient's recent observation data in the local structure, reflecting the unique dynamic characteristics exhibited by the patient in a specific time period, so that the matrix has distinct individual differences.

[0095] The individualized state transition probability matrix not only embodies the individual differences in blood glucose level changes and exercise state distribution in a single modal dimension, but also enhances the sensitivity to the interaction between multiple modalities through a cross-modal coupling correction mechanism. For example, the system can accurately depict the transition relationship between blood glucose rise and exercise intensity increase in the matrix, and highlight the key patterns that frequently occur recently through mechanisms such as dynamic weights and time decay. This makes the matrix have dual sensitivity: on the one hand, it is sensitive to the time dimension and can quickly respond to recent changes in patient state; on the other hand, it is sensitive to the cross-modal relationship and can capture the complex driving action between blood glucose and exercise.

[0096] At the storage level, the corrected individualized matrix will be encrypted and stored in the cloud database. This database supports hierarchical indexing and dynamic calling mechanism, which can iteratively update and track the matrix of the same patient during subsequent data collection. At the same time, each corrected matrix will form a versioned record, which is convenient for tracing the long-term evolution trajectory of the patient's state, thereby providing data support for medical intervention.

[0097] At the application level, the individualized state transition probability matrix will serve as the core reference for subsequent system calls, used in various functional scenarios. First, it can provide input basis for individualized risk assessment, for example, in predicting the risk of hypoglycemia or hyperglycemia, the size of the corresponding transition probability in the matrix will directly affect the risk score value. Second, it can provide dynamic basis for state prediction, for example, before the patient enters a certain blood glucose state, the system can calculate the probability distribution of different transition paths based on the matrix, thereby giving early trend warning. In this way, the matrix not only serves as a model basis in data analysis, but also plays a core role in prediction and decision-making.

[0098] Finally, as a dynamic data structure with both population universality and individual differences, the matrix realizes the double-layer mapping from macroscopic laws to microscopic individuals, ensuring the coexistence of universality and individualization in the clinical application of the system, and can long-term support for continuous monitoring, dynamic intervention and individual health management of patients.

[0099] Embodiment four, as Figure 4As shown, the present embodiment mainly gives a specific method of generating electrical stimulation signal based on individualized state transition probability matrix.

[0100] In the present embodiment, the system first inputs the individualized state transition probability matrix of the patient into the cloud-based electrical stimulation signal generation model. This matrix, after the aforementioned comparison and correction, can accurately represent the probability distribution of the patient's transition to multiple motion states under different blood glucose levels. The input not only includes the probability of a single transition path, but also covers the distribution information of the entire blood glucose-motion state space.

[0101] During data analysis, the model performs hierarchical feature extraction on the matrix. First, the model identifies the transition patterns corresponding to different blood glucose intervals, for example, in the blood glucose decline interval, there may be multiple transition paths such as stable gait, slight fluctuation, and moderate instability, and these paths are represented by different probability values in the matrix. Second, the model analyzes the dynamic characteristics of the motion state distribution, not only focusing on the proportion of a single category, but also evaluating the probability difference and fluctuation trend between categories, to judge the stability and diversity of the patient's motion performance under the current metabolic conditions.

[0102] Unlike traditional methods that rely on single label mapping, the electrical stimulation signal generation model of the present embodiment introduces a complete distribution vector as input features. The so-called distribution vector refers to the coexistence of various possible motion sub-states in the form of probability under a certain blood glucose state, rather than being rigidly divided into a unique label. For example, when the patient's blood glucose is declining, the system may observe that the probability of stable gait is 0.30, the probability of slight fluctuation is 0.34, the probability of moderate instability is 0.30, and the probability of significant abnormality is 0.06. If the traditional method only relies on the maximum value (i.e., slight fluctuation) for label mapping, it may ignore the fact that moderate instability also has a high probability, resulting in incomplete or robust stimulation strategies.

