An intelligent electric acupuncture frequency optimization system and method based on a group response atlas

Abstract

The application discloses an intelligent electric acupuncture frequency optimization system and method based on a group response atlas. The system comprises an electroencephalogram data processing module, a group response atlas construction module, a starting frequency recommendation module, a real-time state evaluation module and a closed-loop control module. The group response atlas construction module performs statistics on the electroencephalogram feature increments before and after stimulation under different electric acupuncture frequencies, and trains a baseline-response mapping model for each frequency by using the historical pre-stimulation electroencephalogram features and corresponding feature increments; the starting frequency recommendation module inputs the standardized pre-stimulation electroencephalogram features of the current subject to be intervened into the baseline-response mapping model for each frequency, obtains the predicted responses under each frequency, and calculates the predicted benefit score by combining the electroencephalogram feature component weight, response dispersion and preset beneficial direction, and selects the frequency with the highest score as the starting frequency; during the treatment process, the system determines whether to switch the candidate frequency according to the real-time electroencephalogram response, the group response atlas and the safety score, so that the dynamic optimization of the electric acupuncture frequency is realized in a single stimulation process.

Description

Technical Field

[0001] This invention relates to electroacupuncture devices, and more particularly to an intelligent electroacupuncture frequency optimization system and method based on population response spectrum and real-time closed-loop control. Background Technology

[0002] Insomnia is a common sleep disorder in clinical practice, often manifested as difficulty falling asleep, difficulty maintaining sleep, early awakening, and the resulting daytime fatigue, decreased attention, and mood abnormalities. Electroacupuncture, as a treatment method combining traditional Chinese acupuncture with modern electrical stimulation technology, is relatively simple to operate, has adjustable parameters, and good patient compliance, making it valuable in the adjunctive treatment of insomnia and related brain dysfunction.

[0003] In current electroacupuncture treatments, the stimulation frequency is usually set by the physician based on experience, with commonly used frequencies including 2Hz, 5Hz, and 10Hz. Different frequencies may induce different neuromodulation effects, and different patients show significant differences in their EEG responses and clinical benefits at the same frequency. Fixed-frequency protocols struggle to simultaneously account for both group patterns and individual differences, easily leading to problems such as insufficient selection of the starting frequency, lack of dynamic feedback during treatment, and difficulty in accumulating efficacy data.

[0004] In real-world clinical settings, it is generally not advisable to sequentially complete three full treatment sessions at 2Hz, 5Hz, and 10Hz for the same patient before comparing the optimal frequency, as this increases the treatment burden and affects clinical efficiency. Therefore, there is a need for an intelligent electroacupuncture system that can utilize past group data to establish frequency response patterns and dynamically optimize the system by incorporating the patient's real-time EEG status during a single treatment session. Summary of the Invention

[0005] To address the problems existing in the prior art, the purpose of this invention is to provide an intelligent electroacupuncture frequency optimization system and method that can optimize the electroacupuncture frequency by combining the patient's real-time EEG status during a single electroacupuncture stimulation.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0007] A smart electroacupuncture frequency optimization system based on population response spectrum, comprising:

[0008] The EEG data processing module is used to acquire historical EEG signals of several subjects before and after stimulation at different electroacupuncture frequencies, and to extract several pairs of EEG features before and after stimulation at different electroacupuncture frequencies.

[0009] The group response mapping module is used to construct a group response mapping of the subject population to different electroacupuncture frequencies by taking the changes in EEG characteristics before and after stimulation with different electroacupuncture frequencies as corresponding responses.

[0010] The starting frequency recommendation module is used to input the standardized pre-stimulation EEG characteristics of the current subject to intervention into the baseline-response mapping model of each electroacupuncture frequency to obtain the predicted response at each electroacupuncture frequency. Based on the weight of the EEG feature components in the group response map, the response dispersion and the preset beneficial direction, the predicted benefit score of each electroacupuncture frequency is calculated, and the electroacupuncture frequency with the highest score is selected as the starting recommendation frequency.

[0011] The real-time state assessment module is used to extract real-time EEG features during electroacupuncture stimulation to obtain real-time responses, and to determine the benefits at the current frequency and the benefits at each candidate frequency based on the real-time responses and the group response maps of different electroacupuncture frequencies.

[0012] The closed-loop control module is used to determine whether to switch to a candidate electroacupuncture frequency based on the benefit at the current frequency and the benefit at each candidate frequency.

[0013] Furthermore, the EEG features include at least one EEG feature component, which includes the mean of each brain region, the standard deviation of each brain region, the asymmetry between brain regions, and the cross-brain region difference features.

[0014] Furthermore, the group response mapping module specifically includes:

[0015] The response determination unit is used to divide all EEG feature pairs under the same stimulus frequency into a group and calculate the relative changes of each EEG feature component before and after stimulation in each EEG feature pair within each group, as the response in the corresponding EEG feature component.

[0016] The distribution statistics unit is used to calculate the mean, standard deviation, and weight of all responses for each EEG feature component within each group, and to train a baseline-response mapping model based on historical pre-stimulation EEG features and corresponding responses at the same stimulation frequency; wherein, the weight of the EEG feature component is obtained by weighting and normalizing at least two of the following: statistical significance score, effect size score, cross-validation feature importance, and expert prior value.

[0017] Furthermore, the starting frequency recommendation module specifically includes:

[0018] The pre-stimulation EEG feature extraction unit is used to acquire the pre-stimulation EEG signal of the current subject to intervention and extract the pre-stimulation EEG features.

