Electroencephalogram data processing method and device based on dynamic feature selection

Abstract

The invention discloses an electroencephalogram data processing method and device based on dynamic feature selection, and relates to the field of electroencephalogram data processing.The method comprises the steps that before-intervention original electroencephalogram signals and after-intervention original electroencephalogram signals of a plurality of users are obtained; the original electroencephalogram signals before intervention and after intervention are subjected to preprocessing, directional phase transfer entropy calculation and directed function connection asymmetric index calculation; calculating an importance score of each lead pair of each electroencephalogram frequency band according to a mean value and a variance of the directed functional connection asymmetric indexes, and performing first feature selection and second feature selection to obtain data before selection intervention and data after selection intervention; according to the directed function connection asymmetry index in the data before selection intervention and the data after selection intervention, the emotion evaluation severity score and the intervention means evaluation result of the user are calculated, the electroencephalogram signal quality is improved, and then the accuracy of intervention means effectiveness evaluation is improved.

Description

Technical Field

[0001] This application relates to the field of electroencephalogram (EEG) data processing, and in particular to an EEG data processing method and apparatus based on dynamic feature selection. Background Technology

[0002] Current EEG signal analysis techniques for depression and anxiety face the following bottlenecks: First, insufficient resolution of directional functional connectivity. Existing methods are mostly limited to undirected phase synchronization indicators, failing to quantify abnormal information flow regulation from the prefrontal cortex to the limbic system, resulting in insufficient specificity in representing core symptoms such as anhedonia and executive dysfunction. Second, feature redundancy and poor model generalization. Traditional approaches often directly use full-dimensional feature modeling, which contains many redundant features. The effectiveness of current interventions for depression and anxiety is often assessed based on EEG data, but current EEG signal analysis techniques suffer from insufficient data accuracy. Summary of the Invention

[0003] The purpose of this application is to provide a method and device for processing electroencephalogram (EEG) data based on dynamic feature selection, which can improve the quality of EEG signals and thus improve the accuracy of evaluating the effectiveness of intervention measures.

[0004] To achieve the above objectives, this application provides the following solution:

[0005] Firstly, this application provides a method for processing electroencephalogram (EEG) data based on dynamic feature selection, including:

[0006] The original EEG signals of several users before and after intervention were acquired. The pre-intervention and post-intervention EEG signals were acquired by an intelligent EEG signal acquisition device based on a set sampling rate. The intelligent EEG signal acquisition device includes a data acquisition module. The data acquisition module includes several electrode units. Lead pairs are formed between two electrode units located in different brain hemispheres.

[0007] The raw EEG signals before and after the intervention were preprocessed to obtain the preprocessed pre-intervention signal and the preprocessed post-intervention signal. Both the preprocessed pre-intervention signal and the preprocessed post-intervention signal included EEG signals in several EEG frequency bands.

[0008] For each EEG frequency band, the pre-intervention signal after preprocessing is used to calculate the directional phase transfer entropy of each lead in the EEG frequency band before intervention based on the phase data of the pre-intervention signal after preprocessing. The directed functional connectivity asymmetry index between each lead pair in the EEG frequency band before intervention is also calculated based on the directional phase transfer entropy of each lead pair in the EEG frequency band before intervention. Similarly, for each EEG frequency band, the post-intervention signal after preprocessing is used to calculate the directional phase transfer entropy of each lead in the EEG frequency band after intervention based on the phase data of the pre-intervention signal after preprocessing.

[0009] The importance score for each pair of leads in each EEG band is calculated based on the mean and variance of the directed functional connectivity asymmetry index between each pair of leads in all EEG bands before intervention and the directed functional connectivity asymmetry index between each pair of leads in all EEG bands after intervention.

[0010] First feature selection data is obtained by performing first feature selection based on the importance score of each lead pair across all EEG bands.

[0011] The first feature selection data is randomly divided into several sample groups, and a second feature selection is performed according to the feature stability weight of each sample group to obtain the second feature selection data; the second feature selection data includes the data before the intervention and the data after the intervention.

[0012] For each user, the severity score of the user's emotional assessment and the assessment result of the intervention were calculated based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

[0013] Optionally, the data acquisition module includes a three-pronged comb module; the three-pronged comb module includes several three-pronged automatic hair-picking mechanisms integrated below the electrode unit, used to clean the hair in the electrode area when acquiring EEG signals.

[0014] Optionally, the data acquisition module further includes a MEMS pressure array and a PID controller; the MEMS pressure array includes a plurality of MEMS pressure sensors; each MEMS pressure sensor is embedded in each electrode unit for acquiring contact force data of the electrode unit;

[0015] The PID controller is used to perform reverse displacement compensation on the electrode unit when the difference between the contact force data and the pressure limit value of the electrode unit is not within the set difference range, so that the difference between the contact force data and the pressure limit value of the electrode unit falls within the set difference range.

[0016] Optionally, the preprocessing includes downsampling, filtering, denoising, and signal segmentation.

[0017] Optionally, the formula for calculating the directional phase transfer entropy is as follows:

[0018]

[0019] Where dPTE_{i→j} is the directional phase transfer entropy from lead i to lead j. τ represents the phase data of the pre-processed signal before intervention or the pre-processed signal after intervention at time t; τ is the optimal delay time. It is the joint probability distribution function. It is a conditional probability distribution function.

[0020] Optionally, based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data, the user's emotional assessment severity score and the intervention assessment results are calculated, specifically including:

[0021] A two-time-point causal difference matrix was constructed based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

[0022] Dynamic weights are calculated based on the average and maximum values ​​of the non-zero elements in the dual-time-point causal difference matrix. The average value of the non-zero elements is the average value obtained by taking the absolute values ​​of all non-zero elements in the dual-time-point causal difference matrix and then averaging them. The maximum value is the maximum value obtained by taking the absolute values ​​of all elements in the dual-time-point causal difference matrix.

[0023] The directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data are respectively input into the trained machine learning model to obtain the pre-intervention severity score and the post-intervention severity score.