[0103] By inputting the complete distribution vector into the model, the present embodiment can cover the uncertainty and diversity of state transition in the electrical stimulation parameter generation process. For example, when both slight fluctuation and moderate instability show high probability, the model will generate a parameter sequence that combines the regulation effects of both modes, maintaining a low amplitude background stimulation to suppress slight fluctuation, and inserting short enhanced pulses when necessary to combat moderate instability. This strategy based on distribution characteristics can effectively avoid the under-regulation or over-regulation problems caused by single-label methods.

[0104] In the generation of electrical stimulation parameters, the cloud model adopts a probability-driven scheduling mechanism. The core idea of this mechanism is that the configuration of electrical stimulation signals should not be determined by a single dominant state, but should fully consider the joint distribution characteristics of multiple transition paths in the individualized state transition probability matrix, so that the output stimulation parameters can truly reflect the uncertainty and complexity of the patient's state.

[0105] First, the system will scan the individualized state transition probability matrix in real time, identify all possible motion state transition paths at the current blood glucose level, and sort their probability values. When the probability of a certain transition path significantly increases, for example, the probability of blood glucose decline → moderate instability increases from 0.15 of the group benchmark to 0.30 of the individual, the system will automatically increase the priority of the electrical stimulation mode matching this path. Priority increase means that during parameter sequence generation, the time proportion, stimulation energy allocation or pulse structure weight of this mode will significantly increase, ensuring that the model can quickly and effectively intervene in high-risk states.

[0106] Second, when the system detects that multiple transition paths have high probabilities at the same time, the model will not simply choose the one with the highest probability as the only reference, but will start a weighted fusion mechanism. Specifically, the system will assign weights according to the probability size of different paths and fuse the corresponding electrical stimulation modes in the parameter sequence in a superimposed manner. For example, when the probability of blood glucose decline → slight fluctuation is 0.34, and the probability of blood glucose decline → moderate instability is 0.30, the system will allocate a combination of two types of stimulation modes in the output parameter sequence: on the one hand, maintain low-frequency, low-amplitude background stimulation to alleviate slight fluctuations; on the other hand, periodically insert higher frequency or longer pulse width stimulation pulses to deal with possible moderate instability. Through this weighted fusion, the generated signal can achieve balanced regulation between different risk states, rather than unilaterally strengthening a single mode.

[0107] At the signal level, the probability-driven scheduling mechanism is also reflected in the dynamic adjustment of stimulation parameters. Unlike the single signal of fixed frequency + fixed amplitude in traditional methods, the stimulation parameter sequence generated by this mechanism will adjust with the real-time changes in transition probabilities. For example, when the probability of blood glucose decline → moderate instability rapidly increases in a short period of time, the system will temporarily insert a high-frequency low-amplitude pulse into the original basic stimulation sequence to provide additional neural regulation support; when this probability gradually decreases in the subsequent window, the system will automatically remove the additional pulse and return to a relatively mild stimulation configuration. This adaptive scheduling based on probability fluctuations allows electrical stimulation to continuously optimize as the patient's state evolves dynamically.

[0108] Through the above mechanism, the embodiment realizes the probabilistic generation and scheduling of electrical stimulation signals. Compared with the traditional single label driving method, it has the advantages of: being able to respond quickly to high-risk paths and improve safety; being able to weight and integrate complex states with multiple paths coexisting, enhancing flexibility; being able to adjust signals in the time dimension in real time, improving robustness and individualized adaptability.

[0109] The final output of the electrical stimulation parameter sequence is no longer a fixed template, but a combination signal dynamically adjusted according to the probability distribution, which can better match the metabolic-motor state characteristics of the patient and achieve precise and personalized neural regulation.

[0110] In this embodiment, the electrical stimulation signal generation model is designed as a double-layer structure to balance long-term stability and short-term flexibility, thereby meeting the dual requirements of safety and responsiveness of individualized neural regulation.