[0019] The prediction scoring unit is used to extract the baseline-response mapping model, EEG feature component weights, and response standard deviation of each electroacupuncture frequency from the population response spectrum of different electroacupuncture frequencies. The standardized pre-stimulation EEG features of the current subject to intervention are input into each baseline-response mapping model to obtain the predicted response. The predicted benefit score of each electroacupuncture frequency is calculated according to the preset beneficial direction, and the electroacupuncture frequency with the highest score is selected as the starting recommended frequency.

[0020] Furthermore, the real-time status assessment module specifically includes:

[0021] The real-time EEG feature extraction unit is used to acquire the real-time EEG signal of the current subject after electroacupuncture stimulation and extract the real-time EEG features after stimulation.

[0022] The benefit determination unit is used to calculate the relative change in the EEG characteristics of the current subject after real-time stimulation and the EEG characteristics before stimulation, as a real-time response. Based on the real-time response and the group response spectrum of the current electroacupuncture frequency and each candidate electroacupuncture frequency, the benefit score of the current electroacupuncture frequency and each candidate electroacupuncture frequency is determined; the candidate electroacupuncture frequencies are all electroacupuncture frequencies other than the current electroacupuncture frequency.

[0023] Furthermore, the closed-loop control module is specifically used for:

[0024] When the benefit score of a candidate electroacupuncture frequency is greater than that of the current electroacupuncture frequency, a frequency switching command is output to switch to the candidate electroacupuncture frequency with the highest benefit score.

[0025] Furthermore, the closed-loop control module can also be used for:

[0026] The safety score at the current moment is calculated. The safety score is obtained by weighting the window quality, patient feedback comfort, and normalized indicators of artifact or stimulation risk level. The artifact or stimulation risk level is obtained by weighting the stimulation artifact rate, electromyographic artifact rate, and the normalized values ​​of the current amplitude, single-phase pulse width, and duration of continuous stimulation of the electroacupuncture stimulator in the current control cycle relative to the corresponding preset safety upper limit.

[0027] If the benefit score of a candidate electroacupuncture frequency is greater than the benefit score of the current electroacupuncture frequency, and the safety score is greater than a preset threshold, and the interval between adjacent frequency switching is greater than the minimum holding time, then a frequency switching command is output to switch to the candidate electroacupuncture frequency with the highest benefit score.

[0028] Furthermore, the system also includes:

[0029] The efficacy tracking and interaction module is used to display the subject's EEG status, current stimulation frequency, recommended stimulation frequency, and reports, and to receive and execute the doctor's confirmation, overriding, or termination instructions.

[0030] A smart electroacupuncture frequency optimization method based on population response spectrum includes:

[0031] We acquired historical EEG signals from several subjects before and after stimulation at different electroacupuncture frequencies, and extracted several pairs of EEG characteristics before and after stimulation at different electroacupuncture frequencies.

[0032] The changes in EEG characteristics before and after stimulation with different electroacupuncture frequencies were used as corresponding responses to construct a population response map of the subject group to different electroacupuncture frequencies. Based on the EEG characteristics before historical stimulation and the corresponding responses, a baseline-response mapping model for each electroacupuncture frequency was trained.

[0033] The standardized pre-stimulation EEG characteristics of the current subjects to be intervened are input into the baseline-response mapping model of each electroacupuncture frequency to obtain the predicted response at each electroacupuncture frequency. The predicted benefit score is calculated by combining the weight of the EEG feature components in the group response map, the response dispersion, and the preset beneficial direction. The electroacupuncture frequency with the highest score is selected as the starting recommended frequency.

[0034] Real-time EEG features were extracted during electroacupuncture stimulation to obtain real-time responses. Based on the real-time responses and the group response maps of different electroacupuncture frequencies, the benefits at the current frequency and the benefits of each candidate frequency were determined.

[0035] Based on the benefits at the current frequency and the benefits at each candidate frequency, determine whether to switch to a candidate electroacupuncture frequency.

[0036] A computer program product includes a computer program that, when executed by a processor, implements the above-described method.

[0037] Compared with the prior art, the beneficial effects of this invention are: the subjects to be intervened in this invention do not need to undergo multiple trials in the early stage, and a better initial stimulation frequency can be obtained in a single stimulation process. Moreover, during the stimulation process, the stimulation frequency can be dynamically optimized by combining the patient's real-time EEG status. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of an embodiment of the intelligent electroacupuncture frequency optimization system based on population response spectrum provided by the present invention;

[0039] Figure 2 This is a schematic diagram of another embodiment of the intelligent electroacupuncture frequency optimization system based on population response spectrum provided by the present invention;

[0040] Figure 3 This is a schematic diagram of the group response map construction process provided by the present invention;

[0041] Figure 4 This is a schematic diagram of the starting frequency recommendation process provided by the present invention;

[0042] Figure 5 This is a schematic diagram of the real-time closed-loop control process provided by the present invention;

[0043] Figure 6 This is a flowchart illustrating the stimulation frequency switching process;

[0044] Figure 7This is a schematic diagram of the clinical workflow and doctor interaction interface. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0046] Example 1

[0047] This invention provides an intelligent electroacupuncture frequency optimization system based on population response mapping, such as... Figure 1 and Figure 2 As shown, it includes an EEG data processing module, a group response map construction module, an initiation frequency recommendation module, a real-time state assessment module, and a closed-loop regulation module.

[0048] The EEG data processing module is used to acquire historical EEG signals of several subjects before and after stimulation at different electroacupuncture frequencies, and extract several pairs of EEG features before and after stimulation at different electroacupuncture frequencies.