[0024] The user's emotional assessment severity score is obtained by weighting and summing the dynamic weights, the severity scores before intervention, and the severity scores after intervention.

[0025] The user's intervention evaluation results are calculated based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

[0026] Optionally, based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data, the user's intervention evaluation results are calculated, specifically including:

[0027] For each EEG frequency band, the feature vector difference of the EEG frequency band is calculated based on the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band before intervention in the selected pre-intervention data and the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band after intervention in the selected post-intervention data.

[0028] The feature vector differences of all the EEG frequency bands are input into the trained intervention efficacy evaluation model to obtain the user's intervention evaluation results.

[0029] Secondly, this application provides an EEG data processing device based on dynamic feature selection, including an intelligent EEG signal acquisition device and a cloud analysis platform;

[0030] The intelligent EEG signal acquisition device includes a data acquisition module and a signal processing module. The data acquisition module is used to acquire raw EEG signals of several users before and after intervention. The data acquisition module includes several electrode units. Lead pairs are formed between two electrode units located in different brain hemispheres.

[0031] The signal processing module is used to preprocess the raw EEG signals before and after the intervention to obtain the preprocessed pre-intervention signal and the preprocessed post-intervention signal; both the preprocessed pre-intervention signal and the preprocessed post-intervention signal include EEG signals of several EEG frequency bands.

[0032] The cloud-based analysis platform is used for: for each EEG frequency band pre-intervention signal after preprocessing, calculating the directional phase transfer entropy of each lead of the EEG frequency band before intervention based on the phase data of the pre-intervention signal after preprocessing, and calculating the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band before intervention based on the directional phase transfer entropy of each lead of the EEG frequency band before intervention; for each EEG frequency band post-intervention signal after preprocessing, calculating the directional phase transfer entropy of each lead of the EEG frequency band after intervention based on the phase data of the pre-intervention signal after preprocessing, and calculating the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band after intervention based on the directional phase transfer entropy of each lead of the EEG frequency band after intervention.

[0033] The importance score for each pair of leads in each EEG band is calculated based on the mean and variance of the directed functional connectivity asymmetry index between each pair of leads in all EEG bands before intervention and the directed functional connectivity asymmetry index between each pair of leads in all EEG bands after intervention.

[0034] First feature selection data is obtained by performing first feature selection based on the importance score of each lead pair across all EEG bands.

[0035] The first feature selection data is randomly divided into several sample groups, and a second feature selection is performed according to the feature stability weight of each sample group to obtain the second feature selection data; the second feature selection data includes the data before the intervention and the data after the intervention.

[0036] For each user, the severity score of the user's emotional assessment and the assessment result of the intervention were calculated based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

[0037] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described EEG data processing method based on dynamic feature selection.

[0038] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described EEG data processing method based on dynamic feature selection.

[0039] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0040] This application provides a method and apparatus for processing EEG data based on dynamic feature selection. After obtaining raw EEG signals from several users before and after intervention, the method performs preprocessing, directional phase transfer entropy calculation, and directed functional connectivity asymmetry index calculation on the raw EEG signals before and after intervention, respectively. Based on the mean and variance of the directed functional connectivity asymmetry index, the method calculates the importance score of each pair of leads in each EEG frequency band, and performs first and second feature selection to obtain selected pre-intervention data and selected post-intervention data. Based on the directed functional connectivity asymmetry index in the selected pre-intervention data and selected post-intervention data, the method calculates the user's emotion assessment severity score and the intervention assessment result. This method solves the problem of insufficient data accuracy in current EEG signal analysis technology, improves the quality of EEG signals, and thus improves the accuracy of intervention effectiveness assessment. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating an EEG data processing method based on dynamic feature selection, provided as an embodiment of this application.

[0043] Figure 2 This is a schematic diagram of the functional modules of an EEG data processing device based on dynamic feature selection, provided in an embodiment of this application.

[0044] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0046] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] In one exemplary embodiment, such as Figure 1 As shown, a method for processing EEG data based on dynamic feature selection is provided, which includes the following steps 201 to 208.

[0048] Step 201: Obtain raw EEG signals before and after intervention from several users; the raw EEG signals before and after intervention are obtained by an intelligent EEG signal acquisition device based on a set sampling rate; the intelligent EEG signal acquisition device includes a data acquisition module; the data acquisition module includes several electrode units; lead pairs are formed between two electrode units located in different hemispheres.

[0049] Step 202: Preprocess the raw EEG signals before and after the intervention to obtain the preprocessed pre-intervention signal and the preprocessed post-intervention signal; both the preprocessed pre-intervention signal and the preprocessed post-intervention signal include EEG signals of several EEG frequency bands.

[0050] Step 203: For the pre-intervention signal after preprocessing of each EEG frequency band, calculate the directional phase transfer entropy (dPTE) of each lead of the EEG frequency band before intervention based on the phase data of the pre-intervention signal after preprocessing of the EEG frequency band, and calculate the directed functional connectivity asymmetry index (dFCAI) between each lead pair of the EEG frequency band before intervention based on the directional phase transfer entropy of each lead pair of the EEG frequency band before intervention; For the post-intervention signal after preprocessing of each EEG frequency band, calculate the post-intervention directional phase transfer entropy of each lead of the EEG frequency band after intervention based on the phase data of the pre-intervention signal after preprocessing of the EEG frequency band, and calculate the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band after intervention based on the directional phase transfer entropy of each lead pair of the EEG frequency band after intervention.

[0051] Step 204: Calculate the importance score for each pair of leads in each EEG band based on the mean and variance of the directed functional connectivity asymmetry index before intervention and the directed functional connectivity asymmetry index after intervention between each pair of leads in all EEG bands.

[0052] Step 205: Perform first feature selection based on the importance score of each lead pair in all EEG bands to obtain first feature selection data.

[0053] Step 206: The first feature selection data is randomly divided into several sample groups, and a second feature selection is performed according to the feature stability weight of each sample group to obtain the second feature selection data; the second feature selection data includes the data before the intervention and the data after the intervention.

[0054] Step 207: For each user, calculate the user's emotional assessment severity score and intervention assessment result based on the pre-intervention directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the post-intervention data.