[0111] Basic layer: The design goal of the basic layer is to provide a stable electrical stimulation background signal to ensure that the patient is always in a safe and controllable stimulation environment in daily life. This layer mainly relies on the information of the stable part of the individualized state transition probability matrix, i.e., the transition relationships that show consistency and less fluctuation in the long term. By modeling these stable transition paths, the basic layer generates standardized electrical stimulation parameter sequences, including pulse frequency, pulse width, and amplitude, etc. For example, when the individual matrix shows that the transition probability from stable blood glucose to stable gait is maintained between 0.55-0.60 in multiple time windows, the system will configure a set of standardized parameters (such as frequency 20Hz, pulse width 200μs, amplitude 1.5mA) for this state and continuously output to maintain the stability of motor function. The existence of the basic layer guarantees the long-term consistency and safety of the stimulation signal, avoiding excessive stimulation or energy waste caused by frequent adjustments due to short-term fluctuations.

[0112] Dynamic adjustment layer: Under the premise of stable output of the basic layer, the role of the dynamic adjustment layer is to respond to the dynamic changes and risk highlights in the transfer path of the matrix in real time. This layer identifies the emergence of high-risk patterns by continuously monitoring the rapid changes in the probability of the matrix, and temporarily generates additional parameter correction signals if necessary. For example, when the system detects that the patient has a sharp drop in blood glucose in a short period of time, the probability of the path of moderate instability rises sharply (such as from 0.18 to 0.32), the dynamic adjustment layer will immediately start an additional control strategy: insert a high-frequency, short-period stabilizing pulse (such as 40Hz, pulse width 150μs, amplitude 1.0mA, duration 10 seconds) based on the standard sequence. This correction signal aims to quickly enhance neural excitability and muscle response, thereby reducing the risk of falls due to unstable posture, while providing a temporary buffer period for hypoglycemia on the metabolic level.

[0113] Multi-layer synergy: The synergy of the basic layer and the dynamic adjustment layer makes the electrical stimulation signal have both background stability and emergency flexibility. The standardized parameters provided by the basic layer ensure that the patient is in a safe and stable stimulation environment most of the time, and unnecessary intervention will not occur due to short-term noise or occasional fluctuations. The dynamic adjustment layer, as a rapid response module, can intervene immediately when a high-risk pattern appears, and modify the basic layer signal by inserting, superimposing or replacing, thereby achieving rapid defense against emergencies.

[0114] Compared with the traditional signal generation method which only relies on fixed parameters or single-layer mapping, the basic layer of the two-layer structure in this embodiment utilizes the stable part of the matrix, and the dynamic adjustment layer focuses on the change part of the risk highlight, achieving the division of labor of stability and sensitivity. The dynamic adjustment layer can generate temporary pulse correction in a short period of time without disrupting the continuous output of the basic layer, ensuring the continuity and flexibility of the overall system. The triggering condition of the dynamic adjustment layer not only depends on the single signal of blood glucose or movement, but also depends on the coupling characteristics of the two, making the correction more targeted.

[0115] After the generated electrical stimulation parameter sequence is issued to the electrical stimulation device worn by the patient via the mobile terminal application, the system of this embodiment will establish a closed-loop feedback loop based on real-time monitoring data of blood glucose and movement. This feedback loop is not just a one-way result collection channel, but an active adjustment module that continuously evaluates the immediate effect of electrical stimulation and cross-modal linkage effects.

[0116] During the operation of the feedback loop, the system first monitors the kinematic parameters, including gait stability, tremor amplitude, joint angular velocity, and acceleration fluctuation, to determine the immediate effect of electrical stimulation on motor improvement. For example, when the parameters show that the tremor amplitude has decreased by more than 25% and the gait has become more stable, the system determines that the stimulation has achieved positive results in the motor level. At the same time, the system synchronously analyzes the immediate change trend of the blood glucose curve to evaluate whether the stimulation has additional effects on the metabolic level. Since electrical stimulation may indirectly affect blood glucose levels through sympathetic nerve excitation and accelerated muscle energy metabolism, monitoring blood glucose signals is crucial to ensuring overall safety.