[0049] Historical EEG signals can be extracted from a local database, a hospital's dedicated network server, or a cloud-controlled database. Each historical record includes at least the patient's basic information, EEG acquisition time, stimulation frequency, stimulation intensity, stimulation duration, EEG signals before and after stimulation treatment, changes in sleep-related scales, physician evaluation, and adverse reaction records. To protect privacy, the database may store patient identity information in anonymized form and may only be used with the patient's permission. EEG signals before and after stimulation using the same stimulation frequency are considered as one record. Multiple records may come from the same patient or different patients. The number of records should be sufficient to cover different stimulation frequencies. Considering that commonly used stimulation frequencies are 2Hz, 5Hz, and 10Hz, in this embodiment, the stimulation frequency set is {2Hz, 5Hz, 10Hz}. It is understood that other suitable stimulation frequencies can also be selected.

[0050] EEG signals require preprocessing, including bandpass filtering, power line notch filtering, bad conduction detection, rereference, removal of EEG and EMG artifacts, suppression of stimulus artifacts, and window quality evaluation. For stimulus artifacts, temporal window masking, interpolation reconstruction, or adaptive filtering can be performed in conjunction with stimulus timing markers. The window quality evaluation outputs the artifact rate; when the artifact rate exceeds a threshold, the window is not used for closed-loop control or triggers re-acquisition.

[0051] The steps for extracting EEG features are as follows: extract the alpha, beta, theta, and delta wave band power or relative power, the theta / beta ratio, the delta / theta ratio, the slow / fast wave ratio, and the spectral entropy for each brain region. Among these, band power, relative power, band ratio, slow / fast wave ratio, and spectral entropy are fundamental indicators that can be calculated using existing EEG analysis methods. For any brain region r and band b, the brain region band power... and relative power It can be represented as:

[0052]

[0053]

[0054] in, This represents the set of channels corresponding to brain region r. () indicates the frequency band range. Indicates channel In frequency Power spectral density at that point This represents the smallest positive number to prevent the denominator from being zero. The channel set corresponding to brain region r, the frequency range of frequency band b, the channel power spectral density, and the smallest positive number to prevent the denominator from being zero are all determined according to unified preprocessing and feature extraction rules.

[0055] Then, based on these basic indicators, and according to their respective brain regions, frequency bands, and left-right hemisphere relationships, the brain region mean, brain region standard deviation, inter-brain region asymmetry, and cross-brain region difference characteristics are calculated for each indicator. Inter-brain region asymmetry is calculated based on the mean and / or standard deviation of all brain regions. For example, inter-brain region asymmetry can be the difference between the maximum and minimum mean values ​​of a brain region, divided by the sum of the maximum and minimum mean values. Cross-brain region difference characteristics can be the differences between the corresponding indicators. After baseline correction and standardization to a unified dimension, the brain region mean, brain region standard deviation, inter-brain region asymmetry, and cross-brain region difference characteristics are selected as one or more EEG feature components to form EEG characteristics. In this embodiment, the brain region mean, brain region standard deviation, inter-brain region asymmetry, and cross-brain region difference characteristics are all used as EEG feature components to form EEG characteristics.

[0056] The pre-stimulation EEG signals of each recording were extracted to obtain the pre-stimulation EEG features. , Let represent the i-th EEG feature component in the pre-stimulation EEG features, and M represent the number of EEG feature components. Then, the post-stimulation EEG signal is extracted to obtain the post-stimulation EEG features. , This represents the i-th EEG feature component in the post-stimulation EEG features. The pre-stimulation EEG features for each recording are also represented. and EEG characteristics after stimulation As a brainwave feature for X= .

[0057] The group response mapping module is used to construct a group response map of the subject population to different electroacupuncture frequencies by using changes in EEG characteristics before and after stimulation with different electroacupuncture frequencies as corresponding responses. For example... Figure 3 As shown, it specifically includes:

[0058] The response determination unit is used to group all EEG feature pairs at the same stimulus frequency into a group, and calculate the relative changes of each EEG feature component before and after stimulation in each EEG feature pair within each group, as the response of the corresponding EEG feature component; that is, the response of EEG feature component i is... ;

[0059] The distribution statistics unit is used to calculate the weight of each EEG feature component in each group and the mean and standard deviation of all responses of each EEG feature component, forming a group response map of the corresponding stimulation frequency, and training the baseline-response mapping model of each electroacupuncture frequency based on the EEG features before historical stimulation and the corresponding responses.

[0060] Specifically, for each stimulation frequency All responses of each EEG feature component i at that frequency After statistical screening, outlier removal, and standardization, the mean response was calculated. and standard deviation .

[0061] Weights of EEG feature component i The score can be obtained by weighting and normalizing at least two of the following: statistical significance score, effect size score, cross-validation feature importance, and expert prior value. To ensure the weights of the EEG feature components can be repeatedly calculated, this embodiment obtains four categories of normalized scores (0 to 1) for each stimulation frequency f and EEG feature component i: the statistical significance score is obtained by normalizing the p-values ​​of paired tests before and after stimulation at the same frequency using a negative logarithmic transformation; the effect size score is obtained by normalizing the absolute value of the response mean divided by the response standard deviation; the feature importance score is obtained by replacing the importance of cross-validation with the improvement label of therapeutic efficacy or the response prediction error as the supervision target; and the expert prior value is given by the physician according to a preset scale. When a certain type of score is unavailable, its combination coefficient is set to 0, and the remaining combination coefficients are renormalized. The unnormalized composite score is calculated first. :

[0062]

[0063]

[0064] In the formula, , These represent the normalized statistical significance score, effect size score, feature importance score, and expert prior value, respectively. The combination coefficients are, for example, all taken as 0.25. The overall score is then normalized across all M EEG feature components at the same stimulus frequency to obtain the weights:

[0065]

[0066] In the formula, It represents a very small number.