[0055] Current EEG signal analysis technology for depression and anxiety disorders still faces the following bottlenecks: 1) Wearable device adaptability defects, rigid electrode structure leads to uneven contact pressure distribution, and hair obstruction causes low signal quality, which seriously restricts the reliability of monitoring in non-professional scenarios; 2) Existing technology is limited by the lack of single-time-point static analysis and directional dynamic tracking capabilities, and cannot capture this early neuroplastic response. Steps 201 to 207 above are performed, including preprocessing, calculating directional phase transfer entropy, and calculating the directed functional connectivity asymmetry index (DQA) of the raw EEG signals before and after intervention. Using the improved Fisher Score algorithm, importance scores for each lead pair in each EEG band are calculated based on the mean and variance of the DQA. First and second feature selections are then performed to select pre-intervention and post-intervention data. Based on the DQA from these data, the user's emotion assessment severity score and intervention evaluation results are calculated. This addresses the current limitations of EEG signal analysis techniques in terms of data accuracy, improves EEG signal quality, and consequently enhances the accuracy of intervention effectiveness assessment. The emotion assessment severity score and intervention evaluation results can be used for clinical guidance by physicians.

[0056] The intelligent EEG signal acquisition device is a flexible wearable device, comprising a data acquisition module, a signal processing module, a secure communication module, and a flexible power supply. The data acquisition module includes nylon elastic electrode caps, an electrode unit array, a three-pronged comb module, a MEMS pressure array, and a PID controller.

[0057] 1. Nylon elastic electrode caps: Electrode caps with the required number of leads are manufactured according to the 10-20 standard system, based on the specific requirements.

[0058] 2. Electrode unit array, comprising several electrode units: each electrode integrates four PZT-5A piezoelectric ceramic sheets (size 1.5×1.5×0.3mm, driving voltage ±18V, deformation resolution 0.01mm), achieving a deformation of 0-0.5mm through voltage control (0-±18V). Combined with closed-loop control using a MEMS pressure sensor, this ensures the electrode contact pressure remains stable at 12±0.8kPa. 0.8kPa is the set differential range. The electrode unit can employ piezoelectrically driven electrodes or pneumatically driven electrodes.

[0059] 3. The three-pronged comb module includes several three-pronged automatic hair-picking mechanisms: employing nickel-titanium alloy memory comb teeth (tooth spacing 0.15mm) and a micro harmonic reduction motor (reduction ratio 100:1), it cleans 3mm of hair per second. 2 or 5mm 2Regional hair removal improves signal quality when hair is obstructed. A three-pronged comb module, integrated below the electrode unit, is used to clear hair from the electrode area during EEG signal acquisition; each electrode unit also integrates a three-pronged automatic hair-removing mechanism.

[0060] 4. The MEMS pressure array includes several MEMS pressure sensors: a 16-unit capacitive sensor (range 0-20 kPa, accuracy 0.1% FS) can be used. Each MEMS pressure sensor is embedded in each electrode unit to collect contact force data of the electrode unit.

[0061] 5. Signal processing module: ADS1299-4 chipset, supporting hardware-level power frequency notch filtering (-120dB@50Hz).

[0062] 6. PID controller: The ADN8835 chip enables closed-loop pressure control (bandwidth 2kHz).

[0063] 7. Secure communication module: BLE 5.2 chip (PHY6212) supporting SM4 encryption.

[0064] 8. Flexible power source: Solid-state lithium polymer battery (3.7V 600mAh, charge-discharge cycle >2000 times).

[0065] The electrode unit is connected to the signal processing module via a silver paste circuit, the trident comb module is linked to the motor via a harmonic reducer, the MEMS pressure sensor feeds back data to the PID controller via an I2C interface, and the signal processing module transmits encrypted data packets to the secure communication module via an LVDS interface.

[0066] The PID controller is used to perform reverse displacement compensation on the electrode unit when the difference between the contact force data and the pressure limit value of the electrode unit is not within the set difference range, so that the difference between the contact force data and the pressure limit value of the electrode unit falls within the set difference range.

[0067] The electrode unit has a built-in 1kHz constant current source (100nA) to measure the scalp contact impedance Zc. When the impedance Zc is greater than the preset impedance threshold, the comb mechanism is activated. The comb movement trajectory is dynamically adjusted through a recursive PID control algorithm. One rotation of the three-pronged comb module can clean 5mm of scalp. 2 Regional hair. MEMS pressure sensors capture contact force signals in real time, with a pressure threshold set at 0.3-0.5N and a pressure limit of 0.5N. When the pressure exceeds the pressure limit, a reverse compensation movement is triggered to prevent scalp damage.

[0068] The intelligent EEG signal acquisition device proposed in this application is particularly suitable for long-term wearable EEG monitoring scenarios. Through the synergy of flexible electronics technology and an adaptive control system, the device achieves closed-loop management of the entire process, including dynamic optimization of scalp contact state, hair removal, contact pressure adjustment, and biosignal acquisition. This effectively solves the problems of signal distortion and skin damage caused by hair interference and uneven contact pressure in traditional EEG devices.

[0069] This intelligent EEG signal acquisition device adopts a multi-layer flexible composite structure design, with each lead area equipped with a three-pronged automatic hair-picking mechanism coupled to the electrode unit. Specifically, it includes: a nylon elastic electrode cap serving as the main support, with the lead areas on its surface forming an embedded silver paste circuit network through high-precision printing. Below each electrode unit is an integrated three-pronged automatic hair-picking mechanism, comprising shape memory alloy comb teeth and a micro-drive mechanism, forming an integrated linkage structure with the electrode unit. The piezoelectric drive module consists of an array of piezoelectric ceramic units, electrically coupled to the substrate conductive lines through a multi-layer interconnected structure, enabling precise control of the vertical displacement of the electrode unit. The contact pressure monitoring module employs a distributed sensing design, with its micro-MEMS pressure sensors embedded in a ring layout at the edge of the electrode unit to acquire contact force data in real time.