[0117] For example, when the system detects that electrical stimulation has indeed reduced the patient's tremor amplitude, but is accompanied by an increase in heart rate and a short-term increase in blood glucose of 0.5-0.8 mmol / L, the system determines that this combination of stimulation parameters has the characteristics of effective motor improvement but significant metabolic side effects. At this time, the feedback loop will initiate a parameter modification mechanism in the next stimulation cycle: reduce the pulse frequency (e.g., from 40 Hz to 25 Hz), or reduce the stimulation amplitude (e.g., from 2.0 mA to 1.5 mA), to weaken the overexcitation of the sympathetic nervous system while maintaining the necessary motor improvement effect.

[0118] In addition, the feedback loop is also designed with a trend adaptive adjustment function. When blood glucose fluctuations show a persistent change rather than a transient abnormality, the system gradually adjusts the baseline of the stimulation parameters. For example, when it is detected that the patient is accompanied by a slow trend of increasing blood glucose after multiple electrical stimulations, the system gradually reduces the upper limit of the amplitude of the basic layer output or shortens the stimulation duration, thereby forming a long-term optimization across modalities. This trend adjustment mechanism avoids the system only making short-term corrections after each stimulation, but rather achieves a higher level of adaptive optimization through accumulated feedback information.

[0119] Through the above design, the feedback loop of the present embodiment has a dual function in terms of function: Immediate response: real-time evaluation of motor improvement and blood glucose changes within one stimulation cycle, rapid adjustment of parameters in the next cycle to ensure short-term safety and efficacy.

[0120] Trend optimization: accumulate data on a time scale of multiple cycles, dynamically adjust the baseline level of the basic output, and achieve long-term optimization of the steady state across modalities.

[0121] Finally, the feedback loop makes the electrical stimulation signal generation process no longer a one-way generation-delivery linear process, but evolves into a closed-loop optimization system of generation-action-feedback-regeneration. The introduction of dynamic feedback across modalities, i.e., the comprehensive monitoring and adjustment of motor and metabolic dual signals, enables the output of electrical stimulation to improve motor impairment while also considering blood glucose homeostasis, thereby avoiding the accumulation of side effects.

[0122] Embodiment five provides a neuromodulation system combined with continuous glucose monitoring, as shown in the figure, which is composed of the following functional modules: Figure 5 a cloud modeling module, configured to construct a baseline state transition probability matrix between blood glucose level and exercise state based on historical data of multiple users, the baseline state transition probability matrix being used to represent the transition relationship between blood glucose fluctuation interval and exercise performance; a data acquisition module, including a continuous glucose monitoring sensor and an exercise state acquisition device, configured to acquire blood glucose time series data and kinematic data of a patient in real time when the patient wears the device, and compare the data in a recent preset time window with the baseline state transition probability matrix; an individualized correction module, configured to correct the baseline state transition probability matrix through a Bayesian updating method according to the comparison result, so as to obtain an individualized state transition probability matrix of the patient; an electrical stimulation signal generation module, deployed in the cloud, configured to receive the individualized state transition probability matrix and output a corresponding electrical stimulation parameter sequence according to the matrix; an electrical stimulation execution module, including an electrical stimulation device worn by the patient, configured to receive and apply the electrical stimulation parameter sequence to a preset part of the patient.

[0123] It should be noted that the above-mentioned neuromodulation method combined with continuous glucose monitoring is also applicable to the device of the present application, and will not be described here.

[0124] Those skilled in the art can realize that the units and algorithm steps described in the embodiments disclosed herein can be realized by electronic hardware, computer software and a combination of electronic hardware and computer software. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0125] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-mentioned system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0126] ​In several embodiments provided in the present application, any function, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or in part or parts of the technical solutions that make contributions to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory; hereinafter referred to as: ROM), a random access memory (Random Access Memory; hereinafter referred to as: RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0127] The above is only a specific embodiment of the present application, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. The module structure of the part of the present application not specifically mentioned is subject to the content recorded in the prior art. The prior art mentioned in the foregoing background section and the specific embodiment section of the present application can be used as a part of the present application to understand the meaning of some technical features or parameters.