[0067] Stimulation frequency Mean value of each EEG characteristic component i Standard deviation and weights of EEG feature components Together they form a collective response map. The population response maps for all stimulus frequencies form a population response map library. In other embodiments, the response probability distribution, at least one valid sample, the number of valid samples, and the last update time can also be stored in the population response map.

[0068] Furthermore, for each stimulus frequency f, a baseline-response mapping model is trained by taking the standardized pre-stimulus EEG feature component i from historical records as input and the corresponding post-stimulus EEG feature increment as output. The baseline-response mapping model can employ regularized linear regression, random forest, gradient boosting tree, or other supervised learning models, and cross-validation can be used to select model parameters.

[0069] Therefore, in this embodiment, the frequency group response map, in addition to storing the response mean, standard deviation, and normalized weights, also stores the baseline-response mapping model, model cross-validation error, preset beneficial directions for each EEG feature component, and the most recent update time. This data structure is used to predict the patient's response at the corresponding electroacupuncture frequency from the baseline, avoiding a direct comparison between the absolute characteristics before stimulation and the increments after stimulation.

[0070] The starting frequency recommendation module does not directly calculate the similarity between the patient's pre-stimulation EEG characteristics and the feature increments before and after stimulation. Instead, it uses the baseline-response mapping model in the population response atlas to predict the potential EEG feature increments at each electroacupuncture frequency, then calculates the predicted benefit score for each electroacupuncture frequency and recommends the starting frequency. For example... Figure 4 As shown, it specifically includes:

[0071] The pre-stimulation EEG feature extraction unit is used to acquire the pre-stimulation EEG signal of the current patient and extract the pre-stimulation EEG features. A 61-channel EEG acquisition device is preferred for acquiring the pre-stimulation EEG signal; however, a multi-channel EEG acquisition device covering the frontal, central, parietal, and occipital regions can also be used depending on clinical conditions. The acquisition module records the resting-state baseline EEG (pre-stimulation EEG signal) before stimulation therapy, records online EEG during stimulation therapy, and records the efficacy evaluation EEG after treatment, simultaneously recording stimulation start, stimulation end, frequency switching, and stimulation pause events. The EEG feature extraction steps are the same as described above. The final pre-stimulation EEG feature X0 of the patient is obtained.

[0072] The predictive scoring unit extracts the baseline-response mapping model, EEG feature component weights, and response standard deviation from the population response spectra at different electroacupuncture frequencies. It inputs the standardized pre-stimulation EEG features of the current patient into the baseline-response mapping model for each frequency to obtain the corresponding predicted response. The predicted return score for frequency f can be expressed as:

[0073]

[0074]

[0075] In the formula, The predicted benefit score represents the stimulus frequency f. This represents the normalized weight of EEG feature component i in the group response map. This indicates the beneficial direction of the EEG feature component corresponding to the improvement in treatment efficacy; a value of +1 indicates a beneficial outcome, and -1 indicates an unfavorable outcome. The predicted response is obtained by the baseline-response mapping model based on the patient's standardized pre-stimulation EEG features. ε represents the standard deviation of the corresponding response in the population response map, where ε is a very small positive number to prevent the denominator from being zero. The larger the standard deviation, the higher the uncertainty of this feature in the population, and the smaller its contribution to the predicted return score.

[0076] Therefore, the initial frequency recommendation compares the standardized response gains predicted by each frequency model based on the same patient baseline, rather than the direct similarity between the patient's pre-stimulation characteristics and the post-stimulation increment of the population. When the difference between the highest predicted gain score and the second highest score is lower than a preset threshold, or when the model cross-validation error exceeds a preset upper limit, the system outputs a set of candidate frequencies and requests the doctor's confirmation, without automatically determining a unique frequency.

[0077] The real-time state assessment module is used to extract real-time EEG features during electroacupuncture stimulation to obtain real-time responses. Based on the real-time responses and the group response profiles at different electroacupuncture frequencies, it determines the benefit at the current frequency and the benefits at each candidate frequency. Figure 5 As shown, the real-time status assessment module specifically includes:

[0078] The real-time EEG feature extraction unit is used to acquire the real-time EEG signal of the current subject after electroacupuncture stimulation and extract the real-time EEG features after stimulation. Here, the EEG data is divided into continuous or partially overlapping sliding time windows. The sliding window is used to extract the signal. The EEG signal in the current sliding window is used as the real-time EEG signal. Then, preprocessing and feature extraction are performed to obtain the real-time EEG features after stimulation.

[0079] The assessment unit is used to calculate the relative change (i.e., the difference) between the real-time EEG characteristics of the subject after stimulation and the pre-stimulation EEG characteristics, as a real-time response. Based on the real-time response and the group response spectrum of the current electroacupuncture frequency and each candidate electroacupuncture frequency, the benefit scores of the current electroacupuncture frequency and each candidate electroacupuncture frequency are determined; the candidate electroacupuncture frequencies are all electroacupuncture frequencies other than the current electroacupuncture frequency; candidate frequencies In the sliding window Under the profit rating It can be represented as:

[0080]

[0081]

[0082] In the formula, This represents the smoothing coefficient, for example, 0.2. This indicates the instantaneous matching benefit. express The i-th component value, This is a local minimum value, used to prevent the denominator from being 0.

[0083] The closed-loop control module determines whether to switch to a candidate electroacupuncture frequency based on the benefit score of the current frequency and the benefits of each candidate frequency. Specifically, when the benefit score of a candidate electroacupuncture frequency is greater than that of the current electroacupuncture frequency, it outputs a frequency switching command to switch to the candidate electroacupuncture frequency with the highest benefit score.