[0070] After the user wears the intelligent EEG signal acquisition device, the nylon elastic electrode cap, integrating a three-pronged hair-flicking mechanism, adaptively conforms to the scalp surface through an elastic constraint structure. The piezoelectric drive module, coupled to the electrode unit, first performs contact state detection, measuring the impedance characteristics of the electrode-scalp interface using a built-in AC excitation source. When the detected impedance value exceeds a preset impedance threshold, it triggers the micro-drive system of the three-pronged hair-flicking mechanism below the electrode, executing a helical progressive motion. This dynamic matching of the hair-cleaning range with the scalp curvature is achieved through a combination of radial expansion and helical contraction of the three-pronged comb teeth. During the cleaning process, the shape memory alloy three-pronged component automatically adjusts the tooth spacing using its hyperelastic properties, forming a progressive hair-guiding channel in the electrode contact area. In the contact pressure adjustment phase, a distributed MEMS pressure sensor, ring-shaped and embedded in the electrode edge, continuously collects contact force data. When the local pressure exceeds the limit, a reverse displacement compensation is generated through the coaxial linkage structure between the piezoelectric drive module and the electrode unit, utilizing the vertical displacement capability of the electrode unit to achieve dynamic balance of the contact force. The multi-channel processing architecture acquires raw EEG signals before and after intervention via electrode units directly connected to the substrate's conductive circuitry. After noise suppression and filtering, the signals are transmitted encrypted via a high-speed serial interface integrated into the elastic electrode cap. The power supply system, powered by a built-in flexible solid-state battery, intelligently distributes power through a conductive network integrally formed with the substrate, supporting continuous system operation.

[0071] Execution entities: embedded processor (signal processing module on the device side) and cloud analysis platform.

[0072] First, dual-timepoint signal acquisition (device-side) is performed. Signal acquisition: Resting-state EEG signals with eyes closed are acquired before intervention (T0) and after the set number of days of intervention (T1), each lasting 10 minutes, with a sampling rate of 250Hz. The intervention here can be physical or pharmacological. For physical intervention (e.g., rTMS): a 3-day interval is maintained, as synaptic plasticity peaks at 72 hours; for pharmacological intervention (e.g., SNRI): the interval can be adjusted to 5 days to accommodate pharmacokinetics. Dual-timepoint cross-calibration mechanism: Multidimensional EEG features are constructed from pre-treatment (T0) and post-treatment (n days after treatment, such as rTMS treatment, n=3; pharmacological treatment, n=5) (T1), and weighted fusion is used to generate an anti-interference physiological state index.

[0073] Hair removal: The three-pronged comb rotates at 30 rpm with a 3 mm travel per tooth, ensuring minimal hair in the electrode contact area.

[0074] The preprocessing in step 202 includes downsampling, filtering, denoising, and signal segmentation.

[0075] Signal downsampling: The signal sampling rate is downsampled from 250Hz to 125Hz.

[0076] Signal filtering and denoising: The signal is filtered to 4-30Hz and denoised by independent component analysis.

[0077] Signal segmentation: The signal is divided into five frequency bands: Theta (4-8Hz), Alpha1 (8-10Hz), Alpha2 (10-13Hz), Beta1 (13-20Hz), and Beta2 (20-30Hz), and the data is segmented according to time windows.

[0078] The following section performs directed functional connection calculations (device side). The formula for calculating the directionality of the information flow from lead i to j, i.e., the directional phase transfer entropy, is expressed as follows:

[0079]

[0080] Where dPTE_{i→j} is the directional phase transfer entropy from lead i to lead j. Let t be the phase data (Hilbert phase) of the pre-processed signal before intervention or the pre-processed signal after intervention at time t, where i and j represent the electrode unit numbers; τ is the optimal delay time, τ = 20 ms, determined by the mutual information method; P represents the probability distribution function. It is the joint probability distribution function. It is a conditional probability distribution function, and both are estimated based on EEG signal phase data.

[0081] The following results were obtained by discretizing the phase (binning), statistically analyzing its frequency, and normalizing:

[0082] Phase discretization: due to phase Since it is a continuous variable (range [0, 2π)), it needs to be discretized into N equal-width bins to calculate the probability. Usually, N = 18 or N = 36 (we can assume N = 18 as an example).

[0083] The formula for dividing the bins is expressed as follows:

[0084]

[0085] In the formula, bin k For the box index of the k-th equal-width box, each phase value or It is mapped to the corresponding bin index.

[0086] The joint probability is estimated by statistically analyzing the joint time frequency, using the following formula:

[0087]

[0088] Among them, count(bin) a ,bin b ) is in the data The frequency of occurrence, T is the total number of valid time points (for example, for a 10-minute signal at a sampling rate of 125Hz, T = 10 × 60 × 125 = 75000 points). Summation is performed for all bin combinations (a = 1, ..., N; b = 1, ..., N).

[0089] The estimation of conditional probability is based on the calculation of joint probability and marginal probability:

[0090]

[0091] Among them, marginal probability count(bin a )yes The total number of times; conditional probability only applies when Defined if it is set to 0, otherwise set to 0 or ignored.

[0092] dFCAI generation: Calculate the directed functional connectivity asymmetry index (also known as the "directional asymmetry index") for the left and right hemisphere lead pairs, and quantify the directional differences in information flow between the left and right hemispheres based on the directional phase transfer entropy algorithm.

[0093] dFCAI=(dPTE_{right→left}-dPTE_{left→right}) / (dPTE_{right→left}+dPTE_{left→right}+ε);

[0094] Where dPTE_{right→left} is the directional phase transition entropy of the lead between the electrodes located in the right hemisphere of the brain and the electrodes located in the left hemisphere, and dPTE_{left→right} is the directional phase transition entropy of the lead between the electrodes located in the left hemisphere of the brain and the electrodes located in the right hemisphere; ε=1e -8 To prevent division by zero in dFCAI calculations; dPTE calculations use Hilbert transform to extract phase information, with a time delay embedding dimension m=3.

[0095] The following describes the dynamic feature selection algorithm: combining the improved Fisher Score with sliding time window SHAP analysis, the optimal feature subset (i.e., the second feature selection data) is selected from the high-dimensional original features.