Claims

1. A method of neuromodulation in conjunction with continuous glucose monitoring, characterized in that, The method comprises: constructing a baseline state transition probability matrix between blood glucose level and exercise state based on historical data of multiple users in the cloud, the baseline state transition probability matrix being used to represent the transition relationship between blood glucose fluctuation interval and exercise performance; acquiring blood glucose time series data and kinematic data of the patient in real time during the process of the patient wearing a continuous blood glucose monitoring sensor and an exercise state acquisition device, and comparing the data in a recent preset time window with the baseline state transition probability matrix; according to the comparison result, individualizing the baseline state transition probability matrix by a Bayesian updating method to obtain an individualized state transition probability matrix of the patient; inputting the individualized state transition probability matrix into a cloud-based electrical stimulation signal generation model, the electrical stimulation signal generation model outputting a corresponding electrical stimulation parameter sequence according to the individualized state transition probability matrix; applying the electrical stimulation parameter sequence to a preset part of the patient through an electrical stimulation device worn by the patient.

2. The method of neuromodulation in conjunction with continuous glucose monitoring of claim 1, wherein, The baseline state transition probability matrix is constructed by the following steps: detecting inflection points of blood glucose time series data of multiple users, determining trend turning points based on the sign change of the first order change rate and the amplitude of the second order change rate, and dividing the blood glucose curve into several metabolic time slices with the trend turning points as boundaries; cutting corresponding kinematic time series data under the boundary conditions of the metabolic time slices to strictly align blood glucose segments with exercise segments; extracting multi-dimensional features from each exercise segment and performing normalization processing; mapping the multi-dimensional features into a distribution type exercise state vector, each dimension of the distribution type vector representing the probability of the segment belonging to different exercise sub-states; labeling the blood glucose trend of adjacent metabolic time slices and pairing the change of the exercise state distribution vector to form a cross-modal transition pair; calculating a coupling degree index of the cross-modal transition pair, the coupling degree being determined by the cross information amount and the overlap degree between the distributions; performing normalization processing on the obtained matrix to obtain a baseline state transition probability matrix with row vectors constituting probability distributions.

3. The method of neuromodulation in conjunction with continuous glucose monitoring of claim 2, wherein, The baseline state transition probability matrix is stored in a double-layer structure, wherein the group average layer is used to represent the general transition rule, and the dynamic event layer is used to retain the high fluctuation characteristics appearing in a small number of users.

4. The method of neuromodulation in conjunction with continuous glucose monitoring of claim 1, wherein, The individualizing correction step comprises: comparing the blood glucose trend and exercise state distribution of the patient in the preset time window with the baseline state transition probability matrix item by item to obtain a difference quantization result; determining a correction amplitude according to the occurrence frequency of the transition mode in the current time window; strengthening or weakening the correction value according to the correlation degree of the blood glucose change rate and the exercise state fluctuation rate; gradually reducing the influence of data outside the time window in the correction process to generate an individualized state transition probability matrix.

5. The method of neuromodulation in conjunction with continuous glucose monitoring of claim 1, wherein, The step of outputting a corresponding electrical stimulation parameter sequence comprises: extracting transition features related to blood glucose intervals and fluctuation features related to exercise state distributions, and taking a distribution vector as an input feature; when the probability of a certain transition path significantly increases, the priority of the electrical stimulation mode corresponding to the path is increased. When multiple transition paths have high probabilities at the same time, the path probabilities are weighted and fused to form a combined parameter sequence; The signal generation model includes a basic layer and a dynamic adjustment layer, the basic layer is used to output normalized pulse frequency, pulse width and amplitude parameters according to the stable part of the matrix, and the dynamic adjustment layer is used to generate an additional correction signal for a high-risk transition path that dynamically changes in the matrix, and to dynamically adjust the subsequent output through real-time feedback after the electric stimulation parameters are sent to the electric stimulation device worn by the patient.