[0084] In other embodiments, to improve patient comfort and avoid discomfort caused by frequent switching, such as Figure 6 As shown, the closed-loop control module can also perform the following steps:

[0085] A. Calculate the security score at the current moment, the security score It is obtained by weighting normalized indicators based on window quality, patient feedback comfort, and the risk level of artifacts or irritants; specifically:

[0086]

[0087] In the formula, , These represent weight values, such as α=0.40, β=0.30, γ=0.30. The window quality, which can be represented by a weighted sum and normalized sum of artifact rate, bad conduction ratio, and signal-to-noise ratio, is as follows: This represents the patient feedback score, which is obtained by extracting scores from the patient's subjective feedback scale and then normalizing them. The risk level of artifacts or stimulation can be obtained by normalizing risk factors such as stimulation artifact rate, electromyography artifact rate, and the actual output current amplitude, monophasic pulse width, and duration of continuous stimulation of the electroacupuncture stimulator in the current control cycle. All of the above variables are converted to the same scoring interval before being included in the safety score, thereby avoiding the direct addition of indicators with different dimensions.

[0088] The current amplitude, single-phase pulse width, and continuous stimulation duration mentioned here all refer to the actual output parameters of the electroacupuncture stimulator in the current control cycle, and are denoted as follows: , and Its corresponding security limit , and The risk level is determined jointly by the device's safety range and the doctor's prescription. Stimulation artifact rate and electromyography artifact rate are cropped to the 0-1 range, and the above three output parameters are divided by their respective safety upper limits and then cropped to the 0-1 range, indicating the stimulation risk level. Calculate using the following formula:

[0089]

[0090]

[0091] and These represent the stimulus artifact rate and electromyography artifact rate within the current EEG window, respectively; clip(z) restricts z to the interval between 0 and 1. to The risk weights are non-negative and sum to 1, for example, all weights are 0.20. Window quality. The weighted sum of the artifact rate, bad conduction ratio, and low signal-to-noise ratio penalty term can be taken as 1, and the patient's feedback on comfort can be obtained. It can be obtained by subtracting the discomfort score from 1 and dividing by 10 from the 0 to 10 discomfort scale.

[0092] For example, taking α=0.40, β=0.30, γ=0.30, if =0.90、 =0.80、 =0.20, then the safety score is =0.54; when the preset safety threshold η=0.50, the safety gating is satisfied. If other conditions remain unchanged... If it rises to 0.70, then =0.39, which is below the threshold. The system will not perform frequency switching and will maintain the current frequency, reduce the stimulation output, or request confirmation from the doctor.

[0093] B. Maximum benefit score when candidate electroacupuncture frequencies exist. Greater than the current electroacupuncture frequency Earnings rating : , It is a threshold and a security score. Greater than the preset threshold If the interval between adjacent frequency switching is greater than the minimum hold time, then the output switching is the one with the highest benefit score. Candidate electroacupuncture frequency The frequency switching command will be executed. Otherwise, the current frequency will be maintained or safety actions such as downgrading or pausing will be performed.

[0094] In clinical applications, after patients complete registration and pre-treatment EEG data acquisition, the system outputs a suggested starting frequency, which the doctor confirms before administering electroacupuncture treatment. During treatment, the system interface displays real-time EEG waveforms, current frequency, a list of recommended frequencies, benefit score, safety score, and efficacy trend, and provides operation buttons such as confirm, override, pause, and terminate.

[0095] In other embodiments, the system also includes a treatment efficacy tracking and interaction module, used to display the subject's EEG status, current stimulation frequency, recommended stimulation frequency, and reports, and to receive and execute physician confirmation, override, or termination commands. Furthermore, after treatment, it can summarize changes in EEG characteristics before and after treatment, changes in sleep scales, patient subjective feedback, and physician evaluations to form a quantitative treatment efficacy record.

[0096] In other embodiments, the system further includes a model update unit for feeding the record back to the population response map construction module to update the population response maps at frequencies such as 2Hz, 5Hz, and 10Hz.

[0097] In this embodiment, before treatment begins, the system does not require the same patient to complete three full treatment courses at frequencies of 2Hz, 5Hz, and 10Hz sequentially. Instead, it matches the patient's pre-treatment EEG characteristics with the group response map to obtain a score. This score is only used for candidate frequency ranking and physician confirmation and is not interpreted as the absolute magnitude of the original EEG physiological quantities. During treatment, the system continuously calculates the patient's current EEG response using a sliding time window and compares this response with the group response map for each frequency. The closed-loop control module only allows the output of a frequency switching command when the candidate frequency shows a stable increase in benefit relative to the current frequency, and the number of continuous observation windows, window quality, artifact risk, patient subjective feedback, and stimulation dose all meet safety constraints. When the benefit of the candidate frequency increases but the safety score is insufficient, or the difference between the highest and second-highest scores is less than a preset threshold, the system does not automatically switch frequencies but maintains the current frequency, reduces the stimulation intensity, pauses stimulation, or requests physician confirmation. This setting provides a safety gating and manual overriding mechanism for closed-loop frequency control, avoiding automatic adjustment based solely on a single EEG indicator. After treatment, the system records changes in EEG before and after treatment, benefit scores during treatment, frequency switching records, changes in sleep scales, patient subjective feedback, and physician evaluations in a historical database, and feeds this information back to the corresponding population response map according to actual usage frequency. Thus, the population response map is continuously updated with clinical data, forming a complete technological chain from historical data modeling, individual initial recommendations, closed-loop regulation during treatment, to post-treatment model updates.

[0098] Example 2

[0099] This invention provides an intelligent electroacupuncture method based on population response mapping and real-time closed-loop control, comprising:

[0100] S1. Obtain historical EEG signals of several subjects before and after stimulation at different electroacupuncture frequencies, and extract several pairs of EEG characteristics before and after stimulation at different electroacupuncture frequencies.