[0096] In step 204, for each EEG frequency band, the improved Fisher Score algorithm is used to calculate the importance score of each EEG frequency band and each lead pair based on the mean and variance of the directed functional connectivity asymmetry index before intervention and the directed functional connectivity asymmetry index after intervention between each lead pair of all EEG frequency bands.

[0097] The formula for calculating importance score is as follows:

[0098] Fisher(X) = |μ_1 - μ_0| / (σ_1) 2 +σ_0 2 );

[0099] Where μ_1 is the within-class mean of the directed functional connectivity asymmetry index among all EEG frequency bands and each lead pair of all users after the intervention set number of days, and μ_0 is the within-class mean of the directed functional connectivity asymmetry index among all EEG frequency bands and each lead pair of all users before the intervention; σ_1 is the variance of the directed functional connectivity asymmetry index among all EEG frequency bands and each lead pair of all users after the intervention set number of days, and σ_0 is the variance of the directed functional connectivity asymmetry index among all EEG frequency bands and each lead pair of all users before the intervention. Based on the importance score of each lead pair of all EEG frequency bands, the first feature selection is performed, retaining features with importance scores greater than the set score threshold (Fisher Score>T), thus obtaining the first feature selection data.

[0100] Then, a second feature selection is performed using sliding time window (SHAP) analysis. Within the random sample group, feature stability weights are calculated, and features with a coefficient of variation <15% are retained (i.e., data with a stability weight greater than 0.85). This yields the second feature selection data. The steps are as follows:

[0101] (1) Random sample group partitioning: The sample set data is randomly divided into multiple overlapping or non-overlapping sample groups. The sample set data is the first feature selection data.

[0102] (2) Calculate SHAP value: Within the sample group, use a pre-trained model (such as MLP in step 5) to calculate the SHAP value of each feature to quantify the contribution of the feature to the model prediction.

[0103] (3) Statistical calculation: For each feature, collect its SHAP value across all random sample groups, and calculate the mean and standard deviation. Mean (μ): reflects the average contribution of the feature. Standard deviation (σ): reflects the volatility of the feature's contribution.

[0104] (4) Calculate the coefficient of variation (CV): The formula for the coefficient of variation is: CV = μ / σ × 100%. The smaller the CV, the more stable the feature contribution.

[0105] (5) Determine the stability weight: The feature stability weight is defined as w = 1 - 100CV.

[0106] Screening criteria: Only retain features with a CV < 15% (i.e., w > 0.85). Weighting significance: The stability weight w is positively correlated with the stability of the feature contribution; the lower the CV, the higher the w.

[0107] (6) Feature selection: Finally, features that meet the coefficient of variation CV < 15% and that have a stability weight w > 0.85 are retained for subsequent modeling (such as PSI calculation or efficacy prediction in step 207 below).

[0108] In another example of this application, the first feature selection and the second feature selection described above can also be achieved using various machine learning regression methods such as LASSO regression.

[0109] The following section performs physiological state index fusion (through a cloud-based analysis platform).

[0110] In step 207, based on the directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the pre-intervention data and the directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the post-intervention data, the user's emotional assessment severity score and the intervention assessment result are calculated, specifically including the following steps 301 to 305.

[0111] Step 301: Construct a two-time-point causal difference matrix based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

[0112] Step 302: Calculate the dynamic weights based on the average and maximum values ​​of the non-zero elements in the dual-time-point causal difference matrix; the average value of the non-zero elements is the average value obtained by taking the absolute values ​​of all non-zero elements in the dual-time-point causal difference matrix and then averaging them, and the maximum value is the maximum value obtained by taking the absolute values ​​of all elements in the dual-time-point causal difference matrix.

[0113] Step 303: Input the directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the pre-intervention data and the directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the post-intervention data into the trained machine learning model to obtain the pre-intervention severity score and the post-intervention severity score.

[0114] Step 304: Calculate the user's emotional assessment severity score by weighting and summing the scores based on the dynamic weights, the pre-intervention severity score, and the post-intervention severity score.

[0115] Step 305: Calculate the user's intervention evaluation results based on the pre-intervention directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each EEG frequency band and each lead pair in the post-intervention data.

[0116] A dual-time-point causal difference matrix G is constructed based on single-time-point dFCAI, and dynamic weights are calculated.

[0117] G=GFC_dFCAI_T1-GFC_dFCAI_T0;

[0118] Wherein, GFC_dFCAI is the matrix form of dFCAI, GFC_dFCAI_T1 represents the matrix obtained by selecting the directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data, and GFC_dFCAI_T0 is the matrix composed of the directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data; the dual-time-point causal difference matrix G, GFC_dFCAI_T1, and GFC_dFCAI_T0 are all in N*N form, where N is the number of EEG leads. The formula for calculating the dynamic weight w is as follows:

[0119] w = 1 - (mean|G| / max(|G|));

[0120] Where mean|G| represents the average of the non-zero elements, that is, the average value calculated after taking the absolute value of all non-zero elements in the two-time-point causal difference matrix G; max(|G|) represents the maximum value selected after taking the absolute value of all elements in the two-time-point causal difference matrix G.

[0121] Output an emotion severity score, expressed as the PSI index (Psychopathology Severity Index):

[0122] The methods for quantitatively assessing the severity are as follows:

[0123] PSI=w·ML(dFCAI_T0)+(1-w)·ML(dFCAI_T1);

[0124] Wherein, dFCAI_T0 is the directed functional connectivity asymmetry index between each pair of leads in each EEG band before intervention in the selected pre-intervention data; dFCAI_T1 is the directed functional connectivity asymmetry index between each pair of leads in each EEG band after intervention in the selected post-intervention data; ML (machine learning) is the machine learning model; ML(dFCAI_T0) is the severity score before intervention; ML(dFCAI_T1) is the severity score after intervention. Furthermore, since this is a prediction of severity, ML here specifically refers to a regression model. The inputs are features and severity scores, and the output is an objective emotion assessment severity score.