6. A neuromodulation system in combination with continuous glucose monitoring, characterized in that, The system comprises the following modules: A cloud modeling module is configured to construct a baseline state transition probability matrix between blood glucose levels and motion states based on historical data of multiple users, and the baseline state transition probability matrix is used to represent the transition relationship between blood glucose fluctuation intervals and motion function performance. A data acquisition module is configured to acquire blood glucose time series data and kinematic data of a patient in real time during the process in which the patient wears a continuous blood glucose monitoring sensor and a motion state acquisition device, and compare data within a recent preset time window with the baseline state transition probability matrix. An individualized correction module is configured to correct the baseline state transition probability matrix through a Bayesian updating method according to the comparison result, so as to obtain an individualized state transition probability matrix of the patient. A signal generation module is configured to input the individualized state transition probability matrix into a cloud electric stimulation signal generation model, and output a corresponding electric stimulation parameter sequence according to the matrix. An execution device is an electric stimulation device worn by the patient, and is configured to apply the electric stimulation parameter sequence to a preset part of the patient.

7. The neuromodulation system in conjunction with continuous glucose monitoring of claim 6, wherein, The cloud modeling module further comprises: A turning point detection unit is configured to detect turning points of blood glucose time series data of multiple users, and determine trend turning points based on a sign change of a first-order change rate and a magnitude of a second-order change rate. A time slice division unit is configured to divide a blood glucose curve into a plurality of metabolic time slices with the trend turning points as boundaries, and cut and align corresponding kinematic time series data under boundary conditions of the metabolic time slices. A feature processing unit is configured to extract multi-dimensional features of each motion segment and perform normalization processing. A distribution modeling unit is configured to map the multi-dimensional features into a distribution type motion state vector, and each dimension of the distribution type vector represents a probability that the segment belongs to a different motion sub-state. A coupling calculation unit is configured to pair blood glucose trend changes and motion state distribution vector changes of adjacent metabolic time slices to form cross-modal transition pairs, and calculate a coupling degree index determined by cross information and overlap. A matrix generation unit is configured to normalize the obtained matrix to obtain a baseline state transition probability matrix in which row vectors constitute a probability distribution.

8. The neuromodulation system in conjunction with continuous glucose monitoring of claim 7, wherein, The baseline state transition probability matrix constructed by the cloud modeling module is stored in a double-layer structure, wherein a group average layer is used to represent a general transition rule, and a dynamic event layer is used to retain high fluctuation characteristics appearing in a small number of users.

9. The neuromodulation system incorporating continuous glucose monitoring of claim 6, wherein, The individualized correction module comprises: An comparison unit is configured to compare blood glucose trend changes and motion state distributions of a patient within a preset time window with the baseline state transition probability matrix item by item to obtain a difference quantization result. a dynamic weight unit configured to determine a correction amplitude according to the frequency of occurrence of the transition mode within a current time window; a coupling adjustment unit configured to enhance or weaken the correction value according to the correlation between the blood glucose change rate and the motion state fluctuation rate; a time decay unit configured to gradually reduce the influence of data outside the time window in the correction process, thereby generating an individualized state transition probability matrix.

10. The neuromodulation system incorporating continuous glucose monitoring of claim 6, wherein, The signal generation module comprises: a feature extraction unit configured to extract transition features related to blood glucose intervals and fluctuation features related to motion state distribution, and to use a distribution vector as an input feature; a scheduling unit configured to increase the priority of the electrical stimulation mode corresponding to a certain transition path when the probability of the path is significantly increased, to perform weighted fusion according to the path probability when multiple transition paths have high probabilities at the same time, and to form a combined parameter sequence; a double-layer generation unit comprising a base layer and a dynamic adjustment layer, wherein the base layer is configured to output standardized pulse frequency, pulse width and amplitude parameters according to the stable part of the matrix, and the dynamic adjustment layer is configured to generate an additional correction signal for the high-risk transition path that dynamically changes in the matrix, and to dynamically adjust the subsequent output after the electrical stimulation parameters are issued to the electrical stimulation device worn by the patient through real-time feedback.

Images