[0101] The steps for extracting EEG features are as follows: extract the alpha, beta, theta, and delta wave band power or relative power, the theta / beta ratio, the delta / theta ratio, the slow / fast wave ratio, and the spectral entropy for each brain region. Among these, band power, relative power, band ratio, slow / fast wave ratio, and spectral entropy are fundamental indicators that can be calculated using existing EEG analysis methods. For any brain region r and band b, the brain region band power... and relative power It can be represented as:

[0102]

[0103]

[0104] in, This represents the set of channels corresponding to brain region r. () indicates the frequency band range. Indicates channel In frequency Power spectral density at that point This represents the smallest positive number to prevent the denominator from being zero. The channel set corresponding to brain region r, the frequency range of frequency band b, the channel power spectral density, and the smallest positive number to prevent the denominator from being zero are all determined according to unified preprocessing and feature extraction rules.

[0105] Then, based on these basic indicators, and according to their respective brain regions, frequency bands, and left-right hemisphere relationships, the brain region mean, brain region standard deviation, inter-brain region asymmetry, and cross-brain region difference characteristics are calculated for each indicator. Inter-brain region asymmetry is calculated based on the mean and / or standard deviation of all brain regions. For example, inter-brain region asymmetry can be the difference between the maximum and minimum mean values ​​of a brain region, divided by the sum of the maximum and minimum mean values. Cross-brain region difference characteristics can be the differences between the corresponding indicators. After baseline correction and standardization to a unified dimension, the brain region mean, brain region standard deviation, inter-brain region asymmetry, and cross-brain region difference characteristics are selected as one or more EEG feature components to form EEG characteristics. In this embodiment, the brain region mean, brain region standard deviation, inter-brain region asymmetry, and cross-brain region difference characteristics are all used as EEG feature components to form EEG characteristics.

[0106] S2. Using the changes in EEG characteristics before and after stimulation with different electroacupuncture frequencies as corresponding responses, construct a population response map of the subject group to different electroacupuncture frequencies.

[0107] Specifically, it includes:

[0108] All EEG feature pairs at the same stimulus frequency are grouped together, and the relative changes of each EEG feature component before and after stimulation are calculated for each EEG feature pair within each group. This is taken as the response of the corresponding EEG feature component; that is, the response of EEG feature component i is... ;

[0109] Calculate the weights of each EEG feature component within each group, as well as the mean and standard deviation of all responses for each EEG feature component, to form a population response profile for the corresponding stimulus frequency. Specifically, for each stimulus frequency... All responses of each EEG feature component i at that frequency After statistical screening, outlier removal, and standardization, the mean response was calculated. and standard deviation Weights of EEG feature component i The significance, effect size, characteristic importance, or weights set by clinical experts can be used to determine the frequency of stimulation. Mean value of each EEG characteristic component i Standard deviation and weights of EEG feature components Together they form a collective response map. The population response maps for all stimulus frequencies form a population response map library. In other embodiments, the response probability distribution, at least one valid sample, the number of valid samples, and the last update time can also be stored in the population response map.

[0110] The weights of the EEG feature components were determined according to the repeatable calculation method described in Example 1. Specifically, the statistical significance, effect size, cross-validation feature importance, and expert prior values ​​were first normalized and a comprehensive score was calculated:

[0111]

[0112]

[0113] In the formula, , These represent the normalized statistical significance score, effect size score, feature importance score, and expert prior value, respectively. The combination coefficients are, for example, all taken as 0.25. The overall score is then normalized across all M EEG feature components at the same stimulus frequency to obtain the weights:

[0114]

[0115] In the formula, It represents a very small number.

[0116] Furthermore, for each stimulus frequency f, a baseline-response mapping model is trained by taking the standardized pre-stimulus EEG feature component i from historical records as input and the corresponding post-stimulus EEG feature increment as output. The baseline-response mapping model can employ regularized linear regression, random forest, gradient boosting tree, or other supervised learning models, and cross-validation can be used to select model parameters.

[0117] S3. Input the standardized pre-stimulation EEG characteristics of the current patient into the baseline-response mapping model of each electroacupuncture frequency, predict the EEG characteristic increment at each frequency and calculate the predicted benefit score, and select the electroacupuncture frequency with the highest score as the initial recommended frequency.

[0118] Specifically, it includes:

[0119] Acquire the pre-stimulation EEG signal of the patient to be intervened and extract the pre-stimulation EEG features. A 61-channel EEG acquisition device is preferred, but a multi-channel EEG acquisition device covering the frontal, central, parietal, and occipital regions can also be used depending on clinical conditions. The acquisition module records the resting-state baseline EEG (pre-stimulation EEG signal) before stimulation therapy, records online EEG during stimulation therapy, and records the efficacy assessment EEG after treatment, simultaneously recording stimulation start, stimulation end, frequency switching, and stimulation pause events. The steps for extracting EEG features are the same as described above. The final pre-stimulation EEG feature X0 of the patient is obtained.

[0120] The baseline-response mapping model, EEG feature component weights, and response standard deviation for each electroacupuncture frequency are extracted from the population response spectrum of different electroacupuncture frequencies. The standardized pre-stimulation EEG features of the current subject to intervention are input into each model to obtain the predicted response at each frequency.

[0121]

[0122] The predicted benefit score is calculated based on the predicted response, weighted EEG feature components, response standard deviation, and preset beneficial direction, with frequency f as the specific calculation method:

[0123]

[0124] The meanings of the variables in the formula are the same as in Example 1. This step first predicts the feature increment after stimulation based on the patient's pre-stimulation EEG characteristics, and then calculates the gain on the predicted increment, without directly calculating the similarity between the pre-stimulation absolute features of different natures and the population increment map. When the difference between the highest predicted gain score and the second highest score is lower than the threshold, a candidate frequency set is output and the doctor is requested to confirm.