[0125] Treatment response trend modeling (performed via a cloud analytics platform) is carried out. Step 305 specifically includes: for each EEG frequency band, calculating the feature vector difference of that frequency band based on the pre-intervention directed functional connectivity asymmetry index between each lead pair of the selected EEG frequency band and the post-intervention directed functional connectivity asymmetry index between each lead pair of the selected EEG frequency band and the post-intervention directed functional connectivity asymmetry index between each lead pair of the selected EEG frequency band; inputting the feature vector differences of all the selected EEG frequency bands into a trained intervention efficacy evaluation model to obtain the user's intervention evaluation result. The trained intervention efficacy evaluation model uses the feature vector differences of the user's sample EEG frequency bands as input, and the user's intervention efficacy result as the model trained with labels.

[0126] The eigenvector difference is calculated and input into the intervention efficacy assessment model (machine learning model) to predict the intervention efficacy. The input to the intervention efficacy assessment model is the eigenvector difference (ΔdFCAI), and the output is the treatment efficacy. The intervention assessment result is divided into effective or ineffective. The formula for calculating the eigenvector difference is as follows:

[0127] ΔdFCAI = dFCAI_T1 - dFCAI_T0;

[0128] Wherein, dFCAI_T1 is the directed functional connectivity asymmetry index between each pair of leads in the selected EEG bands after intervention; dFCAI_T0 is the directed functional connectivity asymmetry index between each pair of leads in the selected pre-intervention EEG bands. The intervention efficacy assessment model can employ models such as XGBoost, SVM, Random Forest (RF), and LightGBM.

[0129] First, the raw signal is downsampled, frequency band filtered, analyzed independently of other components, and segmented at the device end to extract EEG frequency bands such as Theta, Alpha1, Alpha2, Beta1, and Beta2. Then, directed functional connectivity between leads is calculated based on directional phase transfer entropy (dPTE), and an interhemispheric directional difference index is generated. In the cloud processing stage, stable features are screened using an improved FisherScore and sliding SHAP analysis, combined with a dynamically weighted Granger causality difference matrix, to generate a quantitative assessment score for the severity of emotional state. Finally, treatment response trends are predicted based on feature vector changes, forming a closed-loop analysis process from signal preprocessing to efficacy prediction.

[0130] Based on resting-state EEG data from 27 patients with depression (12 in the drug-responsive group and 15 in the ineffective group), multidimensional feature extraction and machine learning methods were used to predict the efficacy of SSRIs (antidepressants). Data preprocessing included downsampling (250Hz→125Hz), 4-30Hz bandpass filtering, and independent component analysis for noise reduction. The signal was segmented into five frequency bands: Theta (4-8Hz), Alpha1 (8-10Hz), Alpha2 (10-13Hz), Beta1 (13-20Hz), and Beta2 (20-30Hz), and further segmented according to different time windows (4 seconds, 6 seconds, 8 seconds, 10 seconds, 12 seconds, and 14 seconds). The directional phase transition entropy between leads was extracted as a feature. Shapley additive interpretation (SHAP) combined with four classifiers—XGBoost, SVM, Random Forest (RF), and LightGBM—was used for dynamic feature optimization, and the performance was evaluated using leave-one-out cross-validation (LOOCV). Full feature analysis (Table 1) shows that the initial SSRI efficacy prediction accuracy of each model was 86.80%-88.31%, while the feature selection optimization (Table 2) significantly improved the accuracy. Among them, SVM achieved the highest accuracy of 96.83% with 65 optimal features in a 12-second window.

[0131] Table 1. Predictive accuracy of SSRI efficacy without feature selection.

[0132]

[0133]

[0134] Table 2. Predictive accuracy of SSRI efficacy after feature selection

[0135]

[0136] This application has the following advantages:

[0137] 1. High accuracy of directional connectivity analysis: Step 202 makes dFCAI highly correlated with clinical assessment because directional features can sensitively capture prefrontal cortex-limbic system regulation disorders.

[0138] 2. Feature selection efficiency innovation: The Fisher-SHAP hybrid algorithm (steps 204-206) compresses the feature dimension to a low dimension while retaining most of the effective information, thus reducing the computation time.

[0139] 3. Revolution in Wearable Comfort: Piezoelectric drive and a three-pronged comb module improve signal pass rate even when hair obscures the signal. Simultaneously, a triple feedback mechanism of impedance, motion, and force enables closed-loop control with optimized contact, resulting in improved user comfort.

[0140] 4. Enhanced resistance to interference: The dual-time-point Granger causal weighting (dual-time-point causal difference matrix) improves the environmental interference resistance of the emotion assessment severity score.

[0141] This application also provides an application scenario in which the above-described EEG data processing method based on dynamic feature selection is applied. Specifically, the EEG data processing method based on dynamic feature selection provided in this embodiment can be applied in an intervention effect evaluation scenario. The intervention effect evaluation scenario includes a content production stage, a content processing chain, and a content distribution stage. The user's raw EEG signals before and after intervention enter the content processing chain from the content production stage. Through human-computer collaboration, corresponding emotion assessment severity scores and intervention measure evaluation results are obtained, and then enter the downstream content distribution stage. The EEG data processing method based on dynamic feature selection provided in this embodiment belongs to the content processing chain. Specifically, in the content processing chain of the user's raw EEG signals before and after intervention, the raw EEG signals before and after intervention can be preprocessed, directional phase transfer entropy calculated, and directed functional connectivity asymmetry index calculated, respectively. The importance score of each EEG frequency band and each lead pair is calculated based on the mean and variance of the directed functional connectivity asymmetry index, and the first feature selection and the second feature selection are performed to obtain the selected pre-intervention data and the selected post-intervention data. The user's emotion assessment severity score and the intervention assessment result are calculated based on the directed functional connectivity asymmetry index in the selected pre-intervention data and the selected post-intervention data.

[0142] Based on the same inventive concept, this application also provides a dynamic feature selection-based EEG data processing device for implementing the above-described dynamic feature selection-based EEG data processing method. The solution provided by this device is similar to the implementation described in the above-described method. Therefore, the specific limitations of one or more embodiments of the dynamic feature selection-based EEG data processing device provided below can be found in the limitations of the dynamic feature selection-based EEG data processing method described above, and will not be repeated here.

[0143] In one exemplary embodiment, a dynamic feature-based EEG data processing device is provided, comprising an intelligent EEG signal acquisition device and a cloud analysis platform. The intelligent EEG signal acquisition device is a flexible wearable device, including a data acquisition module, a signal processing module, a secure communication module, and a flexible power supply. The data acquisition module includes a nylon elastic electrode cap, an electrode unit array, a three-pronged comb module, a MEMS pressure array, and a PID controller. The dynamic feature-based EEG data processing device... Figure 2 As shown.

[0144] The intelligent EEG signal acquisition device includes a data acquisition module and a signal processing module. The data acquisition module is used to acquire raw EEG signals of several users before and after intervention. The data acquisition module includes several electrode units. Lead pairs are formed between two electrode units located in different brain hemispheres.

[0145] The signal processing module is used to preprocess the raw EEG signals before and after the intervention to obtain the preprocessed pre-intervention signal and the preprocessed post-intervention signal; both the preprocessed pre-intervention signal and the preprocessed post-intervention signal include EEG signals of several EEG frequency bands.

[0146] The cloud-based analysis platform is used for: for each EEG frequency band pre-intervention signal after preprocessing, calculating the directional phase transfer entropy of each lead of the EEG frequency band before intervention based on the phase data of the pre-intervention signal after preprocessing, and calculating the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band before intervention based on the directional phase transfer entropy of each lead of the EEG frequency band before intervention; for each EEG frequency band post-intervention signal after preprocessing, calculating the directional phase transfer entropy of each lead of the EEG frequency band after intervention based on the phase data of the pre-intervention signal after preprocessing, and calculating the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band after intervention based on the directional phase transfer entropy of each lead of the EEG frequency band after intervention.

[0147] The importance score for each pair of leads in each EEG band is calculated based on the mean and variance of the directed functional connectivity asymmetry index between each pair of leads in all EEG bands before intervention and the directed functional connectivity asymmetry index between each pair of leads in all EEG bands after intervention.

[0148] First feature selection data is obtained by performing first feature selection based on the importance score of each lead pair across all EEG bands.

[0149] The first feature selection data is randomly divided into several sample groups, and a second feature selection is performed according to the feature stability weight of each sample group to obtain the second feature selection data; the second feature selection data includes the data before the intervention and the data after the intervention.

[0150] For each user, the severity score of the user's emotional assessment and the assessment result of the intervention were calculated based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

[0151] 1. Nylon elastic electrode caps: Electrode caps with the required number of leads are manufactured according to the 10-20 standard system, based on the specific requirements.

[0152] 2. Electrode unit array, comprising several electrode units: each electrode integrates four PZT-5A piezoelectric ceramic sheets (size 1.5×1.5×0.3mm, driving voltage ±18V, deformation resolution 0.01mm), achieving a deformation of 0-0.5mm through voltage control (0-±18V). Combined with closed-loop control using a MEMS pressure sensor, this ensures the electrode contact pressure remains stable at 12±0.8kPa. 0.8kPa is the set differential range. The electrode unit can employ piezoelectrically driven electrodes or pneumatically driven electrodes.

[0153] 3. The three-pronged comb module includes several three-pronged automatic hair-picking mechanisms: employing nickel-titanium alloy memory comb teeth (tooth spacing 0.15mm) and a micro harmonic reduction motor (reduction ratio 100:1), it cleans 3mm of hair per second. 2 or 5mm 2 Regional hair removal improves signal quality when hair is obstructed. A three-pronged comb module, integrated below the electrode unit, is used to clear hair from the electrode area during EEG signal acquisition; each electrode unit also integrates a three-pronged automatic hair-removing mechanism.

[0154] 4. The MEMS pressure array includes several MEMS pressure sensors: a 16-unit capacitive sensor (range 0-20 kPa, accuracy 0.1% FS) can be used. Each MEMS pressure sensor is embedded in each electrode unit to collect contact force data of the electrode unit.

[0155] 5. Signal processing module: ADS1299-4 chipset, supporting hardware-level power frequency notch filtering (-120dB@50Hz).

[0156] 6. PID controller: The ADN8835 chip enables closed-loop pressure control (bandwidth 2kHz).

[0157] 7. Secure communication module: BLE 5.2 chip (PHY6212) supporting SM4 encryption.

[0158] 8. Flexible power source: Solid-state lithium polymer battery (3.7V 600mAh, charge-discharge cycle >2000 times).

[0159] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 3 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores EEG data for processing. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements an EEG data processing method based on dynamic feature selection.

[0160] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0161] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0162] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0163] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0164] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0165] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0167] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for processing electroencephalogram (EEG) data based on dynamic feature selection, characterized in that, The EEG data processing method based on dynamic feature selection includes: The original EEG signals of several users before and after intervention were acquired. The pre-intervention and post-intervention EEG signals were acquired by an intelligent EEG signal acquisition device based on a set sampling rate. The intelligent EEG signal acquisition device includes a data acquisition module. The data acquisition module includes several electrode units. Lead pairs are formed between two electrode units located in different brain hemispheres. The raw EEG signals before and after the intervention were preprocessed to obtain the preprocessed pre-intervention signal and the preprocessed post-intervention signal. Both the preprocessed pre-intervention signal and the preprocessed post-intervention signal included EEG signals in several EEG frequency bands. For each EEG frequency band, the pre-intervention signal after preprocessing is used to calculate the directional phase transfer entropy of each lead in the EEG frequency band before intervention based on the phase data of the pre-intervention signal after preprocessing. The directed functional connectivity asymmetry index between each lead pair in the EEG frequency band before intervention is also calculated based on the directional phase transfer entropy of each lead pair in the EEG frequency band before intervention. Similarly, for each EEG frequency band, the post-intervention signal after preprocessing is used to calculate the directional phase transfer entropy of each lead in the EEG frequency band after intervention based on the phase data of the pre-intervention signal after preprocessing. The importance score for each pair of leads in each EEG band is calculated based on the mean and variance of the directed functional connectivity asymmetry index between each pair of leads in all EEG bands before intervention and the directed functional connectivity asymmetry index between each pair of leads in all EEG bands after intervention. First feature selection data is obtained by performing first feature selection based on the importance score of each lead pair across all EEG bands. The first feature selection data is randomly divided into several sample groups, and a second feature selection is performed according to the feature stability weight of each sample group to obtain the second feature selection data; the second feature selection data includes the data before the intervention and the data after the intervention. For each user, the severity score of the user's emotional assessment and the assessment result of the intervention were calculated based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