[0125] S4. During electroacupuncture stimulation, extract real-time EEG features to obtain real-time responses, and determine the benefits at the current frequency and the benefits at each candidate frequency based on the real-time responses and the group response spectra at different electroacupuncture frequencies.

[0126] Specifically, it includes:

[0127] The real-time EEG signal of the subject to intervention after electroacupuncture stimulation is acquired, and the real-time EEG features after stimulation are extracted. Here, the EEG data is divided into continuous or partially overlapping sliding time windows. The sliding window is used to extract the signal. The EEG signal in the current sliding window is used as the real-time EEG signal. Then, preprocessing and feature extraction are performed to obtain the real-time EEG features after stimulation.

[0128] The relative change (i.e., the difference) between the real-time EEG characteristics of the subject to intervention and the pre-stimulation EEG characteristics is calculated as the real-time response. Based on the real-time response and the group response spectrum of the current electroacupuncture frequency and each candidate electroacupuncture frequency, the benefit scores of the current electroacupuncture frequency and each candidate electroacupuncture frequency are determined; the candidate electroacupuncture frequencies are all electroacupuncture frequencies other than the current electroacupuncture frequency; candidate frequencies In the sliding window Under the profit rating It can be represented as:

[0129]

[0130]

[0131] In the formula, This represents the smoothing coefficient, for example, 0.2. This indicates the instantaneous matching benefit. express The i-th component value, This is a local minimum value, used to prevent the denominator from being 0.

[0132] S5. Based on the benefits at the current frequency and the benefits of each candidate frequency, determine whether to switch to the candidate electroacupuncture frequency.

[0133] Calculate the security score at the current moment, the security score It is obtained by weighting normalized indicators based on window quality, patient feedback comfort, and the risk level of artifacts or irritants; specifically:

[0134]

[0135] In the formula, , Indicates the weight value. The window quality, which can be represented by a weighted sum and normalized sum of artifact rate, bad conduction ratio, and signal-to-noise ratio, is as follows: This represents the patient feedback score, which is obtained by extracting scores from the patient's subjective feedback scale and then normalizing them. The risk level of artifacts or stimulation can be obtained by normalizing risk factors such as stimulation artifact rate, electromyography artifact rate, and the actual output current amplitude, monophasic pulse width, and duration of continuous stimulation of the electroacupuncture stimulator in the current control cycle. All of the above variables are converted to the same scoring interval before being included in the safety score, thereby avoiding the direct addition of indicators with different dimensions.

[0136] The current amplitude, single-phase pulse width, and continuous stimulation duration mentioned here all refer to the actual output parameters of the electroacupuncture stimulator in the current control cycle, and are denoted as follows: , and Its corresponding security limit , and The risk level is determined jointly by the device's safety range and the doctor's prescription. Stimulation artifact rate and electromyography artifact rate are cropped to the 0-1 range, and the above three output parameters are divided by their respective safety upper limits and then cropped to the 0-1 range, indicating the stimulation risk level. Calculate using the following formula:

[0137]

[0138]

[0139] and These represent the stimulus artifact rate and electromyography artifact rate within the current EEG window, respectively; clip(z) restricts z to the interval between 0 and 1. to The risk weights are non-negative and sum to 1, for example, all weights are 0.20. Window quality. The weighted sum of the artifact rate, bad conduction ratio, and low signal-to-noise ratio penalty term can be taken as 1, and the patient's feedback on comfort can be obtained. It can be obtained by subtracting the discomfort score from 1 and dividing by 10 from the 0 to 10 discomfort scale.

[0140] For example, taking α=0.40, β=0.30, γ=0.30, if =0.90、 =0.80、 =0.20, then the safety score is =0.54; when the preset safety threshold η=0.50, the safety gating is satisfied. If other conditions remain unchanged... If it rises to 0.70, then =0.39, which is below the threshold. The system will not perform frequency switching and will maintain the current frequency, reduce the stimulation output, or request confirmation from the doctor.

[0141] When there is a candidate electroacupuncture frequency, the maximum benefit score is available. Greater than the current electroacupuncture frequency Earnings rating : , It is a threshold and a security score. Greater than the preset threshold If the interval between adjacent frequency switching is greater than the minimum hold time, then the output switching is the one with the highest benefit score. Candidate electroacupuncture frequency The frequency switching command will be executed. Otherwise, the current frequency will be maintained or safety actions such as downgrading or pausing will be performed.

[0142] Example 3

[0143] This invention also provides a computer program product, such as an app on a mobile phone or tablet, or an installer on a computer. This product includes a computer program / instructions that, when executed by a processor, implement the method described in Embodiment 1. The code for the computer-executable program used to perform the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the subject's computer, partially on the subject's computer, as a standalone software package, partially on the subject's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the subject's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0144] It should be understood that the embodiments and descriptions above are only the principles, main features and advantages of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the invention, and all such changes and modifications fall within the protection scope of the present invention.

Claims

1. A smart electroacupuncture frequency optimization system based on population response spectrum, characterized in that, include: The EEG data processing module is used to acquire historical EEG signals of several subjects before and after stimulation at different electroacupuncture frequencies, and to extract several pairs of EEG features before and after stimulation at different electroacupuncture frequencies. The group response map construction module is used to construct a group response map of the subject group to different electroacupuncture frequencies by taking the changes in EEG characteristics before and after stimulation at different electroacupuncture frequencies as corresponding responses, and to train the baseline-response mapping model for each electroacupuncture frequency based on the EEG characteristics before historical stimulation and the corresponding responses. The starting frequency recommendation module is used to input the standardized pre-stimulation EEG characteristics of the current patient into the baseline-response mapping model of each electroacupuncture frequency, obtain the predicted response at each electroacupuncture frequency, calculate the predicted benefit score of each electroacupuncture frequency based on the predicted response, and select the electroacupuncture frequency with the highest score as the starting recommended frequency. The real-time state assessment module is used to extract real-time EEG features during electroacupuncture stimulation to obtain real-time responses, and to determine the benefits at the current frequency and the benefits at each candidate frequency based on the real-time responses and the group response maps of different electroacupuncture frequencies. The closed-loop control module is used to determine whether to switch to a candidate electroacupuncture frequency based on the benefit at the current frequency and the benefit at each candidate frequency.

2. The intelligent electroacupuncture frequency optimization system according to claim 1, characterized in that, The EEG features include at least one EEG feature component, which includes the mean of each brain region, the standard deviation of each brain region, the asymmetry between brain regions, and the cross-brain region difference features.

3. The intelligent electroacupuncture frequency optimization system according to claim 1, characterized in that, The group response mapping construction module specifically includes: The response determination unit is used to divide all EEG feature pairs under the same stimulus frequency into a group and calculate the relative changes of each EEG feature component before and after stimulation in each EEG feature pair within each group, as the response in the corresponding EEG feature component. The distribution statistics unit is used to calculate the mean, standard deviation, and weight of all responses of each EEG feature component within each group, forming a group response map for the corresponding stimulus frequency, and training a baseline-response mapping model based on historical pre-stimulation EEG features and corresponding responses at the same stimulus frequency.

4. The intelligent electroacupuncture frequency optimization system according to claim 3, characterized in that, The starting frequency recommendation module specifically includes: The pre-stimulation EEG feature extraction unit is used to acquire the pre-stimulation EEG signal of the current subject to intervention and extract standardized pre-stimulation EEG features. The prediction scoring unit is used to extract the baseline-response mapping model, EEG feature component weights, and response standard deviation of each electroacupuncture frequency from the population response spectrum of different electroacupuncture frequencies. The standardized pre-stimulation EEG features of the current subject to intervention are input into each baseline-response mapping model to obtain the predicted response. The predicted benefit score of each electroacupuncture frequency is calculated according to the preset beneficial direction, and the electroacupuncture frequency with the highest score is selected as the starting recommended frequency.

5. The intelligent electroacupuncture frequency optimization system according to claim 1, characterized in that, The real-time status assessment module specifically includes: The real-time EEG feature extraction unit is used to acquire the real-time EEG signal of the current subject after electroacupuncture stimulation and extract the real-time EEG features after stimulation. The benefit determination unit is used to calculate the relative change in the EEG characteristics of the current subject after real-time stimulation and the EEG characteristics before stimulation, as a real-time response. Based on the real-time response and the group response spectrum of the current electroacupuncture frequency and each candidate electroacupuncture frequency, the benefit score of the current electroacupuncture frequency and each candidate electroacupuncture frequency is determined; the candidate electroacupuncture frequencies are all electroacupuncture frequencies other than the current electroacupuncture frequency.

6. The intelligent electroacupuncture frequency optimization system according to claim 1, characterized in that, The closed-loop control module is specifically used for: When the benefit score of a candidate electroacupuncture frequency is greater than that of the current electroacupuncture frequency, a frequency switching command is output to switch to the candidate electroacupuncture frequency with the highest benefit score.

7. The intelligent electroacupuncture frequency optimization system according to claim 1, characterized in that, The closed-loop control module is specifically used for: Calculate the safety score at the current moment, which is obtained by weighting a normalized index of window quality, patient feedback comfort, and artifact or stimulus risk level. If the benefit score of a candidate electroacupuncture frequency is greater than the benefit score of the current electroacupuncture frequency, and the safety score is greater than a preset threshold, and the interval between adjacent frequency switching is greater than the minimum holding time, then a frequency switching command is output to switch to the candidate electroacupuncture frequency with the highest benefit score.

8. The intelligent electroacupuncture frequency optimization system according to claim 1, characterized in that, The system also includes: The efficacy tracking and interaction module is used to display the subject's EEG status, current stimulation frequency, recommended stimulation frequency, and reports, and to receive and execute the doctor's confirmation, overriding, or termination instructions.

9. A method for optimizing the frequency of intelligent electroacupuncture based on population response maps, characterized in that, include: We acquired historical EEG signals from several subjects before and after stimulation at different electroacupuncture frequencies, and extracted several pairs of EEG characteristics before and after stimulation at different electroacupuncture frequencies. The changes in EEG characteristics before and after stimulation at different electroacupuncture frequencies were used as corresponding responses to construct a population response map of the subject group to different electroacupuncture frequencies. Based on the EEG characteristics before historical stimulation and the corresponding responses, a baseline-response mapping model for each electroacupuncture frequency was trained. The standardized pre-stimulation EEG characteristics of the current patient to be intervened are input into the baseline-response mapping model of each electroacupuncture frequency to obtain the predicted response at each electroacupuncture frequency. The predicted benefit score is calculated based on the predicted response, and the electroacupuncture frequency with the highest score is selected as the initial recommended frequency. Real-time EEG features were extracted during electroacupuncture stimulation to obtain real-time responses. Based on the real-time responses and the group response maps of different electroacupuncture frequencies, the benefits at the current frequency and the benefits of each candidate frequency were determined. Based on the benefits at the current frequency and the benefits at each candidate frequency, determine whether to switch to a candidate electroacupuncture frequency.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, it implements the method of claim 9.

Images