2. The EEG data processing method based on dynamic feature selection according to claim 1, characterized in that, The data acquisition module includes a three-pronged comb module; the three-pronged comb module includes several three-pronged automatic hair-picking mechanisms integrated below the electrode unit, used to clean the hair in the electrode area when acquiring EEG signals.

3. The EEG data processing method based on dynamic feature selection according to claim 1, characterized in that, The data acquisition module also includes a MEMS pressure array and a PID controller; the MEMS pressure array includes several MEMS pressure sensors; each MEMS pressure sensor is embedded in each electrode unit and is used to collect contact force data of the electrode unit. The PID controller is used to perform reverse displacement compensation on the electrode unit when the difference between the contact force data and the pressure limit value of the electrode unit is not within the set difference range, so that the difference between the contact force data and the pressure limit value of the electrode unit falls within the set difference range.

4. The EEG data processing method based on dynamic feature selection according to claim 1, characterized in that, The preprocessing includes downsampling, filtering, noise reduction, and signal segmentation.

5. The EEG data processing method based on dynamic feature selection according to claim 1, characterized in that, The formula for calculating the directional phase transfer entropy is as follows: dPTE_{i→j}=-∑p(φ_i(t),φ_j(t+τ))log[p(φ_j(t+τ)|φ_i(t))]; Where dPTE_{i→j} is the directional phase transfer entropy from lead i to j, φ_i(t) is the phase data of the pre-intervention signal or the pre-intervention signal at time t; τ is the optimal delay time; p(φ_i(t), φ_j(t+τ)) is the joint probability distribution function, and p(φ_j(t+τ)|φ_i(t)) is the conditional probability distribution function.

6. The EEG data processing method based on dynamic feature selection according to claim 1, characterized in that, Based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data, the user's emotional assessment severity score and the intervention assessment results are calculated, specifically including: A two-time-point causal difference matrix was constructed based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data. Dynamic weights are calculated based on the average and maximum values ​​of the non-zero elements in the dual-time-point causal difference matrix. The average value of the non-zero elements is the average value obtained by taking the absolute values ​​of all non-zero elements in the dual-time-point causal difference matrix and then averaging them. The maximum value is the maximum value obtained by taking the absolute values ​​of all elements in the dual-time-point causal difference matrix. The directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data are respectively input into the trained machine learning model to obtain the pre-intervention severity score and the post-intervention severity score. The user's emotional assessment severity score is obtained by weighting and summing the dynamic weights, the severity scores before intervention, and the severity scores after intervention. The user's intervention evaluation results are calculated based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

7. The EEG data processing method based on dynamic feature selection according to claim 6, characterized in that, Based on the directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data, the user's intervention assessment results are calculated, specifically including: For each EEG frequency band, the feature vector difference of the EEG frequency band is calculated based on the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band before intervention in the selected pre-intervention data and the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band after intervention in the selected post-intervention data. The feature vector differences of all the EEG frequency bands are input into the trained intervention efficacy evaluation model to obtain the user's intervention evaluation results.

8. A brainwave data processing device based on dynamic feature selection, characterized in that, The EEG data processing device based on dynamic feature selection includes an intelligent EEG signal acquisition device and a cloud analysis platform; The intelligent EEG signal acquisition device includes a data acquisition module and a signal processing module. The data acquisition module is used to acquire raw EEG signals of several users before and after intervention. The data acquisition module includes several electrode units. Lead pairs are formed between two electrode units located in different brain hemispheres. The signal processing module is used to preprocess the raw EEG signals before and after the intervention to obtain the preprocessed pre-intervention signal and the preprocessed post-intervention signal; both the preprocessed pre-intervention signal and the preprocessed post-intervention signal include EEG signals of several EEG frequency bands. The cloud-based analysis platform is used for: for each EEG frequency band pre-intervention signal after preprocessing, calculating the directional phase transfer entropy of each lead of the EEG frequency band before intervention based on the phase data of the pre-intervention signal after preprocessing, and calculating the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band before intervention based on the directional phase transfer entropy of each lead of the EEG frequency band before intervention; for each EEG frequency band post-intervention signal after preprocessing, calculating the directional phase transfer entropy of each lead of the EEG frequency band after intervention based on the phase data of the pre-intervention signal after preprocessing, and calculating the directed functional connectivity asymmetry index between each lead pair of the EEG frequency band after intervention based on the directional phase transfer entropy of each lead of the EEG frequency band after intervention. The importance score for each pair of leads in each EEG band is calculated based on the mean and variance of the directed functional connectivity asymmetry index between each pair of leads in all EEG bands before intervention and the directed functional connectivity asymmetry index between each pair of leads in all EEG bands after intervention. First feature selection data is obtained by performing first feature selection based on the importance score of each lead pair across all EEG bands. The first feature selection data is randomly divided into several sample groups, and the second feature selection is performed according to the feature stability weight of each sample group to obtain the second feature selection data. The second feature selection data includes selecting pre-intervention data and selecting post-intervention data; For each user, the severity score of the user's emotional assessment and the assessment result of the intervention were calculated based on the pre-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the pre-intervention data and the post-intervention directed functional connectivity asymmetry index between each pair of EEG frequencies and leads in the post-intervention data.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the EEG data processing method based on dynamic feature selection as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the EEG data processing method based on dynamic feature selection as described in any one of claims 1-7.

Images